From 46d7267ec2eaa78845095ca97cb85ff8453b8a21 Mon Sep 17 00:00:00 2001
From: Sohan Kshirsagar <sohan.kshirsagar@gmail.com>
Date: Tue, 14 Apr 2026 13:48:41 -0700
Subject: [PATCH] Add Tusk results

---
 offline/README.md                             |    1 +
 offline/analysis/benchmark_dashboard.html     |   15 +-
 offline/analysis/benchmark_dashboard.json     | 2284 +++++++++++-----
 offline/analysis/benchmark_dashboard.py       |    2 +
 .../candidates.json                           | 1308 ++++++++-
 .../evaluations.json                          | 2160 +++++++++++++++
 .../candidates.json                           | 1356 +++++++++-
 .../evaluations.json                          | 2195 +++++++++++++++
 offline/results/benchmark_data.json           | 1771 ++++++++++++
 .../results/openai_gpt-5.2/candidates.json    | 1556 ++++++++++-
 .../results/openai_gpt-5.2/evaluations.json   | 2389 ++++++++++++++++-
 11 files changed, 14161 insertions(+), 876 deletions(-)

diff --git a/offline/README.md b/offline/README.md
index 3aa9456..8351558 100644
--- a/offline/README.md
+++ b/offline/README.md
@@ -17,6 +17,7 @@ Open replication of the code review benchmark used by companies like [Augment](h
 | [Greptile](https://www.greptile.com/) | AI code review |
 | [Propel](https://propelauth.com/) | AI code review |
 | [Qodo](https://www.qodo.ai/) | AI code review |
+| [Tusk](https://www.usetusk.ai/) | AI code review & testing |
 
 Adding a new tool requires forking the benchmark PRs and collecting the tool's reviews — see Steps 0 and 1 below.
 
diff --git a/offline/analysis/benchmark_dashboard.html b/offline/analysis/benchmark_dashboard.html
index 4905b9e..ae0f6db 100644
--- a/offline/analysis/benchmark_dashboard.html
+++ b/offline/analysis/benchmark_dashboard.html
@@ -193,22 +193,23 @@
         <div class="predefined-filter" data-filter-id="high_precision" onclick="applyPredefinedFilter('high_precision')">Highest Precision <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_kodus-v2_domain_concurrency" onclick="applyPredefinedFilter('tool_kodus-v2_domain_concurrency')">Best for Concurrency (Precision) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_kg_complexity_complex" onclick="applyPredefinedFilter('tool_kg_complexity_complex')">Best for Complex Code (Precision) <span class="arrow">↗</span></div>
-        <div class="predefined-filter" data-filter-id="tool_greptile-v4_change_type_bug_fix" onclick="applyPredefinedFilter('tool_greptile-v4_change_type_bug_fix')">Best for Bug Fixes (Recall) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_copilot_change_type_bug_fix" onclick="applyPredefinedFilter('tool_copilot_change_type_bug_fix')">Best for Bug Fixes (Recall) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_mesa_change_type_bug_fix" onclick="applyPredefinedFilter('tool_mesa_change_type_bug_fix')">Best for Bug Fixes (Recall) <span class="arrow">↗</span></div>
+        <div class="predefined-filter" data-filter-id="tool_greptile-v4_change_type_bug_fix" onclick="applyPredefinedFilter('tool_greptile-v4_change_type_bug_fix')">Best for Bug Fixes (Recall) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="change_type_performance" onclick="applyPredefinedFilter('change_type_performance')">Best for Performance Optimization <span class="arrow">↗</span></div>
+        <div class="predefined-filter" data-filter-id="tool_tusk_domain_ui" onclick="applyPredefinedFilter('tool_tusk_domain_ui')">Best for Ui (Recall) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="high_recall" onclick="applyPredefinedFilter('high_recall')">Highest Recall <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_qodo-v2-2_language_java_domain_authentication" onclick="applyPredefinedFilter('tool_qodo-v2-2_language_java_domain_authentication')">Java + Authentication <span class="arrow">↗</span></div>
-        <div class="predefined-filter" data-filter-id="tool_codeant_difficulty_moderate_context_file" onclick="applyPredefinedFilter('tool_codeant_difficulty_moderate_context_file')">Moderate Bugs + File Context (Recall) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_mra-a_language_java_risk_high" onclick="applyPredefinedFilter('tool_mra-a_language_java_risk_high')">Java + High Risk (Recall) <span class="arrow">↗</span></div>
+        <div class="predefined-filter" data-filter-id="tool_codeant_difficulty_moderate_context_file" onclick="applyPredefinedFilter('tool_codeant_difficulty_moderate_context_file')">Moderate Bugs + File Context (Recall) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="ruby_medium" onclick="applyPredefinedFilter('ruby_medium')">Best for Medium Ruby PRs <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_sourcery_language_typescript_concern_correctness" onclick="applyPredefinedFilter('tool_sourcery_language_typescript_concern_correctness')">Typescript + Correctness <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="risk_critical" onclick="applyPredefinedFilter('risk_critical')">Best for Critical Risk <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_baz_language_java_domain_authentication" onclick="applyPredefinedFilter('tool_baz_language_java_domain_authentication')">Java + Authentication <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_qodo-extended_change_type_bug_fix_context_cross_file" onclick="applyPredefinedFilter('tool_qodo-extended_change_type_bug_fix_context_cross_file')">Bug Fixes + Cross-File <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="domain_ui" onclick="applyPredefinedFilter('domain_ui')">Best for Ui <span class="arrow">↗</span></div>
-        <div class="predefined-filter" data-filter-id="tool_qodo-v2_language_java_risk_high" onclick="applyPredefinedFilter('tool_qodo-v2_language_java_risk_high')">Java + High Risk <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_macroscope_context_file_concern_correctness" onclick="applyPredefinedFilter('tool_macroscope_context_file_concern_correctness')">File Context + Correctness <span class="arrow">↗</span></div>
+        <div class="predefined-filter" data-filter-id="tool_qodo-v2_language_java_risk_high" onclick="applyPredefinedFilter('tool_qodo-v2_language_java_risk_high')">Java + High Risk <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_greptile_pr_size_medium_context_file" onclick="applyPredefinedFilter('tool_greptile_pr_size_medium_context_file')">Medium PRs + File Context <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="tool_propel_language_ruby_change_type_feature" onclick="applyPredefinedFilter('tool_propel_language_ruby_change_type_feature')">Ruby + Features (Precision) <span class="arrow">↗</span></div>
         <div class="predefined-filter" data-filter-id="complexity_complex" onclick="applyPredefinedFilter('complexity_complex')">Best for Complex Code <span class="arrow">↗</span></div>
@@ -413,10 +414,10 @@
 
     <script>
         // All models data embedded
-        const allModelsData = {"anthropic_claude-opus-4-5-20251101": {"prs": [{"url": "https://github.com/keycloak/keycloak/pull/37429", "language": "Java", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "gemini": {"tp": 1, "fp": 4, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 3}, "augment": {"tp": 2, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "copilot": {"tp": 3, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 4}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 2, "fp": 5, "fn": 2}, "mra-nano": {"tp": 0, "fp": 2, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-a": {"tp": 0, "fp": 4, "fn": 4}, "mra-b": {"tp": 0, "fp": 2, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 4}, "sourcery": {"tp": 1, "fp": 6, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 14, "fn": 1}, "mesa": {"tp": 2, "fp": 8, "fn": 2}, "codeant": {"tp": 2, "fp": 14, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 3}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 4}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 4}}}, {"url": "https://github.com/keycloak/keycloak/pull/37634", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 1, "fn": 2}, "gemini": {"tp": 3, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "augment": {"tp": 2, "fp": 2, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "graphite": {"tp": 2, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 2}, "bugbot": {"tp": 2, "fp": 1, "fn": 2}, "copilot": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 5, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 4, "fn": 2}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 0, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 9, "fn": 2}, "kodus": {"tp": 2, "fp": 10, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "claude-code": {"tp": 0, "fp": 2, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 2}, "macroscope": {"tp": 2, "fp": 2, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/38446", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 2}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 2}, "mesa": {"tp": 1, "fp": 6, "fn": 1}, "codeant": {"tp": 1, "fp": 6, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 0}, "macroscope": {"tp": 0, "fp": 2, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/36882", "language": "Java", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 0}, "augment": {"tp": 0, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 7, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 0, "fp": 1, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36880", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 4, "fn": 3}, "gemini": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 2, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 6, "fn": 2}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 3}, "mra-max": {"tp": 2, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 2, "fn": 2}, "mra-a": {"tp": 3, "fp": 2, "fn": 0}, "entelligence": {"tp": 0, "fp": 5, "fn": 3}, "sourcery": {"tp": 0, "fp": 6, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 1, "fp": 0, "fn": 2}, "codeant": {"tp": 0, "fp": 10, "fn": 3}, "kodus": {"tp": 0, "fp": 5, "fn": 3}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/37038", "language": "Java", "pr_size": "small", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"qodo": {"tp": 1, "fp": 2, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 9, "fn": 1}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 2, "fp": 6, "fn": 0}, "mra-b": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 10, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 9, "fn": 1}, "kodus": {"tp": 2, "fp": 9, "fn": 0}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 0}, "macroscope": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/33832", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 1, "fp": 5, "fn": 1}, "gemini": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 5, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 9, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 9, "fn": 2}, "mra-a": {"tp": 0, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 11, "fn": 2}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 1, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/40940", "language": "Java", "pr_size": "small", "domain": "concurrency", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 1, "fn": 2}, "gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 2, "fp": 2, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"augment": {"tp": 2, "fp": 0, "fn": 0}, "gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 6, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 0, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 4, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 3, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 0, "fp": 6, "fn": 2}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 1, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 0, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/getsentry/sentry/pull/93824", "language": "Python", "pr_size": "large", "domain": "concurrency", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 0, "fn": 5}, "augment": {"tp": 5, "fp": 1, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 4}, "coderabbit": {"tp": 0, "fp": 1, "fn": 5}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "claude": {"tp": 0, "fp": 2, "fn": 5}, "qodo": {"tp": 1, "fp": 3, "fn": 4}, "copilot": {"tp": 1, "fp": 5, "fn": 4}, "propel": {"tp": 2, "fp": 2, "fn": 3}, "greptile": {"tp": 1, "fp": 1, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 3, "fp": 3, "fn": 2}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 0, "fp": 5, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 5}, "mra-b": {"tp": 1, "fp": 7, "fn": 4}, "mra-a": {"tp": 1, "fp": 5, "fn": 4}, "entelligence": {"tp": 1, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 5}, "codeant": {"tp": 1, "fp": 5, "fn": 4}, "kodus": {"tp": 3, "fp": 2, "fn": 2}, "claude-code": {"tp": 3, "fp": 3, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 5}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 4}, "cubic-v2": {"tp": 5, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 3}, "macroscope": {"tp": 3, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5", "language": "Python", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 3}, "greptile": {"tp": 0, "fp": 2, "fn": 3}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "coderabbit": {"tp": 1, "fp": 15, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 13, "fn": 3}, "greptile-v4": {"tp": 0, "fp": 1, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 0, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 2, "fp": 28, "fn": 1}, "mesa": {"tp": 0, "fp": 14, "fn": 3}, "codeant": {"tp": 1, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 17, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 1, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 2}, "macroscope": {"tp": 0, "fp": 5, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1", "language": "Python", "pr_size": "medium", "domain": "API", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 3}, "claude": {"tp": 2, "fp": 4, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 1, "fp": 2, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 1, "fn": 1}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "coderabbit": {"tp": 3, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 5, "fn": 3}, "kg": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 3, "fp": 1, "fn": 1}, "mra-max": {"tp": 1, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 7, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 3, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 4}, "mra-a": {"tp": 3, "fp": 2, "fn": 1}, "entelligence": {"tp": 2, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 7, "fn": 2}, "codeant": {"tp": 2, "fp": 15, "fn": 2}, "kodus": {"tp": 2, "fp": 1, "fn": 2}, "claude-code": {"tp": 2, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 4, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/97529", "language": "Go", "pr_size": "small", "domain": "concurrency", "change_type": "performance", "complexity": "complex", "difficulty": "very_subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 5, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 1, "fp": 1, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 8, "fn": 1}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80168", "language": "Python", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 6, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 2, "fp": 4, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/getsentry/sentry/pull/80528", "language": "Python", "pr_size": "small", "domain": "scheduling", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 1, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 1, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 1, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 1, "fn": 2}, "codeant": {"tp": 1, "fp": 6, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/77754", "language": "Python", "pr_size": "medium", "domain": "serialization", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 3}, "claude": {"tp": 2, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 0, "fn": 3}, "baz": {"tp": 1, "fp": 0, "fn": 3}, "coderabbit": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 3, "fn": 2}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 3}, "mra-b": {"tp": 2, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 3}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 2, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 0, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 3}, "macroscope": {"tp": 1, "fp": 0, "fn": 3}}}, {"url": "https://github.com/getsentry/sentry/pull/95633", "language": "Python", "pr_size": "medium", "domain": "concurrency", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 2, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 0, "fp": 6, "fn": 3}, "propel": {"tp": 0, "fp": 1, "fn": 3}, "augment": {"tp": 0, "fp": 3, "fn": 3}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 0, "fp": 7, "fn": 3}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 4, "fn": 3}, "devin": {"tp": 0, "fp": 2, "fn": 3}, "mra-max": {"tp": 0, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 2, "fn": 3}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 1, "fp": 1, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 0, "fp": 4, "fn": 3}, "codeant": {"tp": 1, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 0, "fp": 5, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 0, "fp": 7, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2", "language": "Python", "pr_size": "medium", "domain": "data_processing", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 3}, "claude": {"tp": 2, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 3, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 2, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 0, "fn": 3}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 5, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 9, "fn": 1}, "mra-b": {"tp": 3, "fp": 9, "fn": 0}, "mra-a": {"tp": 3, "fp": 6, "fn": 0}, "entelligence": {"tp": 3, "fp": 1, "fn": 0}, "sourcery": {"tp": 2, "fp": 8, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 16, "fn": 1}, "codeant": {"tp": 2, "fp": 7, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 5, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3", "language": "Python", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 1, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 2, "fp": 5, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 7, "fn": 1}, "mra-b": {"tp": 2, "fp": 4, "fn": 1}, "mra-a": {"tp": 2, "fp": 3, "fn": 1}, "entelligence": {"tp": 1, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 9, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 2}, "codeant": {"tp": 1, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "linearb": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/103633", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 6, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 5, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 9, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 2}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 0, "fp": 8, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 6, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}}}, {"url": "https://github.com/getsentry/sentry/pull/67876", "language": "Python", "pr_size": "small", "domain": "authentication", "change_type": "security_patch", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 2, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 1, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 8, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 1, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 8, "fn": 2}, "kodus": {"tp": 0, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/32918", "language": "Java", "pr_size": "small", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 3, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 0, "fp": 2, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 0, "fp": 4, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 0, "fp": 3, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/94942", "language": "Go", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 0, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 9, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 2, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 1}, "mesa": {"tp": 2, "fp": 1, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 2, "fp": 0, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/90939", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 0, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 0, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 1, "fn": 1}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 3, "fn": 1}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/80329", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "performance", "complexity": "simple", "difficulty": "obvious", "risk": "low", "context": "local", "concern": "maintainability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 0}, "copilot": {"tp": 1, "fp": 4, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "coderabbit": {"tp": 1, "fp": 2, "fn": 0}, "greptile": {"tp": 1, "fp": 3, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 0}, "devin": {"tp": 1, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 4, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 6, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 1}, "mra-a": {"tp": 0, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 9, "fn": 0}, "codeant": {"tp": 1, "fp": 8, "fn": 0}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 3, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90045", "language": "Go", "pr_size": "medium", "domain": "logging", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 4, "fn": 0}, "claude": {"tp": 3, "fp": 3, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 4, "fn": 0}, "bugbot": {"tp": 3, "fp": 3, "fn": 0}, "copilot": {"tp": 3, "fp": 7, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 3, "fp": 5, "fn": 0}, "baz": {"tp": 2, "fp": 2, "fn": 1}, "coderabbit": {"tp": 3, "fp": 2, "fn": 0}, "greptile": {"tp": 3, "fp": 1, "fn": 0}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 6, "fn": 1}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 3}, "mra-b": {"tp": 1, "fp": 1, "fn": 2}, "mra-a": {"tp": 0, "fp": 3, "fn": 3}, "entelligence": {"tp": 0, "fp": 4, "fn": 3}, "sourcery": {"tp": 3, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 3, "fp": 8, "fn": 0}, "codeant": {"tp": 3, "fp": 13, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 6, "fn": 1}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/106778", "language": "Go", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 10, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "copilot": {"tp": 1, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 6, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 2, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 2, "fn": 2}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 8, "fn": 2}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 0, "fp": 11, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/107534", "language": "Go", "pr_size": "small", "domain": "testing", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "low", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 4, "fn": 1}, "augment": {"tp": 0, "fp": 0, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 0}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/79265", "language": "Go", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 4}, "claude": {"tp": 1, "fp": 2, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 2, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 7, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 5}, "augment": {"tp": 0, "fp": 2, "fn": 5}, "propel": {"tp": 1, "fp": 1, "fn": 4}, "baz": {"tp": 0, "fp": 1, "fn": 5}, "coderabbit": {"tp": 3, "fp": 4, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 4}, "kg": {"tp": 1, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 2, "fp": 10, "fn": 3}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 1, "fp": 3, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 15, "fn": 4}, "mra-b": {"tp": 2, "fp": 3, "fn": 3}, "mra-a": {"tp": 2, "fp": 4, "fn": 3}, "entelligence": {"tp": 0, "fp": 0, "fn": 5}, "sourcery": {"tp": 2, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 4, "fp": 5, "fn": 1}, "mesa": {"tp": 2, "fp": 7, "fn": 3}, "codeant": {"tp": 4, "fp": 5, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 4}, "claude-code": {"tp": 0, "fp": 2, "fn": 5}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 4}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 5}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 4}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 3}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 5}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9", "language": "Ruby", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 0, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 3, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 10, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 6, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 4, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/76186", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 4, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "mra-a": {"tp": 2, "fp": 3, "fn": 0}, "entelligence": {"tp": 0, "fp": 5, "fn": 2}, "sourcery": {"tp": 0, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10", "language": "Ruby", "pr_size": "large", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 10, "fn": 2}, "claude": {"tp": 0, "fp": 3, "fn": 4}, "qodo": {"tp": 2, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "bugbot": {"tp": 3, "fp": 4, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 3, "fp": 0, "fn": 1}, "augment": {"tp": 3, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 2, "fp": 9, "fn": 2}, "greptile": {"tp": 2, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 2}, "devin": {"tp": 0, "fp": 2, "fn": 4}, "mra-max": {"tp": 0, "fp": 25, "fn": 4}, "greptile-v4": {"tp": 0, "fp": 3, "fn": 4}, "mra-nano": {"tp": 0, "fp": 17, "fn": 4}, "mra-ultra": {"tp": 2, "fp": 10, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 3}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 9, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 11, "fn": 2}, "kodus": {"tp": 3, "fp": 20, "fn": 1}, "claude-code": {"tp": 0, "fp": 6, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 3, "fp": 8, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 2}, "macroscope": {"tp": 3, "fp": 4, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "bug_fix", "complexity": "simple", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 2, "fn": 0}, "claude": {"tp": 3, "fp": 4, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 1, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 2, "fn": 3}, "greptile": {"tp": 1, "fp": 4, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 3, "fp": 2, "fn": 0}, "devin": {"tp": 3, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 20, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 3}, "mra-b": {"tp": 0, "fp": 6, "fn": 3}, "mra-a": {"tp": 3, "fp": 3, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 1, "fp": 1, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 1}, "mesa": {"tp": 1, "fp": 2, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8", "language": "Ruby", "pr_size": "medium", "domain": "API", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 2}, "claude": {"tp": 2, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 2, "fp": 8, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 14, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 2}, "entelligence": {"tp": 2, "fp": 4, "fn": 1}, "sourcery": {"tp": 0, "fp": 5, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 2, "fp": 8, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 6, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 2}, "macroscope": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3", "language": "Ruby", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 7, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 0, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 6, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 6, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 9, "fn": 0}, "kodus": {"tp": 2, "fp": 5, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5", "language": "Ruby", "pr_size": "small", "domain": "UI", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 13, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 7, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 3, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 5, "fn": 1}, "mesa": {"tp": 1, "fp": 11, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6", "language": "Ruby", "pr_size": "small", "domain": "serialization", "change_type": "feature", "complexity": "simple", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 0}, "copilot": {"tp": 1, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 0}, "baz": {"tp": 1, "fp": 1, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 0}, "greptile": {"tp": 0, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 2, "fn": 0}, "mra-max": {"tp": 1, "fp": 10, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 2, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 2, "fn": 0}, "mra-a": {"tp": 0, "fp": 4, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 0}, "sourcery": {"tp": 1, "fp": 4, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 0}, "mesa": {"tp": 0, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 13, "fn": 0}, "kodus": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4", "language": "Ruby", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 5, "fp": 4, "fn": 1}, "claude": {"tp": 3, "fp": 3, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 6}, "qodo": {"tp": 2, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 4, "fn": 5}, "copilot": {"tp": 1, "fp": 3, "fn": 5}, "augment": {"tp": 3, "fp": 2, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 5}, "baz": {"tp": 1, "fp": 3, "fn": 5}, "coderabbit": {"tp": 4, "fp": 20, "fn": 2}, "greptile": {"tp": 4, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 6}, "qodo-v2": {"tp": 3, "fp": 8, "fn": 3}, "devin": {"tp": 1, "fp": 3, "fn": 5}, "mra-max": {"tp": 1, "fp": 29, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 4}, "mra-nano": {"tp": 2, "fp": 3, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 7, "fn": 5}, "mra-b": {"tp": 4, "fp": 6, "fn": 2}, "mra-a": {"tp": 3, "fp": 4, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 5}, "sourcery": {"tp": 2, "fp": 3, "fn": 4}, "cubic-dev": {"tp": 4, "fp": 25, "fn": 2}, "mesa": {"tp": 4, "fp": 5, "fn": 2}, "codeant": {"tp": 3, "fp": 10, "fn": 3}, "kodus": {"tp": 3, "fp": 19, "fn": 3}, "claude-code": {"tp": 3, "fp": 1, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 6}, "bito": {"tp": 0, "fp": 0, "fn": 6}, "sentry": {"tp": 0, "fp": 0, "fn": 6}, "linearb": {"tp": 0, "fp": 1, "fn": 6}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 4}, "qodo-v2-2": {"tp": 3, "fp": 3, "fn": 3}, "qodo-v22": {"tp": 3, "fp": 3, "fn": 3}, "qodo-extended": {"tp": 5, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 5, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 5, "fn": 5}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1", "language": "Ruby", "pr_size": "medium", "domain": "file_io", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 1, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 2, "fp": 0, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "coderabbit": {"tp": 2, "fp": 6, "fn": 1}, "greptile": {"tp": 2, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 0}, "mra-nano": {"tp": 2, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 1, "fn": 2}, "mra-a": {"tp": 1, "fp": 1, "fn": 2}, "entelligence": {"tp": 2, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 1}, "codeant": {"tp": 2, "fp": 9, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 2, "fp": 6, "fn": 0}, "greptile": {"tp": 2, "fp": 1, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 18, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 2, "fp": 8, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 2, "fp": 4, "fn": 0}, "mra-a": {"tp": 1, "fp": 6, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 8, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22532", "language": "TypeScript", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 11, "fn": 0}, "augment": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 2, "fp": 2, "fn": 0}, "baz": {"tp": 1, "fp": 2, "fn": 1}, "coderabbit": {"tp": 1, "fp": 10, "fn": 1}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 14, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 1, "fp": 3, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 0, "fp": 4, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/8330", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 0}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 2, "fp": 0, "fn": 0}, "qodo": {"tp": 2, "fp": 4, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 0}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 0}, "coderabbit": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 2, "fp": 0, "fn": 0}, "mra-max": {"tp": 2, "fp": 12, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 2, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 2, "fp": 3, "fn": 0}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14943", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 0, "fn": 1}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 4, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22345", "language": "TypeScript", "pr_size": "small", "domain": "database", "change_type": "migration", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 2, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 0, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 6, "fn": 2}, "kodus": {"tp": 0, "fp": 1, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 2, "fn": 2}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/11059", "language": "TypeScript", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 3, "fn": 2}, "claude": {"tp": 3, "fp": 3, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 3, "fp": 7, "fn": 2}, "bugbot": {"tp": 4, "fp": 4, "fn": 1}, "copilot": {"tp": 4, "fp": 9, "fn": 1}, "augment": {"tp": 5, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 4, "fp": 1, "fn": 1}, "coderabbit": {"tp": 5, "fp": 14, "fn": 0}, "greptile": {"tp": 4, "fp": 7, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 4, "fn": 1}, "devin": {"tp": 4, "fp": 1, "fn": 1}, "mra-max": {"tp": 3, "fp": 12, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 7, "fn": 1}, "mra-nano": {"tp": 2, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 2}, "mra-b": {"tp": 4, "fp": 4, "fn": 1}, "mra-a": {"tp": 4, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 5}, "sourcery": {"tp": 5, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 5, "fp": 18, "fn": 0}, "mesa": {"tp": 2, "fp": 13, "fn": 3}, "codeant": {"tp": 4, "fp": 13, "fn": 1}, "kodus": {"tp": 3, "fp": 5, "fn": 2}, "claude-code": {"tp": 3, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 5, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 5, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 5, "fp": 6, "fn": 0}, "macroscope": {"tp": 4, "fp": 7, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/7232", "language": "TypeScript", "pr_size": "medium", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 3, "fn": 2}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 4, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 18, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 4, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 5, "fn": 0}, "mesa": {"tp": 0, "fp": 9, "fn": 2}, "codeant": {"tp": 2, "fp": 9, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/14740", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 4}, "claude": {"tp": 2, "fp": 0, "fn": 3}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 3, "fn": 2}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "augment": {"tp": 4, "fp": 3, "fn": 1}, "propel": {"tp": 4, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 4, "fp": 6, "fn": 1}, "greptile": {"tp": 4, "fp": 1, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 4, "fp": 3, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 11, "fn": 3}, "greptile-v4": {"tp": 4, "fp": 6, "fn": 1}, "mra-nano": {"tp": 0, "fp": 6, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 5}, "mra-b": {"tp": 2, "fp": 0, "fn": 3}, "mra-a": {"tp": 1, "fp": 4, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 5}, "sourcery": {"tp": 3, "fp": 1, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 11, "fn": 1}, "mesa": {"tp": 3, "fp": 3, "fn": 2}, "codeant": {"tp": 2, "fp": 12, "fn": 3}, "kodus": {"tp": 1, "fp": 7, "fn": 4}, "claude-code": {"tp": 1, "fp": 2, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 4, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 4, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/10600", "language": "TypeScript", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 8, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 3}, "copilot": {"tp": 1, "fp": 9, "fn": 3}, "augment": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 0, "fp": 4, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 0, "fp": 13, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 7, "fn": 2}, "mra-nano": {"tp": 1, "fp": 6, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 6, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 7, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 11, "fn": 1}, "mesa": {"tp": 1, "fp": 5, "fn": 3}, "codeant": {"tp": 2, "fp": 12, "fn": 2}, "kodus": {"tp": 0, "fp": 8, "fn": 4}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 2, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/10967", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "bug_fix", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 9, "fn": 4}, "claude": {"tp": 3, "fp": 4, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 0, "fp": 3, "fn": 5}, "bugbot": {"tp": 2, "fp": 4, "fn": 3}, "copilot": {"tp": 4, "fp": 7, "fn": 1}, "augment": {"tp": 2, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 5}, "greptile": {"tp": 2, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 4}, "devin": {"tp": 2, "fp": 3, "fn": 3}, "mra-max": {"tp": 2, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 5, "fn": 2}, "mra-nano": {"tp": 0, "fp": 15, "fn": 5}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 2}, "mra-b": {"tp": 3, "fp": 6, "fn": 2}, "mra-a": {"tp": 3, "fp": 15, "fn": 2}, "entelligence": {"tp": 3, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 3, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 5}, "mesa": {"tp": 3, "fp": 9, "fn": 2}, "codeant": {"tp": 3, "fp": 6, "fn": 2}, "kodus": {"tp": 3, "fp": 6, "fn": 2}, "claude-code": {"tp": 2, "fp": 5, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 5, "fn": 3}, "qodo-v22": {"tp": 3, "fp": 3, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 4, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 3, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/8087", "language": "TypeScript", "pr_size": "medium", "domain": "concurrency", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 4, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 3, "fn": 1}, "mra-max": {"tp": 1, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 4, "fn": 0}, "mra-b": {"tp": 2, "fp": 2, "fn": 0}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 2, "fp": 8, "fn": 0}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 10, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}], "tools": ["augment", "baz", "bito", "bugbot", "claude", "claude-code", "codeant", "coderabbit", "copilot", "cubic-dev", "cubic-v2", "devin", "entelligence", "gemini", "graphite", "greptile", "greptile-v4", "kg", "kodus", "kodus-v2", "linearb", "macroscope", "mesa", "mra-a", "mra-b", "mra-max", "mra-nano", "mra-ultra", "propel", "qodo", "qodo-extended", "qodo-extended-summary", "qodo-v2", "qodo-v2-2", "qodo-v22", "sentry", "sourcery", "vercel"], "dimensions": {"language": ["Go", "Java", "Python", "Ruby", "TypeScript"], "pr_size": ["small", "medium", "large"], "domain": ["API", "UI", "authentication", "caching", "concurrency", "configuration", "data_processing", "database", "file_io", "logging", "scheduling", "serialization", "testing"], "change_type": ["bug_fix", "feature", "migration", "performance", "refactoring", "security_patch"], "complexity": ["simple", "moderate", "complex"], "difficulty": ["obvious", "moderate", "subtle", "very_subtle"], "risk": ["low", "medium", "high", "critical"], "context": ["local", "file", "cross_file", "system"], "concern": ["correctness", "maintainability", "reliability", "security"]}, "overall_metrics": {"greptile": {"precision": 41.5, "recall": 39.4, "f1": 40.4, "tp": 54, "fp": 76, "fn": 83, "num_prs": 50}, "qodo-v2-2": {"precision": 47.1, "recall": 41.6, "f1": 44.2, "tp": 57, "fp": 64, "fn": 80, "num_prs": 50}, "qodo-v2": {"precision": 42.9, "recall": 55.5, "f1": 48.4, "tp": 76, "fp": 101, "fn": 61, "num_prs": 50}, "claude-code": {"precision": 34.8, "recall": 40.9, "f1": 37.6, "tp": 56, "fp": 105, "fn": 81, "num_prs": 50}, "cubic-v2": {"precision": 56.3, "recall": 68.6, "f1": 61.8, "tp": 94, "fp": 73, "fn": 43, "num_prs": 50}, "augment": {"precision": 47.5, "recall": 61.3, "f1": 53.5, "tp": 84, "fp": 93, "fn": 53, "num_prs": 50}, "qodo-extended-summary": {"precision": 40.2, "recall": 67.2, "f1": 50.3, "tp": 92, "fp": 137, "fn": 45, "num_prs": 50}, "claude": {"precision": 34.8, "recall": 35.8, "f1": 35.3, "tp": 49, "fp": 92, "fn": 88, "num_prs": 50}, "mra-b": {"precision": 25.0, "recall": 41.6, "f1": 31.2, "tp": 57, "fp": 171, "fn": 80, "num_prs": 50}, "greptile-v4": {"precision": 33.1, "recall": 56.9, "f1": 41.8, "tp": 78, "fp": 158, "fn": 59, "num_prs": 50}, "kodus-v2": {"precision": 46.7, "recall": 35.8, "f1": 40.5, "tp": 49, "fp": 56, "fn": 88, "num_prs": 50}, "mra-max": {"precision": 8.9, "recall": 34.3, "f1": 14.1, "tp": 47, "fp": 481, "fn": 90, "num_prs": 50}, "gemini": {"precision": 31.1, "recall": 37.2, "f1": 33.9, "tp": 51, "fp": 113, "fn": 86, "num_prs": 50}, "codeant": {"precision": 13.9, "recall": 48.9, "f1": 21.6, "tp": 67, "fp": 415, "fn": 70, "num_prs": 50}, "bugbot": {"precision": 47.2, "recall": 43.8, "f1": 45.5, "tp": 60, "fp": 67, "fn": 77, "num_prs": 50}, "mra-a": {"precision": 23.6, "recall": 45.2, "f1": 31.0, "tp": 61, "fp": 198, "fn": 74, "num_prs": 49}, "sourcery": {"precision": 33.3, "recall": 51.8, "f1": 40.6, "tp": 71, "fp": 142, "fn": 66, "num_prs": 50}, "bito": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "vercel": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "copilot": {"precision": 28.3, "recall": 53.3, "f1": 37.0, "tp": 73, "fp": 185, "fn": 64, "num_prs": 50}, "graphite": {"precision": 100.0, "recall": 8.8, "f1": 16.1, "tp": 12, "fp": 0, "fn": 125, "num_prs": 50}, "mesa": {"precision": 19.3, "recall": 45.3, "f1": 27.1, "tp": 62, "fp": 259, "fn": 75, "num_prs": 50}, "kodus": {"precision": 22.9, "recall": 44.5, "f1": 30.3, "tp": 61, "fp": 205, "fn": 76, "num_prs": 50}, "kg": {"precision": 50.0, "recall": 16.8, "f1": 25.1, "tp": 23, "fp": 23, "fn": 114, "num_prs": 50}, "devin": {"precision": 54.3, "recall": 37.2, "f1": 44.2, "tp": 51, "fp": 43, "fn": 86, "num_prs": 50}, "baz": {"precision": 48.8, "recall": 29.2, "f1": 36.5, "tp": 40, "fp": 42, "fn": 97, "num_prs": 50}, "sentry": {"precision": 0.0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 1, "fn": 137, "num_prs": 50}, "macroscope": {"precision": 48.4, "recall": 43.8, "f1": 46.0, "tp": 60, "fp": 64, "fn": 77, "num_prs": 50}, "mra-nano": {"precision": 12.6, "recall": 17.5, "f1": 14.7, "tp": 24, "fp": 166, "fn": 113, "num_prs": 50}, "qodo-extended": {"precision": 37.2, "recall": 62.8, "f1": 46.7, "tp": 86, "fp": 145, "fn": 51, "num_prs": 50}, "cubic-dev": {"precision": 30.1, "recall": 75.9, "f1": 43.2, "tp": 104, "fp": 241, "fn": 33, "num_prs": 50}, "propel": {"precision": 52.5, "recall": 38.7, "f1": 44.5, "tp": 53, "fp": 48, "fn": 84, "num_prs": 50}, "qodo-v22": {"precision": 44.6, "recall": 54.7, "f1": 49.2, "tp": 75, "fp": 93, "fn": 62, "num_prs": 50}, "linearb": {"precision": 8.3, "recall": 2.9, "f1": 4.3, "tp": 4, "fp": 44, "fn": 133, "num_prs": 50}, "qodo": {"precision": 31.8, "recall": 44.5, "f1": 37.1, "tp": 61, "fp": 131, "fn": 76, "num_prs": 50}, "mra-ultra": {"precision": 18.3, "recall": 37.2, "f1": 24.6, "tp": 51, "fp": 227, "fn": 86, "num_prs": 50}, "entelligence": {"precision": 30.1, "recall": 20.4, "f1": 24.3, "tp": 28, "fp": 65, "fn": 109, "num_prs": 50}, "coderabbit": {"precision": 24.7, "recall": 39.4, "f1": 30.3, "tp": 54, "fp": 165, "fn": 83, "num_prs": 50}}}, "anthropic_claude-sonnet-4-5-20250929": {"prs": [{"url": "https://github.com/keycloak/keycloak/pull/37429", "language": "Java", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "gemini": {"tp": 0, "fp": 4, "fn": 4}, "claude": {"tp": 1, "fp": 0, "fn": 3}, "augment": {"tp": 2, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 4}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 8, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-a": {"tp": 0, "fp": 4, "fn": 4}, "mra-b": {"tp": 0, "fp": 2, "fn": 4}, "entelligence": {"tp": 0, "fp": 2, "fn": 4}, "sourcery": {"tp": 1, "fp": 5, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 14, "fn": 1}, "mesa": {"tp": 1, "fp": 7, "fn": 3}, "codeant": {"tp": 2, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 3}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 3, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 4}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 4}}}, {"url": "https://github.com/keycloak/keycloak/pull/37634", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 3, "fn": 2}, "gemini": {"tp": 3, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "augment": {"tp": 2, "fp": 1, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "graphite": {"tp": 2, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 2}, "bugbot": {"tp": 2, "fp": 2, "fn": 2}, "copilot": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 4, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-b": {"tp": 2, "fp": 4, "fn": 2}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 0, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 9, "fn": 2}, "codeant": {"tp": 2, "fp": 8, "fn": 2}, "kodus": {"tp": 2, "fp": 10, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "claude-code": {"tp": 0, "fp": 2, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 2}, "macroscope": {"tp": 2, "fp": 2, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/38446", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 2}, "propel": {"tp": 0, "fp": 4, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 1, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 2}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 1, "fp": 6, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 0}, "macroscope": {"tp": 0, "fp": 3, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/36882", "language": "Java", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "augment": {"tp": 0, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 10, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 10, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 0, "fp": 1, "fn": 1}, "codeant": {"tp": 0, "fp": 13, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36880", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 5, "fn": 3}, "gemini": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 2, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 4, "fn": 3}, "copilot": {"tp": 0, "fp": 7, "fn": 3}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 2}, "devin": {"tp": 0, "fp": 1, "fn": 3}, "mra-max": {"tp": 1, "fp": 7, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 5, "fn": 3}, "sourcery": {"tp": 1, "fp": 5, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 0, "fp": 7, "fn": 3}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/37038", "language": "Java", "pr_size": "small", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"qodo": {"tp": 1, "fp": 2, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 8, "fn": 1}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 1, "fp": 7, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 10, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 2, "fp": 9, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 0}, "macroscope": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/33832", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 1, "fp": 5, "fn": 1}, "gemini": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 4, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 8, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 18, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 0, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 1, "fp": 7, "fn": 1}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 1, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/40940", "language": "Java", "pr_size": "small", "domain": "concurrency", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 1, "fn": 2}, "gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 2, "fp": 4, "fn": 0}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 13, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 3, "fn": 1}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"augment": {"tp": 2, "fp": 0, "fn": 0}, "gemini": {"tp": 2, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 4, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 0, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 4, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 2, "fp": 4, "fn": 0}, "sourcery": {"tp": 2, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 0, "fp": 5, "fn": 2}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 1, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 0, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/getsentry/sentry/pull/93824", "language": "Python", "pr_size": "large", "domain": "concurrency", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 0, "fn": 5}, "augment": {"tp": 5, "fp": 1, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 4}, "coderabbit": {"tp": 1, "fp": 3, "fn": 4}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "claude": {"tp": 0, "fp": 2, "fn": 5}, "qodo": {"tp": 1, "fp": 2, "fn": 4}, "copilot": {"tp": 1, "fp": 4, "fn": 4}, "propel": {"tp": 1, "fp": 2, "fn": 4}, "greptile": {"tp": 1, "fp": 1, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 2, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 0, "fp": 7, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 5}, "mra-b": {"tp": 1, "fp": 8, "fn": 4}, "mra-a": {"tp": 1, "fp": 9, "fn": 4}, "entelligence": {"tp": 1, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 5}, "codeant": {"tp": 2, "fp": 5, "fn": 3}, "kodus": {"tp": 3, "fp": 2, "fn": 2}, "claude-code": {"tp": 3, "fp": 3, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 5}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 4}, "cubic-v2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 3}, "macroscope": {"tp": 2, "fp": 1, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5", "language": "Python", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 3}, "greptile": {"tp": 0, "fp": 2, "fn": 3}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "coderabbit": {"tp": 1, "fp": 32, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 9, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 9, "fn": 1}, "mra-a": {"tp": 1, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 0, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 2, "fp": 28, "fn": 1}, "mesa": {"tp": 0, "fp": 18, "fn": 3}, "codeant": {"tp": 1, "fp": 15, "fn": 2}, "kodus": {"tp": 1, "fp": 17, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 1, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 2}, "macroscope": {"tp": 0, "fp": 5, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1", "language": "Python", "pr_size": "medium", "domain": "API", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 3}, "claude": {"tp": 2, "fp": 7, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 1, "fp": 2, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 1, "fn": 1}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "coderabbit": {"tp": 3, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 3}, "kg": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 3, "fp": 2, "fn": 1}, "mra-max": {"tp": 1, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 6, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 3, "fn": 4}, "mra-a": {"tp": 3, "fp": 3, "fn": 1}, "entelligence": {"tp": 2, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 1}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 3, "fp": 14, "fn": 1}, "kodus": {"tp": 2, "fp": 3, "fn": 2}, "claude-code": {"tp": 2, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/97529", "language": "Go", "pr_size": "small", "domain": "concurrency", "change_type": "performance", "complexity": "complex", "difficulty": "very_subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 4, "fn": 0}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 6, "fn": 1}, "codeant": {"tp": 1, "fp": 9, "fn": 1}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80168", "language": "Python", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 8, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 2, "fp": 4, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 0, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/getsentry/sentry/pull/80528", "language": "Python", "pr_size": "small", "domain": "scheduling", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 1, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 1, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 1, "fn": 1}, "codeant": {"tp": 1, "fp": 5, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/77754", "language": "Python", "pr_size": "medium", "domain": "serialization", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 3}, "claude": {"tp": 2, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 0, "fn": 3}, "baz": {"tp": 1, "fp": 0, "fn": 3}, "coderabbit": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 2, "fn": 2}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 3}, "mra-b": {"tp": 2, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 3, "fn": 3}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 2, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 0, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 3}, "macroscope": {"tp": 1, "fp": 0, "fn": 3}}}, {"url": "https://github.com/getsentry/sentry/pull/95633", "language": "Python", "pr_size": "medium", "domain": "concurrency", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 2, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 0, "fp": 6, "fn": 3}, "propel": {"tp": 0, "fp": 1, "fn": 3}, "augment": {"tp": 0, "fp": 6, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 3}, "coderabbit": {"tp": 0, "fp": 7, "fn": 3}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 4, "fn": 3}, "devin": {"tp": 0, "fp": 2, "fn": 3}, "mra-max": {"tp": 1, "fp": 4, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 2, "fn": 3}, "mra-b": {"tp": 1, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 5, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 0, "fp": 4, "fn": 3}, "codeant": {"tp": 1, "fp": 12, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 0, "fp": 5, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 0, "fp": 7, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2", "language": "Python", "pr_size": "medium", "domain": "data_processing", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 3}, "claude": {"tp": 2, "fp": 6, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 3, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 2, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 0, "fn": 3}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 5, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 9, "fn": 1}, "mra-b": {"tp": 3, "fp": 7, "fn": 0}, "mra-a": {"tp": 3, "fp": 5, "fn": 0}, "entelligence": {"tp": 3, "fp": 1, "fn": 0}, "sourcery": {"tp": 2, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 24, "fn": 1}, "codeant": {"tp": 2, "fp": 7, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 4, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 5, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3", "language": "Python", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 2, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 0}, "mra-b": {"tp": 2, "fp": 4, "fn": 1}, "mra-a": {"tp": 2, "fp": 6, "fn": 1}, "entelligence": {"tp": 1, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 9, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 2}, "codeant": {"tp": 1, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "linearb": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/103633", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 1, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 5, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 3, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 5, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 8, "fn": 1}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 0, "fp": 7, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 0, "fp": 6, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}}}, {"url": "https://github.com/getsentry/sentry/pull/67876", "language": "Python", "pr_size": "small", "domain": "authentication", "change_type": "security_patch", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 2, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 7, "fn": 1}, "bugbot": {"tp": 1, "fp": 2, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 2}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 1, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 8, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 8, "fn": 2}, "kodus": {"tp": 0, "fp": 3, "fn": 3}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/32918", "language": "Java", "pr_size": "small", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 4, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 0, "fp": 3, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 0, "fp": 3, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/94942", "language": "Go", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 0, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 0, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 11, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 2, "fp": 7, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 1}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 10, "fn": 0}, "kodus": {"tp": 2, "fp": 0, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/90939", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 0, "fn": 1}, "mra-nano": {"tp": 1, "fp": 2, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 1, "fn": 1}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 0, "fp": 3, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 3, "fn": 1}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/80329", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "performance", "complexity": "simple", "difficulty": "obvious", "risk": "low", "context": "local", "concern": "maintainability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 0}, "copilot": {"tp": 1, "fp": 4, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 3, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "coderabbit": {"tp": 1, "fp": 2, "fn": 0}, "greptile": {"tp": 1, "fp": 3, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 0}, "devin": {"tp": 1, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 5, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 6, "fn": 1}, "mra-a": {"tp": 0, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 9, "fn": 0}, "codeant": {"tp": 1, "fp": 8, "fn": 0}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 6, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90045", "language": "Go", "pr_size": "medium", "domain": "logging", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 4, "fn": 0}, "claude": {"tp": 3, "fp": 3, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 5, "fn": 0}, "bugbot": {"tp": 3, "fp": 3, "fn": 0}, "copilot": {"tp": 3, "fp": 8, "fn": 0}, "propel": {"tp": 2, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 5, "fn": 0}, "baz": {"tp": 2, "fp": 2, "fn": 1}, "coderabbit": {"tp": 3, "fp": 2, "fn": 0}, "greptile": {"tp": 3, "fp": 1, "fn": 0}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 6, "fn": 1}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 3, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 9, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 3}, "mra-b": {"tp": 1, "fp": 2, "fn": 2}, "mra-a": {"tp": 0, "fp": 3, "fn": 3}, "entelligence": {"tp": 0, "fp": 4, "fn": 3}, "sourcery": {"tp": 3, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 3, "fp": 1, "fn": 0}, "mesa": {"tp": 3, "fp": 9, "fn": 0}, "codeant": {"tp": 3, "fp": 11, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 3, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 6, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/106778", "language": "Go", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 12, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 4, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "copilot": {"tp": 1, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 5, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 6, "fn": 1}, "devin": {"tp": 2, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 14, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 0, "fp": 12, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/107534", "language": "Go", "pr_size": "small", "domain": "testing", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "low", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 1}, "augment": {"tp": 0, "fp": 0, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 5, "fn": 0}, "mra-nano": {"tp": 0, "fp": 7, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 0}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/79265", "language": "Go", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 4}, "claude": {"tp": 1, "fp": 2, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 2, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 5}, "augment": {"tp": 1, "fp": 1, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 4}, "baz": {"tp": 0, "fp": 1, "fn": 5}, "coderabbit": {"tp": 3, "fp": 4, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 4}, "kg": {"tp": 1, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 3}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 1, "fp": 11, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 2, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 15, "fn": 3}, "mra-b": {"tp": 2, "fp": 3, "fn": 3}, "mra-a": {"tp": 2, "fp": 4, "fn": 3}, "entelligence": {"tp": 0, "fp": 0, "fn": 5}, "sourcery": {"tp": 2, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 4, "fp": 5, "fn": 1}, "mesa": {"tp": 3, "fp": 6, "fn": 2}, "codeant": {"tp": 4, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 4}, "claude-code": {"tp": 0, "fp": 3, "fn": 5}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 4}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 5}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 4}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 3}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 5}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9", "language": "Ruby", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 0, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 3, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 10, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 0, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 0, "fp": 12, "fn": 2}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/76186", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 0, "fp": 2, "fn": 2}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 9, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 2, "fp": 4, "fn": 0}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 6, "fn": 2}, "sourcery": {"tp": 0, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 8, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10", "language": "Ruby", "pr_size": "large", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 8, "fn": 2}, "claude": {"tp": 0, "fp": 3, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "bugbot": {"tp": 3, "fp": 4, "fn": 1}, "copilot": {"tp": 2, "fp": 4, "fn": 2}, "propel": {"tp": 3, "fp": 0, "fn": 1}, "augment": {"tp": 3, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 2, "fp": 10, "fn": 2}, "greptile": {"tp": 2, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 2}, "devin": {"tp": 0, "fp": 2, "fn": 4}, "mra-max": {"tp": 0, "fp": 18, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 5, "fn": 2}, "mra-nano": {"tp": 0, "fp": 17, "fn": 4}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 3}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 8, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 13, "fn": 2}, "kodus": {"tp": 3, "fp": 19, "fn": 1}, "claude-code": {"tp": 0, "fp": 6, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 7, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 2}, "macroscope": {"tp": 3, "fp": 5, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "bug_fix", "complexity": "simple", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 2, "fn": 0}, "claude": {"tp": 3, "fp": 4, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 1, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 2}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 2, "fn": 3}, "greptile": {"tp": 2, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 3, "fp": 2, "fn": 0}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 1, "fp": 18, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 3}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 2, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 3, "fp": 2, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 1}, "kodus": {"tp": 2, "fp": 3, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 2, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8", "language": "Ruby", "pr_size": "medium", "domain": "API", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 1}, "claude": {"tp": 2, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 1, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 2, "fp": 8, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 15, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 1}, "mra-nano": {"tp": 2, "fp": 8, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 5, "fn": 3}, "entelligence": {"tp": 2, "fp": 4, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 2, "fp": 8, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 6, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 2}, "macroscope": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3", "language": "Ruby", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 6, "fn": 1}, "copilot": {"tp": 0, "fp": 2, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 3, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 0, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 2}, "mra-nano": {"tp": 1, "fp": 7, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 7, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5", "language": "Ruby", "pr_size": "small", "domain": "UI", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 12, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 8, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "mra-a": {"tp": 1, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 11, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6", "language": "Ruby", "pr_size": "small", "domain": "serialization", "change_type": "feature", "complexity": "simple", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 1, "fp": 4, "fn": 0}, "copilot": {"tp": 1, "fp": 2, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 0}, "baz": {"tp": 1, "fp": 1, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 0}, "greptile": {"tp": 0, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 0}, "mra-max": {"tp": 1, "fp": 11, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 10, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 2, "fn": 0}, "mra-a": {"tp": 0, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 1}, "sourcery": {"tp": 0, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 13, "fn": 0}, "kodus": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 5, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4", "language": "Ruby", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 4, "fp": 5, "fn": 2}, "claude": {"tp": 5, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 6}, "qodo": {"tp": 3, "fp": 4, "fn": 3}, "bugbot": {"tp": 1, "fp": 4, "fn": 5}, "copilot": {"tp": 1, "fp": 3, "fn": 5}, "augment": {"tp": 3, "fp": 2, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 5}, "baz": {"tp": 1, "fp": 3, "fn": 5}, "coderabbit": {"tp": 3, "fp": 22, "fn": 3}, "greptile": {"tp": 4, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 6}, "qodo-v2": {"tp": 4, "fp": 4, "fn": 2}, "devin": {"tp": 1, "fp": 3, "fn": 5}, "mra-max": {"tp": 1, "fp": 19, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 4}, "mra-nano": {"tp": 2, "fp": 10, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 7, "fn": 5}, "mra-b": {"tp": 4, "fp": 6, "fn": 2}, "mra-a": {"tp": 3, "fp": 4, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 5}, "sourcery": {"tp": 2, "fp": 3, "fn": 4}, "cubic-dev": {"tp": 4, "fp": 25, "fn": 2}, "mesa": {"tp": 2, "fp": 9, "fn": 4}, "codeant": {"tp": 3, "fp": 17, "fn": 3}, "kodus": {"tp": 3, "fp": 18, "fn": 3}, "claude-code": {"tp": 3, "fp": 1, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 6}, "bito": {"tp": 0, "fp": 0, "fn": 6}, "sentry": {"tp": 0, "fp": 0, "fn": 6}, "linearb": {"tp": 0, "fp": 1, "fn": 6}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 4}, "qodo-v2-2": {"tp": 3, "fp": 3, "fn": 3}, "qodo-v22": {"tp": 3, "fp": 3, "fn": 3}, "qodo-extended": {"tp": 6, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 4}, "qodo-extended-summary": {"tp": 5, "fp": 2, "fn": 1}, "macroscope": {"tp": 1, "fp": 5, "fn": 5}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1", "language": "Ruby", "pr_size": "medium", "domain": "file_io", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 3, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 2, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 4, "fn": 1}, "augment": {"tp": 2, "fp": 3, "fn": 1}, "propel": {"tp": 2, "fp": 0, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 5, "fn": 2}, "greptile": {"tp": 2, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 1, "fp": 5, "fn": 2}, "entelligence": {"tp": 2, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 13, "fn": 1}, "kodus": {"tp": 2, "fp": 3, "fn": 1}, "claude-code": {"tp": 3, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 4, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 5, "fn": 1}, "greptile": {"tp": 2, "fp": 1, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 10, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 2, "fp": 8, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 2, "fp": 4, "fn": 0}, "mra-a": {"tp": 1, "fp": 11, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 8, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 2, "fp": 9, "fn": 0}, "kodus": {"tp": 1, "fp": 4, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22532", "language": "TypeScript", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 4, "fn": 2}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 11, "fn": 0}, "augment": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 1, "fp": 2, "fn": 1}, "coderabbit": {"tp": 1, "fp": 4, "fn": 1}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 15, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 8, "fn": 1}, "mra-nano": {"tp": 2, "fp": 11, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 1, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 10, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 0, "fp": 5, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/8330", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 5, "fn": 0}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 2, "fp": 0, "fn": 0}, "qodo": {"tp": 2, "fp": 4, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 0}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 3, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 2, "fp": 0, "fn": 0}, "mra-max": {"tp": 2, "fp": 14, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 2, "fp": 9, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 2, "fp": 3, "fn": 0}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14943", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 3, "fn": 1}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 1, "fp": 8, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 0, "fn": 1}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22345", "language": "TypeScript", "pr_size": "small", "domain": "database", "change_type": "migration", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 2, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 10, "fn": 1}, "mra-nano": {"tp": 0, "fp": 9, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 0, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 0, "fp": 5, "fn": 2}, "kodus": {"tp": 0, "fp": 1, "fn": 2}, "claude-code": {"tp": 1, "fp": 3, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 2, "fn": 2}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/11059", "language": "TypeScript", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 3, "fn": 2}, "claude": {"tp": 3, "fp": 3, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 3, "fp": 6, "fn": 2}, "bugbot": {"tp": 4, "fp": 3, "fn": 1}, "copilot": {"tp": 4, "fp": 10, "fn": 1}, "augment": {"tp": 5, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 4, "fp": 1, "fn": 1}, "coderabbit": {"tp": 5, "fp": 12, "fn": 0}, "greptile": {"tp": 4, "fp": 7, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 7, "fn": 1}, "devin": {"tp": 4, "fp": 1, "fn": 1}, "mra-max": {"tp": 3, "fp": 12, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 13, "fn": 1}, "mra-nano": {"tp": 4, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 2}, "mra-b": {"tp": 4, "fp": 4, "fn": 1}, "mra-a": {"tp": 4, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 5}, "sourcery": {"tp": 5, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 5, "fp": 21, "fn": 0}, "mesa": {"tp": 2, "fp": 13, "fn": 3}, "codeant": {"tp": 4, "fp": 14, "fn": 1}, "kodus": {"tp": 3, "fp": 7, "fn": 2}, "claude-code": {"tp": 2, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 5, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 5, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 5, "fp": 6, "fn": 0}, "macroscope": {"tp": 4, "fp": 6, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/7232", "language": "TypeScript", "pr_size": "medium", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 2}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 21, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 1, "fp": 18, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 9, "fn": 2}, "codeant": {"tp": 1, "fp": 11, "fn": 1}, "kodus": {"tp": 2, "fp": 11, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/14740", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 4}, "claude": {"tp": 2, "fp": 0, "fn": 3}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 3}, "bugbot": {"tp": 3, "fp": 3, "fn": 2}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "augment": {"tp": 4, "fp": 3, "fn": 1}, "propel": {"tp": 4, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 4, "fp": 3, "fn": 1}, "greptile": {"tp": 4, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 4}, "qodo-v2": {"tp": 3, "fp": 3, "fn": 2}, "devin": {"tp": 2, "fp": 0, "fn": 3}, "mra-max": {"tp": 3, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 5, "fn": 1}, "mra-nano": {"tp": 1, "fp": 9, "fn": 4}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 5}, "mra-b": {"tp": 1, "fp": 2, "fn": 4}, "mra-a": {"tp": 0, "fp": 6, "fn": 5}, "entelligence": {"tp": 0, "fp": 3, "fn": 5}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 11, "fn": 1}, "mesa": {"tp": 2, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 15, "fn": 3}, "kodus": {"tp": 1, "fp": 7, "fn": 4}, "claude-code": {"tp": 1, "fp": 2, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 4, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 4, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/10600", "language": "TypeScript", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 8, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 8, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 3}, "copilot": {"tp": 1, "fp": 9, "fn": 3}, "augment": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 0, "fp": 4, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 0, "fp": 14, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 9, "fn": 2}, "mra-nano": {"tp": 1, "fp": 11, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 8, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 8, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 4, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 11, "fn": 1}, "mesa": {"tp": 1, "fp": 5, "fn": 3}, "codeant": {"tp": 2, "fp": 12, "fn": 2}, "kodus": {"tp": 0, "fp": 8, "fn": 4}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 0, "fp": 1, "fn": 4}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/10967", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "bug_fix", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 7, "fn": 4}, "claude": {"tp": 4, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 0, "fp": 3, "fn": 5}, "bugbot": {"tp": 2, "fp": 4, "fn": 3}, "copilot": {"tp": 3, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 5}, "greptile": {"tp": 2, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 4}, "devin": {"tp": 3, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 17, "fn": 3}, "greptile-v4": {"tp": 4, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 4}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 2}, "mra-b": {"tp": 2, "fp": 11, "fn": 3}, "mra-a": {"tp": 3, "fp": 15, "fn": 2}, "entelligence": {"tp": 3, "fp": 5, "fn": 2}, "sourcery": {"tp": 2, "fp": 5, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 5}, "mesa": {"tp": 4, "fp": 15, "fn": 1}, "codeant": {"tp": 3, "fp": 5, "fn": 2}, "kodus": {"tp": 3, "fp": 5, "fn": 2}, "claude-code": {"tp": 2, "fp": 6, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 5, "fn": 3}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 2, "fp": 6, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 4, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 3, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/8087", "language": "TypeScript", "pr_size": "medium", "domain": "concurrency", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 6, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 2, "fp": 3, "fn": 0}, "mra-max": {"tp": 1, "fp": 11, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 1, "fp": 8, "fn": 1}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 2, "fp": 9, "fn": 0}, "codeant": {"tp": 1, "fp": 11, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 9, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 1, "fp": 4, "fn": 1}}}], "tools": ["augment", "baz", "bito", "bugbot", "claude", "claude-code", "codeant", "coderabbit", "copilot", "cubic-dev", "cubic-v2", "devin", "entelligence", "gemini", "graphite", "greptile", "greptile-v4", "kg", "kodus", "kodus-v2", "linearb", "macroscope", "mesa", "mra-a", "mra-b", "mra-max", "mra-nano", "mra-ultra", "propel", "qodo", "qodo-extended", "qodo-extended-summary", "qodo-v2", "qodo-v2-2", "qodo-v22", "sentry", "sourcery", "vercel"], "dimensions": {"language": ["Go", "Java", "Python", "Ruby", "TypeScript"], "pr_size": ["small", "medium", "large"], "domain": ["API", "UI", "authentication", "caching", "concurrency", "configuration", "data_processing", "database", "file_io", "logging", "scheduling", "serialization", "testing"], "change_type": ["bug_fix", "feature", "migration", "performance", "refactoring", "security_patch"], "complexity": ["simple", "moderate", "complex"], "difficulty": ["obvious", "moderate", "subtle", "very_subtle"], "risk": ["low", "medium", "high", "critical"], "context": ["local", "file", "cross_file", "system"], "concern": ["correctness", "maintainability", "reliability", "security"]}, "overall_metrics": {"greptile": {"precision": 43.7, "recall": 40.1, "f1": 41.8, "tp": 55, "fp": 71, "fn": 82, "num_prs": 50}, "qodo-v2-2": {"precision": 43.1, "recall": 40.9, "f1": 41.9, "tp": 56, "fp": 74, "fn": 81, "num_prs": 50}, "qodo-v2": {"precision": 40.5, "recall": 56.2, "f1": 47.1, "tp": 77, "fp": 113, "fn": 60, "num_prs": 50}, "claude-code": {"precision": 30.7, "recall": 40.1, "f1": 34.8, "tp": 55, "fp": 124, "fn": 82, "num_prs": 50}, "cubic-v2": {"precision": 55.6, "recall": 68.6, "f1": 61.4, "tp": 94, "fp": 75, "fn": 43, "num_prs": 50}, "augment": {"precision": 46.0, "recall": 63.5, "f1": 53.4, "tp": 87, "fp": 102, "fn": 50, "num_prs": 50}, "qodo-extended-summary": {"precision": 40.0, "recall": 67.2, "f1": 50.1, "tp": 92, "fp": 138, "fn": 45, "num_prs": 50}, "claude": {"precision": 35.7, "recall": 40.1, "f1": 37.8, "tp": 55, "fp": 99, "fn": 82, "num_prs": 50}, "mra-b": {"precision": 22.0, "recall": 40.9, "f1": 28.6, "tp": 56, "fp": 198, "fn": 81, "num_prs": 50}, "greptile-v4": {"precision": 31.2, "recall": 59.1, "f1": 40.8, "tp": 81, "fp": 179, "fn": 56, "num_prs": 50}, "kodus-v2": {"precision": 44.9, "recall": 35.0, "f1": 39.3, "tp": 48, "fp": 59, "fn": 89, "num_prs": 50}, "mra-max": {"precision": 9.5, "recall": 35.8, "f1": 15.1, "tp": 49, "fp": 465, "fn": 88, "num_prs": 50}, "gemini": {"precision": 29.7, "recall": 35.8, "f1": 32.5, "tp": 49, "fp": 116, "fn": 88, "num_prs": 50}, "codeant": {"precision": 13.4, "recall": 50.4, "f1": 21.2, "tp": 69, "fp": 445, "fn": 68, "num_prs": 50}, "bugbot": {"precision": 45.4, "recall": 43.1, "f1": 44.2, "tp": 59, "fp": 71, "fn": 78, "num_prs": 50}, "mra-a": {"precision": 18.9, "recall": 42.2, "f1": 26.1, "tp": 57, "fp": 245, "fn": 78, "num_prs": 49}, "sourcery": {"precision": 33.3, "recall": 51.8, "f1": 40.6, "tp": 71, "fp": 142, "fn": 66, "num_prs": 50}, "bito": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "vercel": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "copilot": {"precision": 27.2, "recall": 51.1, "f1": 35.5, "tp": 70, "fp": 187, "fn": 67, "num_prs": 50}, "graphite": {"precision": 100.0, "recall": 8.8, "f1": 16.1, "tp": 12, "fp": 0, "fn": 125, "num_prs": 50}, "mesa": {"precision": 17.6, "recall": 46.7, "f1": 25.5, "tp": 64, "fp": 300, "fn": 73, "num_prs": 50}, "kodus": {"precision": 21.9, "recall": 44.5, "f1": 29.3, "tp": 61, "fp": 218, "fn": 76, "num_prs": 50}, "kg": {"precision": 44.7, "recall": 15.3, "f1": 22.8, "tp": 21, "fp": 26, "fn": 116, "num_prs": 50}, "devin": {"precision": 54.2, "recall": 38.0, "f1": 44.6, "tp": 52, "fp": 44, "fn": 85, "num_prs": 50}, "baz": {"precision": 48.2, "recall": 29.9, "f1": 36.9, "tp": 41, "fp": 44, "fn": 96, "num_prs": 50}, "sentry": {"precision": 0.0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 1, "fn": 137, "num_prs": 50}, "macroscope": {"precision": 45.8, "recall": 43.8, "f1": 44.8, "tp": 60, "fp": 71, "fn": 77, "num_prs": 50}, "mra-nano": {"precision": 10.4, "recall": 25.5, "f1": 14.8, "tp": 35, "fp": 301, "fn": 102, "num_prs": 50}, "qodo-extended": {"precision": 35.3, "recall": 61.3, "f1": 44.8, "tp": 84, "fp": 154, "fn": 53, "num_prs": 50}, "cubic-dev": {"precision": 29.2, "recall": 76.6, "f1": 42.3, "tp": 105, "fp": 255, "fn": 32, "num_prs": 50}, "propel": {"precision": 49.5, "recall": 36.5, "f1": 42.0, "tp": 50, "fp": 51, "fn": 87, "num_prs": 50}, "qodo-v22": {"precision": 42.1, "recall": 52.6, "f1": 46.8, "tp": 72, "fp": 99, "fn": 65, "num_prs": 50}, "linearb": {"precision": 8.0, "recall": 2.9, "f1": 4.3, "tp": 4, "fp": 46, "fn": 133, "num_prs": 50}, "qodo": {"precision": 27.9, "recall": 45.3, "f1": 34.5, "tp": 62, "fp": 160, "fn": 75, "num_prs": 50}, "mra-ultra": {"precision": 17.1, "recall": 38.7, "f1": 23.7, "tp": 53, "fp": 257, "fn": 84, "num_prs": 50}, "entelligence": {"precision": 29.3, "recall": 21.2, "f1": 24.6, "tp": 29, "fp": 70, "fn": 108, "num_prs": 50}, "coderabbit": {"precision": 23.7, "recall": 38.7, "f1": 29.4, "tp": 53, "fp": 171, "fn": 84, "num_prs": 50}}}, "openai_gpt-5.2": {"prs": [{"url": "https://github.com/keycloak/keycloak/pull/37429", "language": "Java", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "gemini": {"tp": 0, "fp": 5, "fn": 4}, "claude": {"tp": 1, "fp": 0, "fn": 3}, "augment": {"tp": 2, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 4}, "devin": {"tp": 2, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 3}, "mra-a": {"tp": 0, "fp": 4, "fn": 4}, "mra-b": {"tp": 0, "fp": 2, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 4}, "sourcery": {"tp": 1, "fp": 14, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 1}, "mesa": {"tp": 2, "fp": 10, "fn": 2}, "codeant": {"tp": 2, "fp": 17, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 3}, "claude-code": {"tp": 1, "fp": 5, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 3, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 0, "fp": 2, "fn": 4}}}, {"url": "https://github.com/keycloak/keycloak/pull/37634", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 2, "fn": 2}, "gemini": {"tp": 3, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "augment": {"tp": 2, "fp": 4, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "graphite": {"tp": 2, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 2, "fp": 2, "fn": 2}, "copilot": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 5, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 6, "fn": 3}, "mra-b": {"tp": 2, "fp": 4, "fn": 2}, "mra-a": {"tp": 2, "fp": 5, "fn": 2}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 1, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 7, "fn": 2}, "codeant": {"tp": 2, "fp": 6, "fn": 2}, "kodus": {"tp": 2, "fp": 10, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "claude-code": {"tp": 0, "fp": 3, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 2, "fp": 2, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/38446", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 6, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 0, "fp": 3, "fn": 2}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 2}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 1, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 7, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 2}, "mra-a": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 2}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 0}, "macroscope": {"tp": 0, "fp": 4, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/36882", "language": "Java", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 4, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 0}, "augment": {"tp": 0, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 0, "fp": 2, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36880", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 6, "fn": 3}, "gemini": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 3, "fp": 2, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 6, "fn": 2}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 2}, "devin": {"tp": 0, "fp": 1, "fn": 3}, "mra-max": {"tp": 2, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 0, "fp": 5, "fn": 3}, "mra-a": {"tp": 3, "fp": 2, "fn": 0}, "entelligence": {"tp": 0, "fp": 5, "fn": 3}, "sourcery": {"tp": 1, "fp": 5, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 0, "fp": 11, "fn": 3}, "kodus": {"tp": 0, "fp": 6, "fn": 3}, "claude-code": {"tp": 0, "fp": 1, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v2-2": {"tp": 1, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/37038", "language": "Java", "pr_size": "small", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"qodo": {"tp": 1, "fp": 3, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-a": {"tp": 2, "fp": 6, "fn": 0}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 10, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 1, "fp": 11, "fn": 1}, "kodus": {"tp": 2, "fp": 8, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 0}, "macroscope": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/33832", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 4, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 1, "fp": 1, "fn": 1}, "qodo": {"tp": 0, "fp": 6, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 4, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 2, "fp": 9, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 9, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 1, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/40940", "language": "Java", "pr_size": "small", "domain": "concurrency", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 1, "fn": 2}, "gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 2, "fn": 0}, "mra-a": {"tp": 2, "fp": 7, "fn": 0}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 5, "fn": 0}, "codeant": {"tp": 2, "fp": 3, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"augment": {"tp": 2, "fp": 0, "fn": 0}, "gemini": {"tp": 2, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 8, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 4, "fn": 1}, "sourcery": {"tp": 2, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 2, "fp": 1, "fn": 0}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 1, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 0, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 0}, "macroscope": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/getsentry/sentry/pull/93824", "language": "Python", "pr_size": "large", "domain": "concurrency", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 0, "fn": 5}, "augment": {"tp": 5, "fp": 1, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 4}, "coderabbit": {"tp": 1, "fp": 3, "fn": 4}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "claude": {"tp": 0, "fp": 3, "fn": 5}, "qodo": {"tp": 1, "fp": 3, "fn": 4}, "copilot": {"tp": 1, "fp": 5, "fn": 4}, "propel": {"tp": 2, "fp": 4, "fn": 3}, "greptile": {"tp": 1, "fp": 1, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 4}, "mra-max": {"tp": 0, "fp": 5, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 5}, "mra-b": {"tp": 1, "fp": 10, "fn": 4}, "mra-a": {"tp": 1, "fp": 6, "fn": 4}, "entelligence": {"tp": 1, "fp": 0, "fn": 4}, "sourcery": {"tp": 4, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 5}, "codeant": {"tp": 1, "fp": 11, "fn": 4}, "kodus": {"tp": 3, "fp": 2, "fn": 2}, "claude-code": {"tp": 3, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2-2": {"tp": 0, "fp": 3, "fn": 5}, "qodo-v22": {"tp": 2, "fp": 3, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 4}, "cubic-v2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 3}, "macroscope": {"tp": 3, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5", "language": "Python", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 3}, "greptile": {"tp": 0, "fp": 2, "fn": 3}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "coderabbit": {"tp": 1, "fp": 16, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 16, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 9, "fn": 1}, "mra-a": {"tp": 1, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 0, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 2, "fp": 28, "fn": 1}, "mesa": {"tp": 0, "fp": 19, "fn": 3}, "codeant": {"tp": 1, "fp": 17, "fn": 2}, "kodus": {"tp": 1, "fp": 17, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 1, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 2}, "macroscope": {"tp": 0, "fp": 5, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1", "language": "Python", "pr_size": "medium", "domain": "API", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 3}, "claude": {"tp": 2, "fp": 7, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 1, "fp": 2, "fn": 3}, "copilot": {"tp": 3, "fp": 13, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 1, "fn": 1}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "coderabbit": {"tp": 3, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 3}, "kg": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 3, "fp": 2, "fn": 1}, "mra-max": {"tp": 1, "fp": 5, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 5, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 3, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 4}, "mra-a": {"tp": 3, "fp": 0, "fn": 1}, "entelligence": {"tp": 2, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 3, "fn": 1}, "mesa": {"tp": 2, "fp": 8, "fn": 2}, "codeant": {"tp": 2, "fp": 20, "fn": 2}, "kodus": {"tp": 2, "fp": 1, "fn": 2}, "claude-code": {"tp": 2, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 6, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/97529", "language": "Go", "pr_size": "small", "domain": "concurrency", "change_type": "performance", "complexity": "complex", "difficulty": "very_subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 3, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 8, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 4, "fn": 0}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 11, "fn": 1}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80168", "language": "Python", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 2, "fn": 0}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 5, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 2, "fp": 3, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/getsentry/sentry/pull/80528", "language": "Python", "pr_size": "small", "domain": "scheduling", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 2, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 1, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 1, "fp": 7, "fn": 1}, "entelligence": {"tp": 0, "fp": 4, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 1, "fn": 1}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/77754", "language": "Python", "pr_size": "medium", "domain": "serialization", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 3}, "claude": {"tp": 2, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 0, "fn": 3}, "baz": {"tp": 1, "fp": 0, "fn": 3}, "coderabbit": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 2, "fn": 2}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 3}, "mra-b": {"tp": 2, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 3}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 4, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 0, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 3}, "macroscope": {"tp": 1, "fp": 0, "fn": 3}}}, {"url": "https://github.com/getsentry/sentry/pull/95633", "language": "Python", "pr_size": "medium", "domain": "concurrency", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 0, "fp": 6, "fn": 3}, "propel": {"tp": 0, "fp": 1, "fn": 3}, "augment": {"tp": 0, "fp": 7, "fn": 3}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 0, "fp": 7, "fn": 3}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 4, "fn": 3}, "devin": {"tp": 0, "fp": 4, "fn": 3}, "mra-max": {"tp": 0, "fp": 5, "fn": 3}, "greptile-v4": {"tp": 0, "fp": 7, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 3}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 1, "fp": 7, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 0, "fp": 6, "fn": 3}, "codeant": {"tp": 1, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 0, "fp": 7, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2", "language": "Python", "pr_size": "medium", "domain": "data_processing", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 3}, "claude": {"tp": 2, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 8, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 3, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 3, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 0, "fn": 3}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 6, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 13, "fn": 1}, "mra-b": {"tp": 3, "fp": 8, "fn": 0}, "mra-a": {"tp": 3, "fp": 5, "fn": 0}, "entelligence": {"tp": 3, "fp": 1, "fn": 0}, "sourcery": {"tp": 2, "fp": 10, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 14, "fn": 3}, "codeant": {"tp": 2, "fp": 8, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 5, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 5, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3", "language": "Python", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 4, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 7, "fn": 3}, "coderabbit": {"tp": 2, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 13, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 1}, "mra-b": {"tp": 2, "fp": 4, "fn": 1}, "mra-a": {"tp": 1, "fp": 7, "fn": 2}, "entelligence": {"tp": 1, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 9, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 2}, "codeant": {"tp": 1, "fp": 12, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "linearb": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/103633", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 6, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 6, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 3, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 9, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 10, "fn": 2}, "mra-b": {"tp": 0, "fp": 10, "fn": 2}, "mra-a": {"tp": 0, "fp": 9, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 1}, "macroscope": {"tp": 0, "fp": 3, "fn": 2}}}, {"url": "https://github.com/getsentry/sentry/pull/67876", "language": "Python", "pr_size": "small", "domain": "authentication", "change_type": "security_patch", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 2}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 7, "fn": 2}, "bugbot": {"tp": 1, "fp": 2, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 7, "fn": 2}, "mra-a": {"tp": 0, "fp": 1, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 9, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 1, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 9, "fn": 2}, "kodus": {"tp": 0, "fp": 2, "fn": 3}, "claude-code": {"tp": 2, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/32918", "language": "Java", "pr_size": "small", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 2, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 0, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 4, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/94942", "language": "Go", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 9, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 9, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 2, "fp": 2, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 2, "fp": 1, "fn": 0}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 2, "fp": 0, "fn": 0}, "claude-code": {"tp": 1, "fp": 3, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/90939", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 0, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 1, "fn": 1}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 0, "fp": 3, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 3, "fn": 1}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/80329", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "performance", "complexity": "simple", "difficulty": "obvious", "risk": "low", "context": "local", "concern": "maintainability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 1, "fp": 4, "fn": 0}, "copilot": {"tp": 1, "fp": 4, "fn": 0}, "propel": {"tp": 0, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "coderabbit": {"tp": 1, "fp": 6, "fn": 0}, "greptile": {"tp": 1, "fp": 2, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 6, "fn": 0}, "devin": {"tp": 1, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 6, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 1}, "mra-a": {"tp": 0, "fp": 7, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 0}, "codeant": {"tp": 1, "fp": 9, "fn": 0}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 7, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90045", "language": "Go", "pr_size": "medium", "domain": "logging", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 5, "fn": 0}, "claude": {"tp": 3, "fp": 4, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 4, "fn": 0}, "bugbot": {"tp": 3, "fp": 2, "fn": 0}, "copilot": {"tp": 3, "fp": 8, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 3, "fp": 5, "fn": 0}, "baz": {"tp": 2, "fp": 2, "fn": 1}, "coderabbit": {"tp": 3, "fp": 3, "fn": 0}, "greptile": {"tp": 3, "fp": 1, "fn": 0}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 6, "fn": 1}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 5, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 0}, "mra-nano": {"tp": 0, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 3}, "mra-b": {"tp": 1, "fp": 2, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 3}, "entelligence": {"tp": 0, "fp": 4, "fn": 3}, "sourcery": {"tp": 3, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 3, "fp": 6, "fn": 0}, "codeant": {"tp": 3, "fp": 15, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 5, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 6, "fn": 1}, "qodo-extended": {"tp": 3, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/106778", "language": "Go", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 9, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 7, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 2, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 10, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 8, "fn": 2}, "mra-b": {"tp": 0, "fp": 7, "fn": 2}, "mra-a": {"tp": 0, "fp": 10, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 0, "fp": 14, "fn": 2}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/107534", "language": "Go", "pr_size": "small", "domain": "testing", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "low", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 2, "fn": 1}, "copilot": {"tp": 0, "fp": 4, "fn": 1}, "augment": {"tp": 0, "fp": 0, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 0}, "mra-nano": {"tp": 0, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 0}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 1}, "codeant": {"tp": 1, "fp": 5, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 7, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/79265", "language": "Go", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 4}, "claude": {"tp": 1, "fp": 2, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 2, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 5}, "augment": {"tp": 1, "fp": 1, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 4}, "baz": {"tp": 0, "fp": 1, "fn": 5}, "coderabbit": {"tp": 4, "fp": 5, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 4}, "kg": {"tp": 1, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 4}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 1, "fp": 11, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 3}, "mra-nano": {"tp": 2, "fp": 3, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 15, "fn": 3}, "mra-b": {"tp": 2, "fp": 5, "fn": 3}, "mra-a": {"tp": 2, "fp": 4, "fn": 3}, "entelligence": {"tp": 0, "fp": 0, "fn": 5}, "sourcery": {"tp": 3, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 6, "fn": 2}, "mesa": {"tp": 2, "fp": 9, "fn": 3}, "codeant": {"tp": 4, "fp": 9, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 4}, "claude-code": {"tp": 0, "fp": 4, "fn": 5}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 5}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 4}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 4}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 2, "fn": 5}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9", "language": "Ruby", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 0, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 3, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 4, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 17, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 9, "fn": 1}, "mra-b": {"tp": 1, "fp": 6, "fn": 1}, "mra-a": {"tp": 0, "fp": 4, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 6, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 0, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/76186", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 3, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "copilot": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 8, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 3, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 6, "fn": 2}, "sourcery": {"tp": 0, "fp": 6, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 1, "fp": 12, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10", "language": "Ruby", "pr_size": "large", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 10, "fn": 3}, "claude": {"tp": 0, "fp": 3, "fn": 4}, "qodo": {"tp": 2, "fp": 4, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "bugbot": {"tp": 3, "fp": 4, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 2, "fp": 15, "fn": 2}, "greptile": {"tp": 2, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 8, "fn": 2}, "devin": {"tp": 0, "fp": 2, "fn": 4}, "mra-max": {"tp": 0, "fp": 25, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 16, "fn": 4}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 2}, "mra-b": {"tp": 1, "fp": 6, "fn": 3}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 4, "fp": 11, "fn": 0}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 16, "fn": 2}, "kodus": {"tp": 3, "fp": 20, "fn": 1}, "claude-code": {"tp": 0, "fp": 6, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 10, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 7, "fn": 2}, "macroscope": {"tp": 2, "fp": 5, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "bug_fix", "complexity": "simple", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 2, "fn": 0}, "claude": {"tp": 3, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 1, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 1, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 3}, "greptile": {"tp": 2, "fp": 4, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 0}, "devin": {"tp": 3, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 20, "fn": 3}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 3}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 2, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 1, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 1}, "mesa": {"tp": 2, "fp": 5, "fn": 1}, "codeant": {"tp": 2, "fp": 6, "fn": 1}, "kodus": {"tp": 0, "fp": 5, "fn": 3}, "claude-code": {"tp": 3, "fp": 4, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8", "language": "Ruby", "pr_size": "medium", "domain": "API", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 2}, "claude": {"tp": 2, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 7, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 1}, "propel": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 2, "fp": 8, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 2}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 16, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 7, "fn": 2}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 0}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 2}, "entelligence": {"tp": 2, "fp": 4, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 0}, "mesa": {"tp": 2, "fp": 8, "fn": 1}, "codeant": {"tp": 0, "fp": 11, "fn": 3}, "kodus": {"tp": 3, "fp": 8, "fn": 0}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 5, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 2}, "macroscope": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3", "language": "Ruby", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 5, "fn": 1}, "copilot": {"tp": 0, "fp": 2, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 5, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 4, "fn": 2}, "coderabbit": {"tp": 1, "fp": 9, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 5, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 0, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 6, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 9, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 2, "fp": 15, "fn": 0}, "kodus": {"tp": 2, "fp": 5, "fn": 0}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5", "language": "Ruby", "pr_size": "small", "domain": "UI", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 6, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 13, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 1}, "mra-a": {"tp": 0, "fp": 10, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 5, "fn": 1}, "mesa": {"tp": 0, "fp": 10, "fn": 2}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6", "language": "Ruby", "pr_size": "small", "domain": "serialization", "change_type": "feature", "complexity": "simple", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 0, "fp": 3, "fn": 1}, "copilot": {"tp": 1, "fp": 2, "fn": 0}, "augment": {"tp": 0, "fp": 3, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 6, "fn": 1}, "greptile": {"tp": 0, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 3, "fn": 1}, "mra-max": {"tp": 0, "fp": 11, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 3, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 1}, "sourcery": {"tp": 0, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 3, "fn": 1}, "codeant": {"tp": 0, "fp": 15, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 0, "fp": 3, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 4, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 7, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 5, "fn": 1}, "macroscope": {"tp": 0, "fp": 2, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4", "language": "Ruby", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 7, "fn": 3}, "claude": {"tp": 2, "fp": 5, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 6}, "qodo": {"tp": 3, "fp": 3, "fn": 3}, "bugbot": {"tp": 1, "fp": 4, "fn": 5}, "copilot": {"tp": 1, "fp": 5, "fn": 5}, "augment": {"tp": 2, "fp": 3, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 5}, "baz": {"tp": 1, "fp": 4, "fn": 5}, "coderabbit": {"tp": 3, "fp": 25, "fn": 3}, "greptile": {"tp": 2, "fp": 5, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 6}, "qodo-v2": {"tp": 3, "fp": 14, "fn": 3}, "devin": {"tp": 1, "fp": 3, "fn": 5}, "mra-max": {"tp": 1, "fp": 17, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 4}, "mra-nano": {"tp": 1, "fp": 5, "fn": 5}, "mra-ultra": {"tp": 1, "fp": 7, "fn": 5}, "mra-b": {"tp": 2, "fp": 10, "fn": 4}, "mra-a": {"tp": 1, "fp": 6, "fn": 5}, "entelligence": {"tp": 1, "fp": 1, "fn": 5}, "sourcery": {"tp": 1, "fp": 4, "fn": 5}, "cubic-dev": {"tp": 3, "fp": 29, "fn": 3}, "mesa": {"tp": 1, "fp": 12, "fn": 5}, "codeant": {"tp": 3, "fp": 12, "fn": 3}, "kodus": {"tp": 2, "fp": 20, "fn": 4}, "claude-code": {"tp": 2, "fp": 4, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 6}, "bito": {"tp": 0, "fp": 0, "fn": 6}, "sentry": {"tp": 0, "fp": 0, "fn": 6}, "linearb": {"tp": 0, "fp": 1, "fn": 6}, "kodus-v2": {"tp": 2, "fp": 3, "fn": 4}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 4}, "qodo-extended": {"tp": 4, "fp": 7, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 4, "fn": 5}, "qodo-extended-summary": {"tp": 4, "fp": 5, "fn": 2}, "macroscope": {"tp": 1, "fp": 5, "fn": 5}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1", "language": "Ruby", "pr_size": "medium", "domain": "file_io", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 3, "fn": 0}, "bugbot": {"tp": 2, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 7, "fn": 1}, "augment": {"tp": 2, "fp": 4, "fn": 1}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "coderabbit": {"tp": 2, "fp": 5, "fn": 1}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 4, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 17, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 3, "fn": 0}, "mra-nano": {"tp": 2, "fp": 1, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 6, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 5, "fn": 2}, "entelligence": {"tp": 2, "fp": 2, "fn": 1}, "sourcery": {"tp": 2, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 11, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 3, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 2, "fp": 7, "fn": 0}, "greptile": {"tp": 2, "fp": 1, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 15, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 4, "fn": 0}, "mra-nano": {"tp": 2, "fp": 11, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 8, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 1, "fp": 8, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 8, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 2, "fp": 11, "fn": 0}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 0, "fp": 5, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22532", "language": "TypeScript", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 5, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 11, "fn": 0}, "augment": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 1, "fp": 2, "fn": 1}, "coderabbit": {"tp": 1, "fp": 10, "fn": 1}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 0, "fp": 6, "fn": 2}, "codeant": {"tp": 1, "fp": 9, "fn": 1}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/8330", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 0}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 2, "fp": 0, "fn": 0}, "qodo": {"tp": 2, "fp": 4, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 0}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 0}, "coderabbit": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 2, "fp": 0, "fn": 0}, "mra-max": {"tp": 2, "fp": 11, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 3, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 2, "fp": 3, "fn": 0}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 4, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 5, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14943", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 0, "fn": 1}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22345", "language": "TypeScript", "pr_size": "small", "domain": "database", "change_type": "migration", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 5, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 3, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 10, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 12, "fn": 1}, "mra-nano": {"tp": 0, "fp": 3, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 2}, "kodus": {"tp": 0, "fp": 2, "fn": 2}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 2, "fn": 2}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/11059", "language": "TypeScript", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 3, "fn": 2}, "claude": {"tp": 3, "fp": 3, "fn": 2}, "graphite": {"tp": 1, "fp": 1, "fn": 4}, "qodo": {"tp": 3, "fp": 9, "fn": 2}, "bugbot": {"tp": 4, "fp": 4, "fn": 1}, "copilot": {"tp": 5, "fp": 12, "fn": 0}, "augment": {"tp": 5, "fp": 8, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 4}, "baz": {"tp": 4, "fp": 1, "fn": 1}, "coderabbit": {"tp": 5, "fp": 14, "fn": 0}, "greptile": {"tp": 4, "fp": 7, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 5, "fp": 6, "fn": 0}, "devin": {"tp": 4, "fp": 1, "fn": 1}, "mra-max": {"tp": 3, "fp": 13, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 8, "fn": 1}, "mra-nano": {"tp": 2, "fp": 10, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 2}, "mra-b": {"tp": 4, "fp": 4, "fn": 1}, "mra-a": {"tp": 4, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 5}, "sourcery": {"tp": 5, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 5, "fp": 19, "fn": 0}, "mesa": {"tp": 2, "fp": 18, "fn": 3}, "codeant": {"tp": 4, "fp": 17, "fn": 1}, "kodus": {"tp": 3, "fp": 8, "fn": 2}, "claude-code": {"tp": 2, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 6, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 5, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 5, "fp": 8, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 5, "fp": 6, "fn": 0}, "macroscope": {"tp": 4, "fp": 6, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/7232", "language": "TypeScript", "pr_size": "medium", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 1, "fp": 3, "fn": 1}, "copilot": {"tp": 2, "fp": 6, "fn": 0}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 4, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 1, "fp": 4, "fn": 1}, "mra-max": {"tp": 2, "fp": 27, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 8, "fn": 1}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 10, "fn": 2}, "codeant": {"tp": 1, "fp": 15, "fn": 1}, "kodus": {"tp": 2, "fp": 6, "fn": 0}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 5, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/14740", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 4}, "claude": {"tp": 2, "fp": 0, "fn": 3}, "graphite": {"tp": 1, "fp": 1, "fn": 4}, "qodo": {"tp": 3, "fp": 2, "fn": 2}, "bugbot": {"tp": 3, "fp": 3, "fn": 2}, "copilot": {"tp": 3, "fp": 8, "fn": 2}, "augment": {"tp": 4, "fp": 3, "fn": 1}, "propel": {"tp": 4, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 4, "fp": 6, "fn": 1}, "greptile": {"tp": 4, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 4, "fp": 4, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 11, "fn": 3}, "greptile-v4": {"tp": 4, "fp": 7, "fn": 1}, "mra-nano": {"tp": 0, "fp": 7, "fn": 5}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 4}, "mra-b": {"tp": 1, "fp": 2, "fn": 4}, "mra-a": {"tp": 1, "fp": 5, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 5}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 11, "fn": 1}, "mesa": {"tp": 3, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 13, "fn": 3}, "kodus": {"tp": 2, "fp": 6, "fn": 3}, "claude-code": {"tp": 1, "fp": 3, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 4, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 4, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/10600", "language": "TypeScript", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 12, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 3}, "copilot": {"tp": 1, "fp": 9, "fn": 3}, "augment": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 0, "fp": 4, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 6, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 0, "fp": 14, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 9, "fn": 2}, "mra-nano": {"tp": 1, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 10, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 5, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 13, "fn": 1}, "mesa": {"tp": 1, "fp": 7, "fn": 3}, "codeant": {"tp": 2, "fp": 14, "fn": 2}, "kodus": {"tp": 0, "fp": 11, "fn": 4}, "claude-code": {"tp": 1, "fp": 5, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 6, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 3, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/10967", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "bug_fix", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 8, "fn": 4}, "claude": {"tp": 4, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 0, "fp": 3, "fn": 5}, "bugbot": {"tp": 2, "fp": 4, "fn": 3}, "copilot": {"tp": 3, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 1, "fp": 2, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 5}, "greptile": {"tp": 2, "fp": 2, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 4}, "devin": {"tp": 3, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 14, "fn": 5}, "greptile-v4": {"tp": 4, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 15, "fn": 5}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 2}, "mra-b": {"tp": 3, "fp": 15, "fn": 2}, "mra-a": {"tp": 2, "fp": 15, "fn": 3}, "entelligence": {"tp": 3, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 5}, "mesa": {"tp": 4, "fp": 11, "fn": 1}, "codeant": {"tp": 3, "fp": 7, "fn": 2}, "kodus": {"tp": 3, "fp": 6, "fn": 2}, "claude-code": {"tp": 3, "fp": 3, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 4, "fn": 3}, "qodo-v2-2": {"tp": 3, "fp": 5, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 6, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 4, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 4, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/8087", "language": "TypeScript", "pr_size": "medium", "domain": "concurrency", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 6, "fn": 1}, "mra-max": {"tp": 1, "fp": 9, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 3, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 7, "fn": 0}, "mra-b": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 1, "fp": 8, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 2, "fp": 7, "fn": 0}, "codeant": {"tp": 1, "fp": 8, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 5, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}}}], "tools": ["augment", "baz", "bito", "bugbot", "claude", "claude-code", "codeant", "coderabbit", "copilot", "cubic-dev", "cubic-v2", "devin", "entelligence", "gemini", "graphite", "greptile", "greptile-v4", "kg", "kodus", "kodus-v2", "linearb", "macroscope", "mesa", "mra-a", "mra-b", "mra-max", "mra-nano", "mra-ultra", "propel", "qodo", "qodo-extended", "qodo-extended-summary", "qodo-v2", "qodo-v2-2", "qodo-v22", "sentry", "sourcery", "vercel"], "dimensions": {"language": ["Go", "Java", "Python", "Ruby", "TypeScript"], "pr_size": ["small", "medium", "large"], "domain": ["API", "UI", "authentication", "caching", "concurrency", "configuration", "data_processing", "database", "file_io", "logging", "scheduling", "serialization", "testing"], "change_type": ["bug_fix", "feature", "migration", "performance", "refactoring", "security_patch"], "complexity": ["simple", "moderate", "complex"], "difficulty": ["obvious", "moderate", "subtle", "very_subtle"], "risk": ["low", "medium", "high", "critical"], "context": ["local", "file", "cross_file", "system"], "concern": ["correctness", "maintainability", "reliability", "security"]}, "overall_metrics": {"greptile": {"precision": 37.4, "recall": 38.0, "f1": 37.7, "tp": 52, "fp": 87, "fn": 85, "num_prs": 50}, "qodo-v2-2": {"precision": 41.5, "recall": 43.1, "f1": 42.3, "tp": 59, "fp": 83, "fn": 78, "num_prs": 50}, "qodo-v2": {"precision": 36.2, "recall": 56.2, "f1": 44.0, "tp": 77, "fp": 136, "fn": 60, "num_prs": 50}, "claude-code": {"precision": 27.4, "recall": 41.6, "f1": 33.0, "tp": 57, "fp": 151, "fn": 80, "num_prs": 50}, "cubic-v2": {"precision": 53.6, "recall": 65.7, "f1": 59.0, "tp": 90, "fp": 78, "fn": 47, "num_prs": 50}, "augment": {"precision": 41.9, "recall": 60.6, "f1": 49.6, "tp": 83, "fp": 115, "fn": 54, "num_prs": 50}, "qodo-extended-summary": {"precision": 38.1, "recall": 66.4, "f1": 48.4, "tp": 91, "fp": 148, "fn": 46, "num_prs": 50}, "claude": {"precision": 32.3, "recall": 37.2, "f1": 34.6, "tp": 51, "fp": 107, "fn": 86, "num_prs": 50}, "mra-b": {"precision": 19.9, "recall": 39.4, "f1": 26.5, "tp": 54, "fp": 217, "fn": 83, "num_prs": 50}, "greptile-v4": {"precision": 29.3, "recall": 58.4, "f1": 39.0, "tp": 80, "fp": 193, "fn": 57, "num_prs": 50}, "kodus-v2": {"precision": 35.9, "recall": 34.3, "f1": 35.1, "tp": 47, "fp": 84, "fn": 90, "num_prs": 50}, "mra-max": {"precision": 7.7, "recall": 31.4, "f1": 12.4, "tp": 43, "fp": 515, "fn": 94, "num_prs": 50}, "gemini": {"precision": 26.3, "recall": 33.6, "f1": 29.5, "tp": 46, "fp": 129, "fn": 91, "num_prs": 50}, "codeant": {"precision": 11.9, "recall": 49.6, "f1": 19.2, "tp": 68, "fp": 504, "fn": 69, "num_prs": 50}, "bugbot": {"precision": 43.2, "recall": 43.8, "f1": 43.5, "tp": 60, "fp": 79, "fn": 77, "num_prs": 50}, "mra-a": {"precision": 17.6, "recall": 40.0, "f1": 24.4, "tp": 54, "fp": 253, "fn": 81, "num_prs": 49}, "sourcery": {"precision": 29.8, "recall": 53.3, "f1": 38.2, "tp": 73, "fp": 172, "fn": 64, "num_prs": 50}, "bito": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "vercel": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "copilot": {"precision": 24.7, "recall": 52.6, "f1": 33.6, "tp": 72, "fp": 220, "fn": 65, "num_prs": 50}, "graphite": {"precision": 80.0, "recall": 8.8, "f1": 15.8, "tp": 12, "fp": 3, "fn": 125, "num_prs": 50}, "mesa": {"precision": 15.5, "recall": 42.3, "f1": 22.7, "tp": 58, "fp": 315, "fn": 79, "num_prs": 50}, "kodus": {"precision": 21.0, "recall": 43.1, "f1": 28.2, "tp": 59, "fp": 222, "fn": 78, "num_prs": 50}, "kg": {"precision": 51.1, "recall": 16.8, "f1": 25.3, "tp": 23, "fp": 22, "fn": 114, "num_prs": 50}, "devin": {"precision": 46.4, "recall": 37.2, "f1": 41.3, "tp": 51, "fp": 59, "fn": 86, "num_prs": 50}, "baz": {"precision": 37.8, "recall": 27.0, "f1": 31.5, "tp": 37, "fp": 61, "fn": 100, "num_prs": 50}, "sentry": {"precision": 0.0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 1, "fn": 137, "num_prs": 50}, "macroscope": {"precision": 41.8, "recall": 40.9, "f1": 41.3, "tp": 56, "fp": 78, "fn": 81, "num_prs": 50}, "mra-nano": {"precision": 10.8, "recall": 16.8, "f1": 13.1, "tp": 23, "fp": 190, "fn": 114, "num_prs": 50}, "qodo-extended": {"precision": 31.4, "recall": 59.9, "f1": 41.2, "tp": 82, "fp": 179, "fn": 55, "num_prs": 50}, "cubic-dev": {"precision": 27.3, "recall": 73.0, "f1": 39.8, "tp": 100, "fp": 266, "fn": 37, "num_prs": 50}, "propel": {"precision": 42.7, "recall": 36.5, "f1": 39.4, "tp": 50, "fp": 67, "fn": 87, "num_prs": 50}, "qodo-v22": {"precision": 38.4, "recall": 53.3, "f1": 44.6, "tp": 73, "fp": 117, "fn": 64, "num_prs": 50}, "linearb": {"precision": 8.0, "recall": 2.9, "f1": 4.3, "tp": 4, "fp": 46, "fn": 133, "num_prs": 50}, "qodo": {"precision": 25.6, "recall": 44.5, "f1": 32.5, "tp": 61, "fp": 177, "fn": 76, "num_prs": 50}, "mra-ultra": {"precision": 15.3, "recall": 37.2, "f1": 21.7, "tp": 51, "fp": 282, "fn": 86, "num_prs": 50}, "entelligence": {"precision": 27.7, "recall": 20.4, "f1": 23.5, "tp": 28, "fp": 73, "fn": 109, "num_prs": 50}, "coderabbit": {"precision": 21.3, "recall": 40.9, "f1": 28.0, "tp": 56, "fp": 207, "fn": 81, "num_prs": 50}}}};
-        const toolDisplayNames = {"graphite": "Graphite", "qodo": "Qodo", "gemini": "Gemini", "claude": "Claude Code", "augment": "Augment", "bugbot": "Cursor Bugbot", "coderabbit": "CodeRabbit", "propel": "Propel", "copilot": "GitHub Copilot", "baz": "Baz", "greptile": "Greptile", "kg": "KG", "entelligence": "Entelligence", "cubic-dev": "Cubic", "sourcery": "Sourcery", "mesa": "Mesa"};
-        const toolColors = {"graphite": "#6366f1", "qodo": "#8b5cf6", "gemini": "#06b6d4", "claude": "#f59e0b", "augment": "#10b981", "bugbot": "#3b82f6", "coderabbit": "#ec4899", "propel": "#14b8a6", "copilot": "#6b7280", "baz": "#f97316", "greptile": "#22c55e", "kg": "#a855f7", "entelligence": "#0ea5e9", "cubic-dev": "#d946ef", "sourcery": "#84cc16", "mesa": "#f43f5e"};
-        const predefinedFilters = [{"id": "high_precision", "label": "Highest Precision", "filters": {}, "sort": "precision", "description": "Tools ranked by precision - fewer false positives, more reliable findings", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "graphite", "best_score": 100.0}, {"id": "tool_kodus-v2_domain_concurrency", "label": "Best for Concurrency (Precision)", "filters": {"domain": ["concurrency"]}, "description": "Threading and async operations.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "kodus-v2", "best_score": 100.0, "sort": "precision"}, {"id": "tool_kg_complexity_complex", "label": "Best for Complex Code (Precision)", "filters": {"complexity": ["complex"]}, "description": "Deep logic and dependencies.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "kg", "best_score": 100.0, "sort": "precision"}, {"id": "tool_greptile-v4_change_type_bug_fix", "label": "Best for Bug Fixes (Recall)", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "greptile-v4", "best_score": 82.4, "sort": "recall"}, {"id": "tool_copilot_change_type_bug_fix", "label": "Best for Bug Fixes (Recall)", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "copilot", "best_score": 82.4, "sort": "recall"}, {"id": "tool_mesa_change_type_bug_fix", "label": "Best for Bug Fixes (Recall)", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "mesa", "best_score": 82.4, "sort": "recall"}, {"id": "change_type_performance", "label": "Best for Performance Optimization", "filters": {"change_type": ["performance"]}, "description": "Performance optimization changes", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 81.1}, {"id": "high_recall", "label": "Highest Recall", "filters": {}, "sort": "recall", "description": "Tools ranked by recall - catches more issues, may have more noise", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-dev", "best_score": 76.6}, {"id": "tool_qodo-v2-2_language_java_domain_authentication", "label": "Java + Authentication", "filters": {"language": ["Java"], "domain": ["authentication"]}, "description": "Java codebases with OOP patterns. Auth and access control.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v2-2", "best_score": 74.1}, {"id": "tool_codeant_difficulty_moderate_context_file", "label": "Moderate Bugs + File Context (Recall)", "filters": {"difficulty": ["moderate"], "context": ["file"]}, "description": "Requires careful reading. Requires full file understanding.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "codeant", "best_score": 73.3, "sort": "recall"}, {"id": "tool_mra-a_language_java_risk_high", "label": "Java + High Risk (Recall)", "filters": {"language": ["Java"], "risk": ["high"]}, "description": "Java codebases with OOP patterns. Significant impact, potential data loss.", "best_model": "openai_gpt-5.2", "best_tool": "mra-a", "best_score": 73.3, "sort": "recall"}, {"id": "ruby_medium", "label": "Best for Medium Ruby PRs", "filters": {"language": ["Ruby"], "pr_size": ["medium"]}, "description": "Ruby codebases with Rails patterns. Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "devin", "best_score": 72.7}, {"id": "tool_sourcery_language_typescript_concern_correctness", "label": "Typescript + Correctness", "filters": {"language": ["TypeScript"], "concern": ["correctness"]}, "description": "TypeScript codebases with frontend patterns. Logical correctness and expected behavior", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "sourcery", "best_score": 71.8}, {"id": "risk_critical", "label": "Best for Critical Risk", "filters": {"risk": ["critical"]}, "description": "Critical security or data corruption risk.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v22", "best_score": 69.8}, {"id": "tool_baz_language_java_domain_authentication", "label": "Java + Authentication", "filters": {"language": ["Java"], "domain": ["authentication"]}, "description": "Java codebases with OOP patterns. Auth and access control.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "baz", "best_score": 69.6}, {"id": "tool_qodo-extended_change_type_bug_fix_context_cross_file", "label": "Bug Fixes + Cross-File", "filters": {"change_type": ["bug_fix"], "context": ["cross_file"]}, "description": "Bug fixes and issue resolution Spans multiple files.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-extended", "best_score": 68.6}, {"id": "domain_ui", "label": "Best for Ui", "filters": {"domain": ["UI"]}, "description": "User interface and frontend.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "bugbot", "best_score": 66.7}, {"id": "tool_qodo-v2_language_java_risk_high", "label": "Java + High Risk", "filters": {"language": ["Java"], "risk": ["high"]}, "description": "Java codebases with OOP patterns. Significant impact, potential data loss.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v2", "best_score": 66.7}, {"id": "tool_macroscope_context_file_concern_correctness", "label": "File Context + Correctness", "filters": {"context": ["file"], "concern": ["correctness"]}, "description": "Requires full file understanding. Logical correctness and expected behavior", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "macroscope", "best_score": 66.7}, {"id": "tool_greptile_pr_size_medium_context_file", "label": "Medium PRs + File Context", "filters": {"pr_size": ["medium"], "context": ["file"]}, "description": "Medium PRs with 3-5 files, typical feature development Requires full file understanding.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "greptile", "best_score": 64.5}, {"id": "tool_propel_language_ruby_change_type_feature", "label": "Ruby + Features (Precision)", "filters": {"language": ["Ruby"], "change_type": ["feature"]}, "description": "Ruby codebases with Rails patterns. New feature implementation", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "propel", "best_score": 64.3, "sort": "precision"}, {"id": "complexity_complex", "label": "Best for Complex Code", "filters": {"complexity": ["complex"]}, "description": "Deep logic and dependencies.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "augment", "best_score": 62.9}, {"id": "tool_entelligence_change_type_performance_context_cross_file", "label": "Performance Optimization + Cross-File (Precision)", "filters": {"change_type": ["performance"], "context": ["cross_file"]}, "description": "Performance optimization changes Spans multiple files.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "entelligence", "best_score": 62.5, "sort": "precision"}, {"id": "java_medium", "label": "Best for Medium Java PRs", "filters": {"language": ["Java"], "pr_size": ["medium"]}, "description": "Java codebases with OOP patterns. Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-extended-summary", "best_score": 60.0}, {"id": "tool_claude_difficulty_moderate_context_file", "label": "Moderate Bugs + File Context", "filters": {"difficulty": ["moderate"], "context": ["file"]}, "description": "Requires careful reading. Requires full file understanding.", "best_model": "openai_gpt-5.2", "best_tool": "claude", "best_score": 57.1}, {"id": "tool_qodo_language_ruby_change_type_feature", "label": "Ruby + Features", "filters": {"language": ["Ruby"], "change_type": ["feature"]}, "description": "Ruby codebases with Rails patterns. New feature implementation", "best_model": "openai_gpt-5.2", "best_tool": "qodo", "best_score": 45.6}, {"id": "go_small", "label": "Best for Small Go PRs", "filters": {"language": ["Go"], "pr_size": ["small"]}, "description": "Go codebases with concurrency patterns. Small PRs with 1-2 files, easier to review thoroughly", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 77.4}, {"id": "language_go", "label": "Best for Go", "filters": {"language": ["Go"]}, "description": "Go codebases with concurrency patterns.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 75.5}, {"id": "pr_size_small", "label": "Best for Small PRs", "filters": {"pr_size": ["small"]}, "description": "Small PRs with 1-2 files, easier to review thoroughly", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 75.3}, {"id": "python_medium", "label": "Best for Medium Python PRs", "filters": {"language": ["Python"], "pr_size": ["medium"]}, "description": "Python codebases with dynamic typing. Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 71.0}, {"id": "language_python", "label": "Best for Python", "filters": {"language": ["Python"]}, "description": "Python codebases with dynamic typing.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 70.6}, {"id": "risk_high", "label": "Best for High Risk", "filters": {"risk": ["high"]}, "description": "Significant impact, potential data loss.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 70.2}, {"id": "change_type_bug_fix", "label": "Best for Bug Fixes", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 68.6}, {"id": "domain_scheduling", "label": "Best for Scheduling", "filters": {"domain": ["scheduling"]}, "description": "Task scheduling.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 68.4}, {"id": "context_file", "label": "Best for File Context", "filters": {"context": ["file"]}, "description": "Requires full file understanding.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 67.6}, {"id": "domain_concurrency", "label": "Best for Concurrency", "filters": {"domain": ["concurrency"]}, "description": "Threading and async operations.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-dev", "best_score": 64.5}, {"id": "difficulty_subtle", "label": "Best for Subtle Bugs", "filters": {"difficulty": ["subtle"]}, "description": "Non-obvious, needs domain knowledge.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 64.1}, {"id": "domain_caching", "label": "Best for Caching", "filters": {"domain": ["caching"]}, "description": "Cache and memoization.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 63.6}, {"id": "concern_reliability", "label": "Best for Reliability", "filters": {"concern": ["reliability"]}, "description": "Error handling and system stability", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v22", "best_score": 63.6}, {"id": "high_risk_auth", "label": "High Risk Auth", "filters": {"risk": ["high", "critical"], "domain": ["authentication"]}, "description": "Significant impact, potential data loss. Auth and access control.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 63.2}, {"id": "complex_subtle", "label": "Complex & Subtle", "filters": {"complexity": ["complex"], "difficulty": ["subtle", "very_subtle"]}, "description": "Deep logic and dependencies. Non-obvious, needs domain knowledge.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "augment", "best_score": 62.9}, {"id": "concern_correctness", "label": "Best for Correctness", "filters": {"concern": ["correctness"]}, "description": "Logical correctness and expected behavior", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 62.8}, {"id": "concern_security", "label": "Best for Security", "filters": {"concern": ["security"]}, "description": "Security vulnerabilities and attack vectors", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.9}, {"id": "security_critical", "label": "Security Critical", "filters": {"concern": ["security"], "risk": ["high", "critical"]}, "description": "Security vulnerabilities and attack vectors Significant impact, potential data loss.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.9}, {"id": "complexity_moderate", "label": "Best for Moderate Code", "filters": {"complexity": ["moderate"]}, "description": "Some abstraction.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.8}, {"id": "high_f1", "label": "Highest F1", "filters": {}, "sort": "f1", "description": "Tools ranked by F1 score - balanced precision and recall", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.8}, {"id": "language_typescript", "label": "Best for Typescript", "filters": {"language": ["TypeScript"]}, "description": "TypeScript codebases with frontend patterns.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "devin", "best_score": 61.8}, {"id": "language_java", "label": "Best for Java", "filters": {"language": ["Java"]}, "description": "Java codebases with OOP patterns.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v22", "best_score": 61.5}, {"id": "language_ruby", "label": "Best for Ruby", "filters": {"language": ["Ruby"]}, "description": "Ruby codebases with Rails patterns.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 61.0}, {"id": "domain_authentication", "label": "Best for Authentication", "filters": {"domain": ["authentication"]}, "description": "Auth and access control.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 60.3}, {"id": "context_cross_file", "label": "Best for Cross-File", "filters": {"context": ["cross_file"]}, "description": "Spans multiple files.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 60.0}, {"id": "change_type_feature", "label": "Best for Features", "filters": {"change_type": ["feature"]}, "description": "New feature implementation", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 59.9}, {"id": "risk_medium", "label": "Best for Medium Risk", "filters": {"risk": ["medium"]}, "description": "Moderate user impact.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 58.8}, {"id": "pr_size_large", "label": "Best for Large PRs", "filters": {"pr_size": ["large"]}, "description": "Large PRs with 6+ files, complex changes requiring careful review", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 58.3}, {"id": "pr_size_medium", "label": "Best for Medium PRs", "filters": {"pr_size": ["medium"]}, "description": "Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 56.7}, {"id": "difficulty_moderate", "label": "Best for Moderate Bugs", "filters": {"difficulty": ["moderate"]}, "description": "Requires careful reading.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 52.8}];
+        const allModelsData = {"anthropic_claude-opus-4-5-20251101": {"prs": [{"url": "https://github.com/keycloak/keycloak/pull/37429", "language": "Java", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "gemini": {"tp": 1, "fp": 4, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 3}, "augment": {"tp": 2, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "copilot": {"tp": 3, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 4}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 2, "fp": 5, "fn": 2}, "mra-nano": {"tp": 0, "fp": 2, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-a": {"tp": 0, "fp": 4, "fn": 4}, "mra-b": {"tp": 0, "fp": 2, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 4}, "sourcery": {"tp": 1, "fp": 6, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 14, "fn": 1}, "mesa": {"tp": 2, "fp": 8, "fn": 2}, "codeant": {"tp": 2, "fp": 14, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 3}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 4}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 4}, "tusk": {"tp": 2, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/37634", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 1, "fn": 2}, "gemini": {"tp": 3, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "augment": {"tp": 2, "fp": 2, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "graphite": {"tp": 2, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 2}, "bugbot": {"tp": 2, "fp": 1, "fn": 2}, "copilot": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 5, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 4, "fn": 2}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 0, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 9, "fn": 2}, "kodus": {"tp": 2, "fp": 10, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "claude-code": {"tp": 0, "fp": 2, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 2}, "macroscope": {"tp": 2, "fp": 2, "fn": 2}, "tusk": {"tp": 2, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/38446", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 2}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 2}, "mesa": {"tp": 1, "fp": 6, "fn": 1}, "codeant": {"tp": 1, "fp": 6, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 0}, "macroscope": {"tp": 0, "fp": 2, "fn": 2}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36882", "language": "Java", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 0}, "augment": {"tp": 0, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 7, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 0, "fp": 1, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36880", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 4, "fn": 3}, "gemini": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 2, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 6, "fn": 2}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 3}, "mra-max": {"tp": 2, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 2, "fn": 2}, "mra-a": {"tp": 3, "fp": 2, "fn": 0}, "entelligence": {"tp": 0, "fp": 5, "fn": 3}, "sourcery": {"tp": 0, "fp": 6, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 1, "fp": 0, "fn": 2}, "codeant": {"tp": 0, "fp": 10, "fn": 3}, "kodus": {"tp": 0, "fp": 5, "fn": 3}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/37038", "language": "Java", "pr_size": "small", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"qodo": {"tp": 1, "fp": 2, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 9, "fn": 1}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 2, "fp": 6, "fn": 0}, "mra-b": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 10, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 9, "fn": 1}, "kodus": {"tp": 2, "fp": 9, "fn": 0}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 0}, "macroscope": {"tp": 2, "fp": 2, "fn": 0}, "tusk": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/33832", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 1, "fp": 5, "fn": 1}, "gemini": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 5, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 9, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 9, "fn": 2}, "mra-a": {"tp": 0, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 11, "fn": 2}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 1, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 5, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/40940", "language": "Java", "pr_size": "small", "domain": "concurrency", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 1, "fn": 2}, "gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 2, "fp": 2, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"augment": {"tp": 2, "fp": 0, "fn": 0}, "gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 6, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 0, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 4, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 3, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 0, "fp": 6, "fn": 2}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 1, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 0, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/getsentry/sentry/pull/93824", "language": "Python", "pr_size": "large", "domain": "concurrency", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 0, "fn": 5}, "augment": {"tp": 5, "fp": 1, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 4}, "coderabbit": {"tp": 0, "fp": 1, "fn": 5}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "claude": {"tp": 0, "fp": 2, "fn": 5}, "qodo": {"tp": 1, "fp": 3, "fn": 4}, "copilot": {"tp": 1, "fp": 5, "fn": 4}, "propel": {"tp": 2, "fp": 2, "fn": 3}, "greptile": {"tp": 1, "fp": 1, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 3, "fp": 3, "fn": 2}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 0, "fp": 5, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 5}, "mra-b": {"tp": 1, "fp": 7, "fn": 4}, "mra-a": {"tp": 1, "fp": 5, "fn": 4}, "entelligence": {"tp": 1, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 5}, "codeant": {"tp": 1, "fp": 5, "fn": 4}, "kodus": {"tp": 3, "fp": 2, "fn": 2}, "claude-code": {"tp": 3, "fp": 3, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 5}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 4}, "cubic-v2": {"tp": 5, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 3}, "macroscope": {"tp": 3, "fp": 1, "fn": 2}, "tusk": {"tp": 2, "fp": 1, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5", "language": "Python", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 3}, "greptile": {"tp": 0, "fp": 2, "fn": 3}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "coderabbit": {"tp": 1, "fp": 15, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 13, "fn": 3}, "greptile-v4": {"tp": 0, "fp": 1, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 0, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 2, "fp": 28, "fn": 1}, "mesa": {"tp": 0, "fp": 14, "fn": 3}, "codeant": {"tp": 1, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 17, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 1, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 2}, "macroscope": {"tp": 0, "fp": 5, "fn": 3}, "tusk": {"tp": 1, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1", "language": "Python", "pr_size": "medium", "domain": "API", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 3}, "claude": {"tp": 2, "fp": 4, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 1, "fp": 2, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 1, "fn": 1}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "coderabbit": {"tp": 3, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 5, "fn": 3}, "kg": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 3, "fp": 1, "fn": 1}, "mra-max": {"tp": 1, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 7, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 3, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 4}, "mra-a": {"tp": 3, "fp": 2, "fn": 1}, "entelligence": {"tp": 2, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 7, "fn": 2}, "codeant": {"tp": 2, "fp": 15, "fn": 2}, "kodus": {"tp": 2, "fp": 1, "fn": 2}, "claude-code": {"tp": 2, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 4, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 2}, "tusk": {"tp": 3, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/97529", "language": "Go", "pr_size": "small", "domain": "concurrency", "change_type": "performance", "complexity": "complex", "difficulty": "very_subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 5, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 1, "fp": 1, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 8, "fn": 1}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80168", "language": "Python", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 6, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 2, "fp": 4, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80528", "language": "Python", "pr_size": "small", "domain": "scheduling", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 1, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 1, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 1, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 1, "fn": 2}, "codeant": {"tp": 1, "fp": 6, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/77754", "language": "Python", "pr_size": "medium", "domain": "serialization", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 3}, "claude": {"tp": 2, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 0, "fn": 3}, "baz": {"tp": 1, "fp": 0, "fn": 3}, "coderabbit": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 3, "fn": 2}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 3}, "mra-b": {"tp": 2, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 3}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 2, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 0, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 3}, "macroscope": {"tp": 1, "fp": 0, "fn": 3}, "tusk": {"tp": 3, "fp": 1, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/95633", "language": "Python", "pr_size": "medium", "domain": "concurrency", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 2, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 0, "fp": 6, "fn": 3}, "propel": {"tp": 0, "fp": 1, "fn": 3}, "augment": {"tp": 0, "fp": 3, "fn": 3}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 0, "fp": 7, "fn": 3}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 4, "fn": 3}, "devin": {"tp": 0, "fp": 2, "fn": 3}, "mra-max": {"tp": 0, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 2, "fn": 3}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 1, "fp": 1, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 0, "fp": 4, "fn": 3}, "codeant": {"tp": 1, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 0, "fp": 5, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 0, "fp": 7, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 3}, "tusk": {"tp": 1, "fp": 5, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2", "language": "Python", "pr_size": "medium", "domain": "data_processing", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 3}, "claude": {"tp": 2, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 3, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 2, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 0, "fn": 3}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 5, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 9, "fn": 1}, "mra-b": {"tp": 3, "fp": 9, "fn": 0}, "mra-a": {"tp": 3, "fp": 6, "fn": 0}, "entelligence": {"tp": 3, "fp": 1, "fn": 0}, "sourcery": {"tp": 2, "fp": 8, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 16, "fn": 1}, "codeant": {"tp": 2, "fp": 7, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 5, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 1}, "tusk": {"tp": 3, "fp": 1, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3", "language": "Python", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 1, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 2, "fp": 5, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 7, "fn": 1}, "mra-b": {"tp": 2, "fp": 4, "fn": 1}, "mra-a": {"tp": 2, "fp": 3, "fn": 1}, "entelligence": {"tp": 1, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 9, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 2}, "codeant": {"tp": 1, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "linearb": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/103633", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 6, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 5, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 9, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 2}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 0, "fp": 8, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 6, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/67876", "language": "Python", "pr_size": "small", "domain": "authentication", "change_type": "security_patch", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 2, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 1, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 8, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 1, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 8, "fn": 2}, "kodus": {"tp": 0, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}, "tusk": {"tp": 2, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/32918", "language": "Java", "pr_size": "small", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 3, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 0, "fp": 2, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 0, "fp": 4, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 0, "fp": 3, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/94942", "language": "Go", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 0, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 9, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 2, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 1}, "mesa": {"tp": 2, "fp": 1, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 2, "fp": 0, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 2, "fp": 0, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90939", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 0, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 0, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 1, "fn": 1}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 3, "fn": 1}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/80329", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "performance", "complexity": "simple", "difficulty": "obvious", "risk": "low", "context": "local", "concern": "maintainability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 0}, "copilot": {"tp": 1, "fp": 4, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "coderabbit": {"tp": 1, "fp": 2, "fn": 0}, "greptile": {"tp": 1, "fp": 3, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 0}, "devin": {"tp": 1, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 4, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 6, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 1}, "mra-a": {"tp": 0, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 9, "fn": 0}, "codeant": {"tp": 1, "fp": 8, "fn": 0}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 3, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 3, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90045", "language": "Go", "pr_size": "medium", "domain": "logging", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 4, "fn": 0}, "claude": {"tp": 3, "fp": 3, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 4, "fn": 0}, "bugbot": {"tp": 3, "fp": 3, "fn": 0}, "copilot": {"tp": 3, "fp": 7, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 3, "fp": 5, "fn": 0}, "baz": {"tp": 2, "fp": 2, "fn": 1}, "coderabbit": {"tp": 3, "fp": 2, "fn": 0}, "greptile": {"tp": 3, "fp": 1, "fn": 0}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 6, "fn": 1}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 3}, "mra-b": {"tp": 1, "fp": 1, "fn": 2}, "mra-a": {"tp": 0, "fp": 3, "fn": 3}, "entelligence": {"tp": 0, "fp": 4, "fn": 3}, "sourcery": {"tp": 3, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 3, "fp": 8, "fn": 0}, "codeant": {"tp": 3, "fp": 13, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 6, "fn": 1}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}, "tusk": {"tp": 3, "fp": 1, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/106778", "language": "Go", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 10, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "copilot": {"tp": 1, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 6, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 2, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 2, "fn": 2}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 8, "fn": 2}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 0, "fp": 11, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/107534", "language": "Go", "pr_size": "small", "domain": "testing", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "low", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 4, "fn": 1}, "augment": {"tp": 0, "fp": 0, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 0}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 3, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/79265", "language": "Go", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 4}, "claude": {"tp": 1, "fp": 2, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 2, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 7, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 5}, "augment": {"tp": 0, "fp": 2, "fn": 5}, "propel": {"tp": 1, "fp": 1, "fn": 4}, "baz": {"tp": 0, "fp": 1, "fn": 5}, "coderabbit": {"tp": 3, "fp": 4, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 4}, "kg": {"tp": 1, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 2, "fp": 10, "fn": 3}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 1, "fp": 3, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 15, "fn": 4}, "mra-b": {"tp": 2, "fp": 3, "fn": 3}, "mra-a": {"tp": 2, "fp": 4, "fn": 3}, "entelligence": {"tp": 0, "fp": 0, "fn": 5}, "sourcery": {"tp": 2, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 4, "fp": 5, "fn": 1}, "mesa": {"tp": 2, "fp": 7, "fn": 3}, "codeant": {"tp": 4, "fp": 5, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 4}, "claude-code": {"tp": 0, "fp": 2, "fn": 5}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 4}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 5}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 4}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 3}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 5}, "tusk": {"tp": 2, "fp": 2, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9", "language": "Ruby", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 0, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 3, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 10, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 6, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 4, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/76186", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 4, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "mra-a": {"tp": 2, "fp": 3, "fn": 0}, "entelligence": {"tp": 0, "fp": 5, "fn": 2}, "sourcery": {"tp": 0, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}, "tusk": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10", "language": "Ruby", "pr_size": "large", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 10, "fn": 2}, "claude": {"tp": 0, "fp": 3, "fn": 4}, "qodo": {"tp": 2, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "bugbot": {"tp": 3, "fp": 4, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 3, "fp": 0, "fn": 1}, "augment": {"tp": 3, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 2, "fp": 9, "fn": 2}, "greptile": {"tp": 2, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 2}, "devin": {"tp": 0, "fp": 2, "fn": 4}, "mra-max": {"tp": 0, "fp": 25, "fn": 4}, "greptile-v4": {"tp": 0, "fp": 3, "fn": 4}, "mra-nano": {"tp": 0, "fp": 17, "fn": 4}, "mra-ultra": {"tp": 2, "fp": 10, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 3}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 9, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 11, "fn": 2}, "kodus": {"tp": 3, "fp": 20, "fn": 1}, "claude-code": {"tp": 0, "fp": 6, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 3, "fp": 8, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 2}, "macroscope": {"tp": 3, "fp": 4, "fn": 1}, "tusk": {"tp": 2, "fp": 4, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "bug_fix", "complexity": "simple", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 2, "fn": 0}, "claude": {"tp": 3, "fp": 4, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 1, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 2, "fn": 3}, "greptile": {"tp": 1, "fp": 4, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 3, "fp": 2, "fn": 0}, "devin": {"tp": 3, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 20, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 3}, "mra-b": {"tp": 0, "fp": 6, "fn": 3}, "mra-a": {"tp": 3, "fp": 3, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 1, "fp": 1, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 1}, "mesa": {"tp": 1, "fp": 2, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}, "tusk": {"tp": 3, "fp": 2, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8", "language": "Ruby", "pr_size": "medium", "domain": "API", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 2}, "claude": {"tp": 2, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 2, "fp": 8, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 14, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 2}, "entelligence": {"tp": 2, "fp": 4, "fn": 1}, "sourcery": {"tp": 0, "fp": 5, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 2, "fp": 8, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 6, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 2}, "macroscope": {"tp": 1, "fp": 3, "fn": 2}, "tusk": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3", "language": "Ruby", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 7, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 0, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 6, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 6, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 9, "fn": 0}, "kodus": {"tp": 2, "fp": 5, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5", "language": "Ruby", "pr_size": "small", "domain": "UI", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 13, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 7, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 3, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 5, "fn": 1}, "mesa": {"tp": 1, "fp": 11, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 2, "fp": 4, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6", "language": "Ruby", "pr_size": "small", "domain": "serialization", "change_type": "feature", "complexity": "simple", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 0}, "copilot": {"tp": 1, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 0}, "baz": {"tp": 1, "fp": 1, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 0}, "greptile": {"tp": 0, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 2, "fn": 0}, "mra-max": {"tp": 1, "fp": 10, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 2, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 2, "fn": 0}, "mra-a": {"tp": 0, "fp": 4, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 0}, "sourcery": {"tp": 1, "fp": 4, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 0}, "mesa": {"tp": 0, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 13, "fn": 0}, "kodus": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 2, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4", "language": "Ruby", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 5, "fp": 4, "fn": 1}, "claude": {"tp": 3, "fp": 3, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 6}, "qodo": {"tp": 2, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 4, "fn": 5}, "copilot": {"tp": 1, "fp": 3, "fn": 5}, "augment": {"tp": 3, "fp": 2, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 5}, "baz": {"tp": 1, "fp": 3, "fn": 5}, "coderabbit": {"tp": 4, "fp": 20, "fn": 2}, "greptile": {"tp": 4, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 6}, "qodo-v2": {"tp": 3, "fp": 8, "fn": 3}, "devin": {"tp": 1, "fp": 3, "fn": 5}, "mra-max": {"tp": 1, "fp": 29, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 4}, "mra-nano": {"tp": 2, "fp": 3, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 7, "fn": 5}, "mra-b": {"tp": 4, "fp": 6, "fn": 2}, "mra-a": {"tp": 3, "fp": 4, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 5}, "sourcery": {"tp": 2, "fp": 3, "fn": 4}, "cubic-dev": {"tp": 4, "fp": 25, "fn": 2}, "mesa": {"tp": 4, "fp": 5, "fn": 2}, "codeant": {"tp": 3, "fp": 10, "fn": 3}, "kodus": {"tp": 3, "fp": 19, "fn": 3}, "claude-code": {"tp": 3, "fp": 1, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 6}, "bito": {"tp": 0, "fp": 0, "fn": 6}, "sentry": {"tp": 0, "fp": 0, "fn": 6}, "linearb": {"tp": 0, "fp": 1, "fn": 6}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 4}, "qodo-v2-2": {"tp": 3, "fp": 3, "fn": 3}, "qodo-v22": {"tp": 3, "fp": 3, "fn": 3}, "qodo-extended": {"tp": 5, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 5, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 5, "fn": 5}, "tusk": {"tp": 2, "fp": 3, "fn": 4}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1", "language": "Ruby", "pr_size": "medium", "domain": "file_io", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 1, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 2, "fp": 0, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "coderabbit": {"tp": 2, "fp": 6, "fn": 1}, "greptile": {"tp": 2, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 0}, "mra-nano": {"tp": 2, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 1, "fn": 2}, "mra-a": {"tp": 1, "fp": 1, "fn": 2}, "entelligence": {"tp": 2, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 1}, "codeant": {"tp": 2, "fp": 9, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 2}, "tusk": {"tp": 2, "fp": 2, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 2, "fp": 6, "fn": 0}, "greptile": {"tp": 2, "fp": 1, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 18, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 2, "fp": 8, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 2, "fp": 4, "fn": 0}, "mra-a": {"tp": 1, "fp": 6, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 8, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22532", "language": "TypeScript", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 11, "fn": 0}, "augment": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 2, "fp": 2, "fn": 0}, "baz": {"tp": 1, "fp": 2, "fn": 1}, "coderabbit": {"tp": 1, "fp": 10, "fn": 1}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 14, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 1, "fp": 3, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 0, "fp": 4, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/8330", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 0}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 2, "fp": 0, "fn": 0}, "qodo": {"tp": 2, "fp": 4, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 0}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 0}, "coderabbit": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 2, "fp": 0, "fn": 0}, "mra-max": {"tp": 2, "fp": 12, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 2, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 2, "fp": 3, "fn": 0}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14943", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 0, "fn": 1}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 4, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/22345", "language": "TypeScript", "pr_size": "small", "domain": "database", "change_type": "migration", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 2, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 0, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 6, "fn": 2}, "kodus": {"tp": 0, "fp": 1, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 2, "fn": 2}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/11059", "language": "TypeScript", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 3, "fn": 2}, "claude": {"tp": 3, "fp": 3, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 3, "fp": 7, "fn": 2}, "bugbot": {"tp": 4, "fp": 4, "fn": 1}, "copilot": {"tp": 4, "fp": 9, "fn": 1}, "augment": {"tp": 5, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 4, "fp": 1, "fn": 1}, "coderabbit": {"tp": 5, "fp": 14, "fn": 0}, "greptile": {"tp": 4, "fp": 7, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 4, "fn": 1}, "devin": {"tp": 4, "fp": 1, "fn": 1}, "mra-max": {"tp": 3, "fp": 12, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 7, "fn": 1}, "mra-nano": {"tp": 2, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 2}, "mra-b": {"tp": 4, "fp": 4, "fn": 1}, "mra-a": {"tp": 4, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 5}, "sourcery": {"tp": 5, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 5, "fp": 18, "fn": 0}, "mesa": {"tp": 2, "fp": 13, "fn": 3}, "codeant": {"tp": 4, "fp": 13, "fn": 1}, "kodus": {"tp": 3, "fp": 5, "fn": 2}, "claude-code": {"tp": 3, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 5, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 5, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 5, "fp": 6, "fn": 0}, "macroscope": {"tp": 4, "fp": 7, "fn": 1}, "tusk": {"tp": 4, "fp": 2, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/7232", "language": "TypeScript", "pr_size": "medium", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 3, "fn": 2}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 4, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 18, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 4, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 5, "fn": 0}, "mesa": {"tp": 0, "fp": 9, "fn": 2}, "codeant": {"tp": 2, "fp": 9, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}, "tusk": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14740", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 4}, "claude": {"tp": 2, "fp": 0, "fn": 3}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 3, "fn": 2}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "augment": {"tp": 4, "fp": 3, "fn": 1}, "propel": {"tp": 4, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 4, "fp": 6, "fn": 1}, "greptile": {"tp": 4, "fp": 1, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 4, "fp": 3, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 11, "fn": 3}, "greptile-v4": {"tp": 4, "fp": 6, "fn": 1}, "mra-nano": {"tp": 0, "fp": 6, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 5}, "mra-b": {"tp": 2, "fp": 0, "fn": 3}, "mra-a": {"tp": 1, "fp": 4, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 5}, "sourcery": {"tp": 3, "fp": 1, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 11, "fn": 1}, "mesa": {"tp": 3, "fp": 3, "fn": 2}, "codeant": {"tp": 2, "fp": 12, "fn": 3}, "kodus": {"tp": 1, "fp": 7, "fn": 4}, "claude-code": {"tp": 1, "fp": 2, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 4, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 4, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 3}, "tusk": {"tp": 3, "fp": 2, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/10600", "language": "TypeScript", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 8, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 3}, "copilot": {"tp": 1, "fp": 9, "fn": 3}, "augment": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 0, "fp": 4, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 0, "fp": 13, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 7, "fn": 2}, "mra-nano": {"tp": 1, "fp": 6, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 6, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 7, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 11, "fn": 1}, "mesa": {"tp": 1, "fp": 5, "fn": 3}, "codeant": {"tp": 2, "fp": 12, "fn": 2}, "kodus": {"tp": 0, "fp": 8, "fn": 4}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 2, "fn": 4}, "tusk": {"tp": 0, "fp": 4, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/10967", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "bug_fix", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 9, "fn": 4}, "claude": {"tp": 3, "fp": 4, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 0, "fp": 3, "fn": 5}, "bugbot": {"tp": 2, "fp": 4, "fn": 3}, "copilot": {"tp": 4, "fp": 7, "fn": 1}, "augment": {"tp": 2, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 5}, "greptile": {"tp": 2, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 4}, "devin": {"tp": 2, "fp": 3, "fn": 3}, "mra-max": {"tp": 2, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 5, "fn": 2}, "mra-nano": {"tp": 0, "fp": 15, "fn": 5}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 2}, "mra-b": {"tp": 3, "fp": 6, "fn": 2}, "mra-a": {"tp": 3, "fp": 15, "fn": 2}, "entelligence": {"tp": 3, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 3, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 5}, "mesa": {"tp": 3, "fp": 9, "fn": 2}, "codeant": {"tp": 3, "fp": 6, "fn": 2}, "kodus": {"tp": 3, "fp": 6, "fn": 2}, "claude-code": {"tp": 2, "fp": 5, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 5, "fn": 3}, "qodo-v22": {"tp": 3, "fp": 3, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 4, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 3, "fn": 3}, "tusk": {"tp": 2, "fp": 3, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/8087", "language": "TypeScript", "pr_size": "medium", "domain": "concurrency", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 4, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 3, "fn": 1}, "mra-max": {"tp": 1, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 4, "fn": 0}, "mra-b": {"tp": 2, "fp": 2, "fn": 0}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 2, "fp": 8, "fn": 0}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 10, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}], "tools": ["augment", "baz", "bito", "bugbot", "claude", "claude-code", "codeant", "coderabbit", "copilot", "cubic-dev", "cubic-v2", "devin", "entelligence", "gemini", "graphite", "greptile", "greptile-v4", "kg", "kodus", "kodus-v2", "linearb", "macroscope", "mesa", "mra-a", "mra-b", "mra-max", "mra-nano", "mra-ultra", "propel", "qodo", "qodo-extended", "qodo-extended-summary", "qodo-v2", "qodo-v2-2", "qodo-v22", "sentry", "sourcery", "tusk", "vercel"], "dimensions": {"language": ["Go", "Java", "Python", "Ruby", "TypeScript"], "pr_size": ["small", "medium", "large"], "domain": ["API", "UI", "authentication", "caching", "concurrency", "configuration", "data_processing", "database", "file_io", "logging", "scheduling", "serialization", "testing"], "change_type": ["bug_fix", "feature", "migration", "performance", "refactoring", "security_patch"], "complexity": ["simple", "moderate", "complex"], "difficulty": ["obvious", "moderate", "subtle", "very_subtle"], "risk": ["low", "medium", "high", "critical"], "context": ["local", "file", "cross_file", "system"], "concern": ["correctness", "maintainability", "reliability", "security"]}, "overall_metrics": {"qodo": {"precision": 31.8, "recall": 44.5, "f1": 37.1, "tp": 61, "fp": 131, "fn": 76, "num_prs": 50}, "greptile": {"precision": 41.5, "recall": 39.4, "f1": 40.4, "tp": 54, "fp": 76, "fn": 83, "num_prs": 50}, "sentry": {"precision": 0.0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 1, "fn": 137, "num_prs": 50}, "kodus": {"precision": 22.9, "recall": 44.5, "f1": 30.3, "tp": 61, "fp": 205, "fn": 76, "num_prs": 50}, "copilot": {"precision": 28.3, "recall": 53.3, "f1": 37.0, "tp": 73, "fp": 185, "fn": 64, "num_prs": 50}, "macroscope": {"precision": 48.4, "recall": 43.8, "f1": 46.0, "tp": 60, "fp": 64, "fn": 77, "num_prs": 50}, "graphite": {"precision": 100.0, "recall": 8.8, "f1": 16.1, "tp": 12, "fp": 0, "fn": 125, "num_prs": 50}, "entelligence": {"precision": 30.1, "recall": 20.4, "f1": 24.3, "tp": 28, "fp": 65, "fn": 109, "num_prs": 50}, "sourcery": {"precision": 33.3, "recall": 51.8, "f1": 40.6, "tp": 71, "fp": 142, "fn": 66, "num_prs": 50}, "tusk": {"precision": 40.2, "recall": 56.9, "f1": 47.1, "tp": 78, "fp": 116, "fn": 59, "num_prs": 50}, "claude": {"precision": 34.8, "recall": 35.8, "f1": 35.3, "tp": 49, "fp": 92, "fn": 88, "num_prs": 50}, "baz": {"precision": 48.8, "recall": 29.2, "f1": 36.5, "tp": 40, "fp": 42, "fn": 97, "num_prs": 50}, "qodo-v22": {"precision": 44.6, "recall": 54.7, "f1": 49.2, "tp": 75, "fp": 93, "fn": 62, "num_prs": 50}, "codeant": {"precision": 13.9, "recall": 48.9, "f1": 21.6, "tp": 67, "fp": 415, "fn": 70, "num_prs": 50}, "linearb": {"precision": 8.3, "recall": 2.9, "f1": 4.3, "tp": 4, "fp": 44, "fn": 133, "num_prs": 50}, "vercel": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "coderabbit": {"precision": 24.7, "recall": 39.4, "f1": 30.3, "tp": 54, "fp": 165, "fn": 83, "num_prs": 50}, "mesa": {"precision": 19.3, "recall": 45.3, "f1": 27.1, "tp": 62, "fp": 259, "fn": 75, "num_prs": 50}, "gemini": {"precision": 31.1, "recall": 37.2, "f1": 33.9, "tp": 51, "fp": 113, "fn": 86, "num_prs": 50}, "qodo-extended-summary": {"precision": 40.2, "recall": 67.2, "f1": 50.3, "tp": 92, "fp": 137, "fn": 45, "num_prs": 50}, "cubic-dev": {"precision": 30.1, "recall": 75.9, "f1": 43.2, "tp": 104, "fp": 241, "fn": 33, "num_prs": 50}, "mra-ultra": {"precision": 18.3, "recall": 37.2, "f1": 24.6, "tp": 51, "fp": 227, "fn": 86, "num_prs": 50}, "qodo-extended": {"precision": 37.2, "recall": 62.8, "f1": 46.7, "tp": 86, "fp": 145, "fn": 51, "num_prs": 50}, "augment": {"precision": 47.5, "recall": 61.3, "f1": 53.5, "tp": 84, "fp": 93, "fn": 53, "num_prs": 50}, "mra-max": {"precision": 8.9, "recall": 34.3, "f1": 14.1, "tp": 47, "fp": 481, "fn": 90, "num_prs": 50}, "kodus-v2": {"precision": 46.7, "recall": 35.8, "f1": 40.5, "tp": 49, "fp": 56, "fn": 88, "num_prs": 50}, "kg": {"precision": 50.0, "recall": 16.8, "f1": 25.1, "tp": 23, "fp": 23, "fn": 114, "num_prs": 50}, "cubic-v2": {"precision": 56.3, "recall": 68.6, "f1": 61.8, "tp": 94, "fp": 73, "fn": 43, "num_prs": 50}, "devin": {"precision": 54.3, "recall": 37.2, "f1": 44.2, "tp": 51, "fp": 43, "fn": 86, "num_prs": 50}, "mra-b": {"precision": 25.0, "recall": 41.6, "f1": 31.2, "tp": 57, "fp": 171, "fn": 80, "num_prs": 50}, "propel": {"precision": 52.5, "recall": 38.7, "f1": 44.5, "tp": 53, "fp": 48, "fn": 84, "num_prs": 50}, "mra-a": {"precision": 23.6, "recall": 45.2, "f1": 31.0, "tp": 61, "fp": 198, "fn": 74, "num_prs": 49}, "claude-code": {"precision": 34.8, "recall": 40.9, "f1": 37.6, "tp": 56, "fp": 105, "fn": 81, "num_prs": 50}, "qodo-v2-2": {"precision": 47.1, "recall": 41.6, "f1": 44.2, "tp": 57, "fp": 64, "fn": 80, "num_prs": 50}, "greptile-v4": {"precision": 33.1, "recall": 56.9, "f1": 41.8, "tp": 78, "fp": 158, "fn": 59, "num_prs": 50}, "bito": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "mra-nano": {"precision": 12.6, "recall": 17.5, "f1": 14.7, "tp": 24, "fp": 166, "fn": 113, "num_prs": 50}, "bugbot": {"precision": 47.2, "recall": 43.8, "f1": 45.5, "tp": 60, "fp": 67, "fn": 77, "num_prs": 50}, "qodo-v2": {"precision": 42.9, "recall": 55.5, "f1": 48.4, "tp": 76, "fp": 101, "fn": 61, "num_prs": 50}}}, "anthropic_claude-sonnet-4-5-20250929": {"prs": [{"url": "https://github.com/keycloak/keycloak/pull/37429", "language": "Java", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "gemini": {"tp": 0, "fp": 4, "fn": 4}, "claude": {"tp": 1, "fp": 0, "fn": 3}, "augment": {"tp": 2, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 4}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 8, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-a": {"tp": 0, "fp": 4, "fn": 4}, "mra-b": {"tp": 0, "fp": 2, "fn": 4}, "entelligence": {"tp": 0, "fp": 2, "fn": 4}, "sourcery": {"tp": 1, "fp": 5, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 14, "fn": 1}, "mesa": {"tp": 1, "fp": 7, "fn": 3}, "codeant": {"tp": 2, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 3}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 3, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 4}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 4}, "tusk": {"tp": 2, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/37634", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 3, "fn": 2}, "gemini": {"tp": 3, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "augment": {"tp": 2, "fp": 1, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "graphite": {"tp": 2, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 2}, "bugbot": {"tp": 2, "fp": 2, "fn": 2}, "copilot": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 4, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 3}, "mra-b": {"tp": 2, "fp": 4, "fn": 2}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 0, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 9, "fn": 2}, "codeant": {"tp": 2, "fp": 8, "fn": 2}, "kodus": {"tp": 2, "fp": 10, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "claude-code": {"tp": 0, "fp": 2, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 2}, "macroscope": {"tp": 2, "fp": 2, "fn": 2}, "tusk": {"tp": 2, "fp": 1, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/38446", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 2}, "propel": {"tp": 0, "fp": 4, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 1, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 2}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 1, "fp": 6, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 0}, "macroscope": {"tp": 0, "fp": 3, "fn": 2}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36882", "language": "Java", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "augment": {"tp": 0, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 10, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 10, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 0, "fp": 1, "fn": 1}, "codeant": {"tp": 0, "fp": 13, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36880", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 5, "fn": 3}, "gemini": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 2, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 4, "fn": 3}, "copilot": {"tp": 0, "fp": 7, "fn": 3}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 2}, "devin": {"tp": 0, "fp": 1, "fn": 3}, "mra-max": {"tp": 1, "fp": 7, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 5, "fn": 3}, "sourcery": {"tp": 1, "fp": 5, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 0, "fp": 7, "fn": 3}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 1}, "tusk": {"tp": 1, "fp": 5, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/37038", "language": "Java", "pr_size": "small", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"qodo": {"tp": 1, "fp": 2, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 8, "fn": 1}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 1, "fp": 7, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 10, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 2, "fp": 9, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 0}, "macroscope": {"tp": 2, "fp": 2, "fn": 0}, "tusk": {"tp": 2, "fp": 1, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/33832", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 1, "fp": 5, "fn": 1}, "gemini": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 4, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 8, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 18, "fn": 1}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 0, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 1, "fp": 7, "fn": 1}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 1, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 5, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/40940", "language": "Java", "pr_size": "small", "domain": "concurrency", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 1, "fn": 2}, "gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 2, "fp": 4, "fn": 0}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 13, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 3, "fn": 1}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"augment": {"tp": 2, "fp": 0, "fn": 0}, "gemini": {"tp": 2, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 0}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 4, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 0, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 4, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 2, "fp": 4, "fn": 0}, "sourcery": {"tp": 2, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 0, "fp": 5, "fn": 2}, "kodus": {"tp": 2, "fp": 2, "fn": 0}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 1, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 0, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/getsentry/sentry/pull/93824", "language": "Python", "pr_size": "large", "domain": "concurrency", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 0, "fn": 5}, "augment": {"tp": 5, "fp": 1, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 4}, "coderabbit": {"tp": 1, "fp": 3, "fn": 4}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "claude": {"tp": 0, "fp": 2, "fn": 5}, "qodo": {"tp": 1, "fp": 2, "fn": 4}, "copilot": {"tp": 1, "fp": 4, "fn": 4}, "propel": {"tp": 1, "fp": 2, "fn": 4}, "greptile": {"tp": 1, "fp": 1, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 2, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 0, "fp": 7, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 5}, "mra-b": {"tp": 1, "fp": 8, "fn": 4}, "mra-a": {"tp": 1, "fp": 9, "fn": 4}, "entelligence": {"tp": 1, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 5}, "codeant": {"tp": 2, "fp": 5, "fn": 3}, "kodus": {"tp": 3, "fp": 2, "fn": 2}, "claude-code": {"tp": 3, "fp": 3, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 5}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 4}, "cubic-v2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 3}, "macroscope": {"tp": 2, "fp": 1, "fn": 3}, "tusk": {"tp": 2, "fp": 1, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5", "language": "Python", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 3}, "greptile": {"tp": 0, "fp": 2, "fn": 3}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "coderabbit": {"tp": 1, "fp": 32, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 9, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 9, "fn": 1}, "mra-a": {"tp": 1, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 0, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 2, "fp": 28, "fn": 1}, "mesa": {"tp": 0, "fp": 18, "fn": 3}, "codeant": {"tp": 1, "fp": 15, "fn": 2}, "kodus": {"tp": 1, "fp": 17, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 1, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 2}, "macroscope": {"tp": 0, "fp": 5, "fn": 3}, "tusk": {"tp": 1, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1", "language": "Python", "pr_size": "medium", "domain": "API", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 3}, "claude": {"tp": 2, "fp": 7, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 1, "fp": 2, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 1, "fn": 1}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "coderabbit": {"tp": 3, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 3}, "kg": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 3, "fp": 2, "fn": 1}, "mra-max": {"tp": 1, "fp": 4, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 6, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 3, "fn": 4}, "mra-a": {"tp": 3, "fp": 3, "fn": 1}, "entelligence": {"tp": 2, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 1}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 3, "fp": 14, "fn": 1}, "kodus": {"tp": 2, "fp": 3, "fn": 2}, "claude-code": {"tp": 2, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 2}, "tusk": {"tp": 3, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/97529", "language": "Go", "pr_size": "small", "domain": "concurrency", "change_type": "performance", "complexity": "complex", "difficulty": "very_subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 4, "fn": 0}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 6, "fn": 1}, "codeant": {"tp": 1, "fp": 9, "fn": 1}, "kodus": {"tp": 1, "fp": 0, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80168", "language": "Python", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 8, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 2, "fp": 4, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 0, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80528", "language": "Python", "pr_size": "small", "domain": "scheduling", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 1, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 1, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 1, "fn": 1}, "codeant": {"tp": 1, "fp": 5, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/77754", "language": "Python", "pr_size": "medium", "domain": "serialization", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 3}, "claude": {"tp": 2, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 0, "fn": 3}, "baz": {"tp": 1, "fp": 0, "fn": 3}, "coderabbit": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 2, "fn": 2}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 3}, "mra-b": {"tp": 2, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 3, "fn": 3}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 2, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 0, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 3}, "macroscope": {"tp": 1, "fp": 0, "fn": 3}, "tusk": {"tp": 3, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/95633", "language": "Python", "pr_size": "medium", "domain": "concurrency", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 2, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 0, "fp": 6, "fn": 3}, "propel": {"tp": 0, "fp": 1, "fn": 3}, "augment": {"tp": 0, "fp": 6, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 3}, "coderabbit": {"tp": 0, "fp": 7, "fn": 3}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 4, "fn": 3}, "devin": {"tp": 0, "fp": 2, "fn": 3}, "mra-max": {"tp": 1, "fp": 4, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 2, "fn": 3}, "mra-b": {"tp": 1, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 5, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 0, "fp": 4, "fn": 3}, "codeant": {"tp": 1, "fp": 12, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 0, "fp": 5, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 0, "fp": 7, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 3}, "tusk": {"tp": 1, "fp": 8, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2", "language": "Python", "pr_size": "medium", "domain": "data_processing", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 3}, "claude": {"tp": 2, "fp": 6, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 3, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 2, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 0, "fn": 3}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 5, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 9, "fn": 1}, "mra-b": {"tp": 3, "fp": 7, "fn": 0}, "mra-a": {"tp": 3, "fp": 5, "fn": 0}, "entelligence": {"tp": 3, "fp": 1, "fn": 0}, "sourcery": {"tp": 2, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 24, "fn": 1}, "codeant": {"tp": 2, "fp": 7, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 4, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 5, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 1}, "tusk": {"tp": 3, "fp": 1, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3", "language": "Python", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 2, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 0}, "mra-b": {"tp": 2, "fp": 4, "fn": 1}, "mra-a": {"tp": 2, "fp": 6, "fn": 1}, "entelligence": {"tp": 1, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 9, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 2}, "codeant": {"tp": 1, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "linearb": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/103633", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 1, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 5, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 3, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 5, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 8, "fn": 1}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 0, "fp": 7, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 0, "fp": 6, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/67876", "language": "Python", "pr_size": "small", "domain": "authentication", "change_type": "security_patch", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 2, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 7, "fn": 1}, "bugbot": {"tp": 1, "fp": 2, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 2}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 1, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 8, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 8, "fn": 2}, "kodus": {"tp": 0, "fp": 3, "fn": 3}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}, "tusk": {"tp": 3, "fp": 2, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/32918", "language": "Java", "pr_size": "small", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 4, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 0, "fp": 3, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 0, "fp": 3, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/94942", "language": "Go", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 0, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 0, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 11, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 2, "fp": 7, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 1}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 10, "fn": 0}, "kodus": {"tp": 2, "fp": 0, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/90939", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 0, "fn": 1}, "mra-nano": {"tp": 1, "fp": 2, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 1, "fn": 1}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 0, "fp": 3, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 3, "fn": 1}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/80329", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "performance", "complexity": "simple", "difficulty": "obvious", "risk": "low", "context": "local", "concern": "maintainability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 0}, "copilot": {"tp": 1, "fp": 4, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 3, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "coderabbit": {"tp": 1, "fp": 2, "fn": 0}, "greptile": {"tp": 1, "fp": 3, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 0}, "devin": {"tp": 1, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 5, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 6, "fn": 1}, "mra-a": {"tp": 0, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 9, "fn": 0}, "codeant": {"tp": 1, "fp": 8, "fn": 0}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 6, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 3, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90045", "language": "Go", "pr_size": "medium", "domain": "logging", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 4, "fn": 0}, "claude": {"tp": 3, "fp": 3, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 5, "fn": 0}, "bugbot": {"tp": 3, "fp": 3, "fn": 0}, "copilot": {"tp": 3, "fp": 8, "fn": 0}, "propel": {"tp": 2, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 5, "fn": 0}, "baz": {"tp": 2, "fp": 2, "fn": 1}, "coderabbit": {"tp": 3, "fp": 2, "fn": 0}, "greptile": {"tp": 3, "fp": 1, "fn": 0}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 6, "fn": 1}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 3, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 9, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 3}, "mra-b": {"tp": 1, "fp": 2, "fn": 2}, "mra-a": {"tp": 0, "fp": 3, "fn": 3}, "entelligence": {"tp": 0, "fp": 4, "fn": 3}, "sourcery": {"tp": 3, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 3, "fp": 1, "fn": 0}, "mesa": {"tp": 3, "fp": 9, "fn": 0}, "codeant": {"tp": 3, "fp": 11, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 3, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 6, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}, "tusk": {"tp": 3, "fp": 2, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/106778", "language": "Go", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 12, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 4, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "copilot": {"tp": 1, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 5, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 6, "fn": 1}, "devin": {"tp": 2, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 14, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 0, "fp": 6, "fn": 2}, "mra-a": {"tp": 0, "fp": 12, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/107534", "language": "Go", "pr_size": "small", "domain": "testing", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "low", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 1}, "augment": {"tp": 0, "fp": 0, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 5, "fn": 0}, "mra-nano": {"tp": 0, "fp": 7, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 0}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 3, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/79265", "language": "Go", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 4}, "claude": {"tp": 1, "fp": 2, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 2, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 5}, "augment": {"tp": 1, "fp": 1, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 4}, "baz": {"tp": 0, "fp": 1, "fn": 5}, "coderabbit": {"tp": 3, "fp": 4, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 4}, "kg": {"tp": 1, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 3}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 1, "fp": 11, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 2, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 15, "fn": 3}, "mra-b": {"tp": 2, "fp": 3, "fn": 3}, "mra-a": {"tp": 2, "fp": 4, "fn": 3}, "entelligence": {"tp": 0, "fp": 0, "fn": 5}, "sourcery": {"tp": 2, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 4, "fp": 5, "fn": 1}, "mesa": {"tp": 3, "fp": 6, "fn": 2}, "codeant": {"tp": 4, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 4}, "claude-code": {"tp": 0, "fp": 3, "fn": 5}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 4}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 5}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 4}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 3}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 1, "fn": 5}, "tusk": {"tp": 2, "fp": 2, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9", "language": "Ruby", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 0, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 3, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 10, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 0, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 0, "fp": 12, "fn": 2}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/76186", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 0, "fp": 2, "fn": 2}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 9, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 2, "fp": 4, "fn": 0}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 6, "fn": 2}, "sourcery": {"tp": 0, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 8, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 1, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}, "tusk": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10", "language": "Ruby", "pr_size": "large", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 8, "fn": 2}, "claude": {"tp": 0, "fp": 3, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "bugbot": {"tp": 3, "fp": 4, "fn": 1}, "copilot": {"tp": 2, "fp": 4, "fn": 2}, "propel": {"tp": 3, "fp": 0, "fn": 1}, "augment": {"tp": 3, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 2, "fp": 10, "fn": 2}, "greptile": {"tp": 2, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 2}, "devin": {"tp": 0, "fp": 2, "fn": 4}, "mra-max": {"tp": 0, "fp": 18, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 5, "fn": 2}, "mra-nano": {"tp": 0, "fp": 17, "fn": 4}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 3}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 3, "fp": 8, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 13, "fn": 2}, "kodus": {"tp": 3, "fp": 19, "fn": 1}, "claude-code": {"tp": 0, "fp": 6, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 7, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 2}, "macroscope": {"tp": 3, "fp": 5, "fn": 1}, "tusk": {"tp": 2, "fp": 4, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "bug_fix", "complexity": "simple", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 2, "fn": 0}, "claude": {"tp": 3, "fp": 4, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 1, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 2}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 2, "fn": 3}, "greptile": {"tp": 2, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 3, "fp": 2, "fn": 0}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 1, "fp": 18, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 3}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 2, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 3, "fp": 2, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 1}, "kodus": {"tp": 2, "fp": 3, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 2, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8", "language": "Ruby", "pr_size": "medium", "domain": "API", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 1}, "claude": {"tp": 2, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 1, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 2, "fp": 8, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 15, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 1}, "mra-nano": {"tp": 2, "fp": 8, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 0, "fp": 5, "fn": 3}, "entelligence": {"tp": 2, "fp": 4, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 3}, "kodus": {"tp": 2, "fp": 8, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 6, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 2}, "macroscope": {"tp": 1, "fp": 3, "fn": 2}, "tusk": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3", "language": "Ruby", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 6, "fn": 1}, "copilot": {"tp": 0, "fp": 2, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 3, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 0, "fn": 2}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 2}, "mra-nano": {"tp": 1, "fp": 7, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 7, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5", "language": "Ruby", "pr_size": "small", "domain": "UI", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 12, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 8, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "mra-a": {"tp": 1, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 11, "fn": 1}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6", "language": "Ruby", "pr_size": "small", "domain": "serialization", "change_type": "feature", "complexity": "simple", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 1, "fp": 4, "fn": 0}, "copilot": {"tp": 1, "fp": 2, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 0}, "baz": {"tp": 1, "fp": 1, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 0}, "greptile": {"tp": 0, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 0}, "mra-max": {"tp": 1, "fp": 11, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 10, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 2, "fn": 0}, "mra-a": {"tp": 0, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 1}, "sourcery": {"tp": 0, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 0, "fn": 0}, "mesa": {"tp": 0, "fp": 3, "fn": 1}, "codeant": {"tp": 1, "fp": 13, "fn": 0}, "kodus": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 5, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 2, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4", "language": "Ruby", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 4, "fp": 5, "fn": 2}, "claude": {"tp": 5, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 6}, "qodo": {"tp": 3, "fp": 4, "fn": 3}, "bugbot": {"tp": 1, "fp": 4, "fn": 5}, "copilot": {"tp": 1, "fp": 3, "fn": 5}, "augment": {"tp": 3, "fp": 2, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 5}, "baz": {"tp": 1, "fp": 3, "fn": 5}, "coderabbit": {"tp": 3, "fp": 22, "fn": 3}, "greptile": {"tp": 4, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 6}, "qodo-v2": {"tp": 4, "fp": 4, "fn": 2}, "devin": {"tp": 1, "fp": 3, "fn": 5}, "mra-max": {"tp": 1, "fp": 19, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 4}, "mra-nano": {"tp": 2, "fp": 10, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 7, "fn": 5}, "mra-b": {"tp": 4, "fp": 6, "fn": 2}, "mra-a": {"tp": 3, "fp": 4, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 5}, "sourcery": {"tp": 2, "fp": 3, "fn": 4}, "cubic-dev": {"tp": 4, "fp": 25, "fn": 2}, "mesa": {"tp": 2, "fp": 9, "fn": 4}, "codeant": {"tp": 3, "fp": 17, "fn": 3}, "kodus": {"tp": 3, "fp": 18, "fn": 3}, "claude-code": {"tp": 3, "fp": 1, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 6}, "bito": {"tp": 0, "fp": 0, "fn": 6}, "sentry": {"tp": 0, "fp": 0, "fn": 6}, "linearb": {"tp": 0, "fp": 1, "fn": 6}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 4}, "qodo-v2-2": {"tp": 3, "fp": 3, "fn": 3}, "qodo-v22": {"tp": 3, "fp": 3, "fn": 3}, "qodo-extended": {"tp": 6, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 4}, "qodo-extended-summary": {"tp": 5, "fp": 2, "fn": 1}, "macroscope": {"tp": 1, "fp": 5, "fn": 5}, "tusk": {"tp": 2, "fp": 3, "fn": 4}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1", "language": "Ruby", "pr_size": "medium", "domain": "file_io", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 3, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 2, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 4, "fn": 1}, "augment": {"tp": 2, "fp": 3, "fn": 1}, "propel": {"tp": 2, "fp": 0, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 5, "fn": 2}, "greptile": {"tp": 2, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 1}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 1, "fp": 5, "fn": 2}, "entelligence": {"tp": 2, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 13, "fn": 1}, "kodus": {"tp": 2, "fp": 3, "fn": 1}, "claude-code": {"tp": 3, "fp": 1, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 4, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}, "tusk": {"tp": 2, "fp": 2, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 5, "fn": 1}, "greptile": {"tp": 2, "fp": 1, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 10, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 2, "fp": 8, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 2, "fp": 4, "fn": 0}, "mra-a": {"tp": 1, "fp": 11, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 8, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 2, "fp": 9, "fn": 0}, "kodus": {"tp": 1, "fp": 4, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/22532", "language": "TypeScript", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 4, "fn": 2}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 11, "fn": 0}, "augment": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 1, "fp": 2, "fn": 1}, "coderabbit": {"tp": 1, "fp": 4, "fn": 1}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 15, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 8, "fn": 1}, "mra-nano": {"tp": 2, "fp": 11, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 1, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 10, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 0, "fp": 5, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/8330", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 5, "fn": 0}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 2, "fp": 0, "fn": 0}, "qodo": {"tp": 2, "fp": 4, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 0}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "coderabbit": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 3, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 2, "fp": 0, "fn": 0}, "mra-max": {"tp": 2, "fp": 14, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 0}, "mra-nano": {"tp": 2, "fp": 9, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 2}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 2, "fp": 3, "fn": 0}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14943", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 3, "fn": 1}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 1, "fp": 8, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 0, "fn": 1}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/22345", "language": "TypeScript", "pr_size": "small", "domain": "database", "change_type": "migration", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 2, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 10, "fn": 1}, "mra-nano": {"tp": 0, "fp": 9, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 0, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 0, "fp": 5, "fn": 2}, "kodus": {"tp": 0, "fp": 1, "fn": 2}, "claude-code": {"tp": 1, "fp": 3, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 2, "fn": 2}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/11059", "language": "TypeScript", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 3, "fn": 2}, "claude": {"tp": 3, "fp": 3, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 3, "fp": 6, "fn": 2}, "bugbot": {"tp": 4, "fp": 3, "fn": 1}, "copilot": {"tp": 4, "fp": 10, "fn": 1}, "augment": {"tp": 5, "fp": 5, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 4, "fp": 1, "fn": 1}, "coderabbit": {"tp": 5, "fp": 12, "fn": 0}, "greptile": {"tp": 4, "fp": 7, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 7, "fn": 1}, "devin": {"tp": 4, "fp": 1, "fn": 1}, "mra-max": {"tp": 3, "fp": 12, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 13, "fn": 1}, "mra-nano": {"tp": 4, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 2}, "mra-b": {"tp": 4, "fp": 4, "fn": 1}, "mra-a": {"tp": 4, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 5}, "sourcery": {"tp": 5, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 5, "fp": 21, "fn": 0}, "mesa": {"tp": 2, "fp": 13, "fn": 3}, "codeant": {"tp": 4, "fp": 14, "fn": 1}, "kodus": {"tp": 3, "fp": 7, "fn": 2}, "claude-code": {"tp": 2, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 5, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 5, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 5, "fp": 6, "fn": 0}, "macroscope": {"tp": 4, "fp": 6, "fn": 1}, "tusk": {"tp": 4, "fp": 2, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/7232", "language": "TypeScript", "pr_size": "medium", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 2}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 5, "fn": 0}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 21, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 1, "fp": 18, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 9, "fn": 2}, "codeant": {"tp": 1, "fp": 11, "fn": 1}, "kodus": {"tp": 2, "fp": 11, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}, "tusk": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14740", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 4}, "claude": {"tp": 2, "fp": 0, "fn": 3}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 3}, "bugbot": {"tp": 3, "fp": 3, "fn": 2}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "augment": {"tp": 4, "fp": 3, "fn": 1}, "propel": {"tp": 4, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 4, "fp": 3, "fn": 1}, "greptile": {"tp": 4, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 4}, "qodo-v2": {"tp": 3, "fp": 3, "fn": 2}, "devin": {"tp": 2, "fp": 0, "fn": 3}, "mra-max": {"tp": 3, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 5, "fn": 1}, "mra-nano": {"tp": 1, "fp": 9, "fn": 4}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 5}, "mra-b": {"tp": 1, "fp": 2, "fn": 4}, "mra-a": {"tp": 0, "fp": 6, "fn": 5}, "entelligence": {"tp": 0, "fp": 3, "fn": 5}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 11, "fn": 1}, "mesa": {"tp": 2, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 15, "fn": 3}, "kodus": {"tp": 1, "fp": 7, "fn": 4}, "claude-code": {"tp": 1, "fp": 2, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 4, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 4, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 3}, "tusk": {"tp": 3, "fp": 2, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/10600", "language": "TypeScript", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 8, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 8, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 3}, "copilot": {"tp": 1, "fp": 9, "fn": 3}, "augment": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 0, "fp": 4, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 0, "fp": 14, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 9, "fn": 2}, "mra-nano": {"tp": 1, "fp": 11, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 8, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 8, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 4, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 11, "fn": 1}, "mesa": {"tp": 1, "fp": 5, "fn": 3}, "codeant": {"tp": 2, "fp": 12, "fn": 2}, "kodus": {"tp": 0, "fp": 8, "fn": 4}, "claude-code": {"tp": 1, "fp": 4, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 0, "fp": 1, "fn": 4}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 4}, "tusk": {"tp": 0, "fp": 4, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/10967", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "bug_fix", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 7, "fn": 4}, "claude": {"tp": 4, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 0, "fp": 3, "fn": 5}, "bugbot": {"tp": 2, "fp": 4, "fn": 3}, "copilot": {"tp": 3, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 5}, "greptile": {"tp": 2, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 4}, "devin": {"tp": 3, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 17, "fn": 3}, "greptile-v4": {"tp": 4, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 4}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 2}, "mra-b": {"tp": 2, "fp": 11, "fn": 3}, "mra-a": {"tp": 3, "fp": 15, "fn": 2}, "entelligence": {"tp": 3, "fp": 5, "fn": 2}, "sourcery": {"tp": 2, "fp": 5, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 5}, "mesa": {"tp": 4, "fp": 15, "fn": 1}, "codeant": {"tp": 3, "fp": 5, "fn": 2}, "kodus": {"tp": 3, "fp": 5, "fn": 2}, "claude-code": {"tp": 2, "fp": 6, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 5, "fn": 3}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 2, "fp": 6, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 4, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 3, "fn": 3}, "tusk": {"tp": 2, "fp": 3, "fn": 3}}}, {"url": "https://github.com/calcom/cal.com/pull/8087", "language": "TypeScript", "pr_size": "medium", "domain": "concurrency", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 6, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 2, "fp": 3, "fn": 0}, "mra-max": {"tp": 1, "fp": 11, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 12, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 1, "fp": 8, "fn": 1}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 2, "fp": 9, "fn": 0}, "codeant": {"tp": 1, "fp": 11, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 9, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 1, "fp": 4, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}], "tools": ["augment", "baz", "bito", "bugbot", "claude", "claude-code", "codeant", "coderabbit", "copilot", "cubic-dev", "cubic-v2", "devin", "entelligence", "gemini", "graphite", "greptile", "greptile-v4", "kg", "kodus", "kodus-v2", "linearb", "macroscope", "mesa", "mra-a", "mra-b", "mra-max", "mra-nano", "mra-ultra", "propel", "qodo", "qodo-extended", "qodo-extended-summary", "qodo-v2", "qodo-v2-2", "qodo-v22", "sentry", "sourcery", "tusk", "vercel"], "dimensions": {"language": ["Go", "Java", "Python", "Ruby", "TypeScript"], "pr_size": ["small", "medium", "large"], "domain": ["API", "UI", "authentication", "caching", "concurrency", "configuration", "data_processing", "database", "file_io", "logging", "scheduling", "serialization", "testing"], "change_type": ["bug_fix", "feature", "migration", "performance", "refactoring", "security_patch"], "complexity": ["simple", "moderate", "complex"], "difficulty": ["obvious", "moderate", "subtle", "very_subtle"], "risk": ["low", "medium", "high", "critical"], "context": ["local", "file", "cross_file", "system"], "concern": ["correctness", "maintainability", "reliability", "security"]}, "overall_metrics": {"qodo": {"precision": 27.9, "recall": 45.3, "f1": 34.5, "tp": 62, "fp": 160, "fn": 75, "num_prs": 50}, "greptile": {"precision": 43.7, "recall": 40.1, "f1": 41.8, "tp": 55, "fp": 71, "fn": 82, "num_prs": 50}, "sentry": {"precision": 0.0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 1, "fn": 137, "num_prs": 50}, "kodus": {"precision": 21.9, "recall": 44.5, "f1": 29.3, "tp": 61, "fp": 218, "fn": 76, "num_prs": 50}, "copilot": {"precision": 27.2, "recall": 51.1, "f1": 35.5, "tp": 70, "fp": 187, "fn": 67, "num_prs": 50}, "macroscope": {"precision": 45.8, "recall": 43.8, "f1": 44.8, "tp": 60, "fp": 71, "fn": 77, "num_prs": 50}, "graphite": {"precision": 100.0, "recall": 8.8, "f1": 16.1, "tp": 12, "fp": 0, "fn": 125, "num_prs": 50}, "entelligence": {"precision": 29.3, "recall": 21.2, "f1": 24.6, "tp": 29, "fp": 70, "fn": 108, "num_prs": 50}, "sourcery": {"precision": 33.3, "recall": 51.8, "f1": 40.6, "tp": 71, "fp": 142, "fn": 66, "num_prs": 50}, "tusk": {"precision": 37.6, "recall": 56.2, "f1": 45.0, "tp": 77, "fp": 128, "fn": 60, "num_prs": 50}, "claude": {"precision": 35.7, "recall": 40.1, "f1": 37.8, "tp": 55, "fp": 99, "fn": 82, "num_prs": 50}, "baz": {"precision": 48.2, "recall": 29.9, "f1": 36.9, "tp": 41, "fp": 44, "fn": 96, "num_prs": 50}, "qodo-v22": {"precision": 42.1, "recall": 52.6, "f1": 46.8, "tp": 72, "fp": 99, "fn": 65, "num_prs": 50}, "codeant": {"precision": 13.4, "recall": 50.4, "f1": 21.2, "tp": 69, "fp": 445, "fn": 68, "num_prs": 50}, "linearb": {"precision": 8.0, "recall": 2.9, "f1": 4.3, "tp": 4, "fp": 46, "fn": 133, "num_prs": 50}, "vercel": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "coderabbit": {"precision": 23.7, "recall": 38.7, "f1": 29.4, "tp": 53, "fp": 171, "fn": 84, "num_prs": 50}, "mesa": {"precision": 17.6, "recall": 46.7, "f1": 25.5, "tp": 64, "fp": 300, "fn": 73, "num_prs": 50}, "gemini": {"precision": 29.7, "recall": 35.8, "f1": 32.5, "tp": 49, "fp": 116, "fn": 88, "num_prs": 50}, "qodo-extended-summary": {"precision": 40.0, "recall": 67.2, "f1": 50.1, "tp": 92, "fp": 138, "fn": 45, "num_prs": 50}, "cubic-dev": {"precision": 29.2, "recall": 76.6, "f1": 42.3, "tp": 105, "fp": 255, "fn": 32, "num_prs": 50}, "mra-ultra": {"precision": 17.1, "recall": 38.7, "f1": 23.7, "tp": 53, "fp": 257, "fn": 84, "num_prs": 50}, "qodo-extended": {"precision": 35.3, "recall": 61.3, "f1": 44.8, "tp": 84, "fp": 154, "fn": 53, "num_prs": 50}, "augment": {"precision": 46.0, "recall": 63.5, "f1": 53.4, "tp": 87, "fp": 102, "fn": 50, "num_prs": 50}, "mra-max": {"precision": 9.5, "recall": 35.8, "f1": 15.1, "tp": 49, "fp": 465, "fn": 88, "num_prs": 50}, "kodus-v2": {"precision": 44.9, "recall": 35.0, "f1": 39.3, "tp": 48, "fp": 59, "fn": 89, "num_prs": 50}, "kg": {"precision": 44.7, "recall": 15.3, "f1": 22.8, "tp": 21, "fp": 26, "fn": 116, "num_prs": 50}, "cubic-v2": {"precision": 55.6, "recall": 68.6, "f1": 61.4, "tp": 94, "fp": 75, "fn": 43, "num_prs": 50}, "devin": {"precision": 54.2, "recall": 38.0, "f1": 44.6, "tp": 52, "fp": 44, "fn": 85, "num_prs": 50}, "mra-b": {"precision": 22.0, "recall": 40.9, "f1": 28.6, "tp": 56, "fp": 198, "fn": 81, "num_prs": 50}, "propel": {"precision": 49.5, "recall": 36.5, "f1": 42.0, "tp": 50, "fp": 51, "fn": 87, "num_prs": 50}, "mra-a": {"precision": 18.9, "recall": 42.2, "f1": 26.1, "tp": 57, "fp": 245, "fn": 78, "num_prs": 49}, "claude-code": {"precision": 30.7, "recall": 40.1, "f1": 34.8, "tp": 55, "fp": 124, "fn": 82, "num_prs": 50}, "qodo-v2-2": {"precision": 43.1, "recall": 40.9, "f1": 41.9, "tp": 56, "fp": 74, "fn": 81, "num_prs": 50}, "greptile-v4": {"precision": 31.2, "recall": 59.1, "f1": 40.8, "tp": 81, "fp": 179, "fn": 56, "num_prs": 50}, "bito": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "mra-nano": {"precision": 10.4, "recall": 25.5, "f1": 14.8, "tp": 35, "fp": 301, "fn": 102, "num_prs": 50}, "bugbot": {"precision": 45.4, "recall": 43.1, "f1": 44.2, "tp": 59, "fp": 71, "fn": 78, "num_prs": 50}, "qodo-v2": {"precision": 40.5, "recall": 56.2, "f1": 47.1, "tp": 77, "fp": 113, "fn": 60, "num_prs": 50}}}, "openai_gpt-5.2": {"prs": [{"url": "https://github.com/keycloak/keycloak/pull/37429", "language": "Java", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "gemini": {"tp": 0, "fp": 5, "fn": 4}, "claude": {"tp": 1, "fp": 0, "fn": 3}, "augment": {"tp": 2, "fp": 3, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 3, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 4}, "devin": {"tp": 2, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 7, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 3}, "mra-a": {"tp": 0, "fp": 4, "fn": 4}, "mra-b": {"tp": 0, "fp": 2, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 4}, "sourcery": {"tp": 1, "fp": 14, "fn": 3}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 1}, "mesa": {"tp": 2, "fp": 10, "fn": 2}, "codeant": {"tp": 2, "fp": 17, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 3}, "claude-code": {"tp": 1, "fp": 5, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 3, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 3}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 0, "fp": 2, "fn": 4}, "tusk": {"tp": 2, "fp": 2, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/37634", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 2, "fn": 2}, "gemini": {"tp": 3, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "augment": {"tp": 2, "fp": 4, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "graphite": {"tp": 2, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 2, "fp": 2, "fn": 2}, "copilot": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 2, "fp": 1, "fn": 2}, "mra-max": {"tp": 2, "fp": 5, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 6, "fn": 3}, "mra-b": {"tp": 2, "fp": 4, "fn": 2}, "mra-a": {"tp": 2, "fp": 5, "fn": 2}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 1, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 7, "fn": 2}, "codeant": {"tp": 2, "fp": 6, "fn": 2}, "kodus": {"tp": 2, "fp": 10, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "claude-code": {"tp": 0, "fp": 3, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 2}, "macroscope": {"tp": 2, "fp": 2, "fn": 2}, "tusk": {"tp": 2, "fp": 0, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/38446", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 6, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 0, "fp": 3, "fn": 2}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 2}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 1, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 7, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 2}, "mra-a": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 0, "fp": 5, "fn": 2}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 1, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 2}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 2}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 5, "fn": 0}, "macroscope": {"tp": 0, "fp": 4, "fn": 2}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36882", "language": "Java", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 4, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 4, "fn": 0}, "augment": {"tp": 0, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 0, "fp": 2, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/36880", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 0, "fp": 6, "fn": 3}, "gemini": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 3, "fp": 2, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 6, "fn": 2}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 2}, "devin": {"tp": 0, "fp": 1, "fn": 3}, "mra-max": {"tp": 2, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 4, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 0, "fp": 5, "fn": 3}, "mra-a": {"tp": 3, "fp": 2, "fn": 0}, "entelligence": {"tp": 0, "fp": 5, "fn": 3}, "sourcery": {"tp": 1, "fp": 5, "fn": 2}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 0, "fp": 11, "fn": 3}, "kodus": {"tp": 0, "fp": 6, "fn": 3}, "claude-code": {"tp": 0, "fp": 1, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v2-2": {"tp": 1, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 2}}}, {"url": "https://github.com/keycloak/keycloak/pull/37038", "language": "Java", "pr_size": "small", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "very_subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"qodo": {"tp": 1, "fp": 3, "fn": 1}, "gemini": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 1}, "mra-a": {"tp": 2, "fp": 6, "fn": 0}, "mra-b": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 1, "fp": 2, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 10, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 1, "fp": 11, "fn": 1}, "kodus": {"tp": 2, "fp": 8, "fn": 0}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 0}, "macroscope": {"tp": 2, "fp": 2, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/keycloak/keycloak/pull/33832", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"claude": {"tp": 2, "fp": 4, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 1, "fp": 1, "fn": 1}, "qodo": {"tp": 0, "fp": 6, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 4, "fn": 2}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 2, "fp": 9, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 9, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 2}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 1, "fp": 0, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 5, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/40940", "language": "Java", "pr_size": "small", "domain": "concurrency", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 1, "fn": 2}, "gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 2, "fn": 0}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 1, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 2, "fp": 2, "fn": 0}, "mra-b": {"tp": 2, "fp": 2, "fn": 0}, "mra-a": {"tp": 2, "fp": 7, "fn": 0}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 5, "fn": 0}, "codeant": {"tp": 2, "fp": 3, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1", "language": "Java", "pr_size": "medium", "domain": "authentication", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"augment": {"tp": 2, "fp": 0, "fn": 0}, "gemini": {"tp": 2, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 0}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 2, "fp": 7, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 4, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 8, "fn": 0}, "mra-b": {"tp": 2, "fp": 5, "fn": 0}, "mra-a": {"tp": 2, "fp": 1, "fn": 0}, "entelligence": {"tp": 1, "fp": 4, "fn": 1}, "sourcery": {"tp": 2, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 2, "fp": 1, "fn": 0}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 1, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 0, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 0, "fn": 0}, "macroscope": {"tp": 2, "fp": 3, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/getsentry/sentry/pull/93824", "language": "Python", "pr_size": "large", "domain": "concurrency", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"baz": {"tp": 0, "fp": 0, "fn": 5}, "augment": {"tp": 5, "fp": 1, "fn": 0}, "gemini": {"tp": 1, "fp": 1, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 4}, "coderabbit": {"tp": 1, "fp": 3, "fn": 4}, "graphite": {"tp": 1, "fp": 0, "fn": 4}, "claude": {"tp": 0, "fp": 3, "fn": 5}, "qodo": {"tp": 1, "fp": 3, "fn": 4}, "copilot": {"tp": 1, "fp": 5, "fn": 4}, "propel": {"tp": 2, "fp": 4, "fn": 3}, "greptile": {"tp": 1, "fp": 1, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 4, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 4}, "mra-max": {"tp": 0, "fp": 5, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 5}, "mra-ultra": {"tp": 0, "fp": 7, "fn": 5}, "mra-b": {"tp": 1, "fp": 10, "fn": 4}, "mra-a": {"tp": 1, "fp": 6, "fn": 4}, "entelligence": {"tp": 1, "fp": 0, "fn": 4}, "sourcery": {"tp": 4, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 4, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 5}, "codeant": {"tp": 1, "fp": 11, "fn": 4}, "kodus": {"tp": 3, "fp": 2, "fn": 2}, "claude-code": {"tp": 3, "fp": 1, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2-2": {"tp": 0, "fp": 3, "fn": 5}, "qodo-v22": {"tp": 2, "fp": 3, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 4, "fn": 4}, "cubic-v2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 3}, "macroscope": {"tp": 3, "fp": 1, "fn": 2}, "tusk": {"tp": 2, "fp": 1, "fn": 3}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5", "language": "Python", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 3}, "greptile": {"tp": 0, "fp": 2, "fn": 3}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "coderabbit": {"tp": 1, "fp": 16, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 0, "fp": 0, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 2}, "devin": {"tp": 1, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 16, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 9, "fn": 1}, "mra-a": {"tp": 1, "fp": 8, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 0, "fp": 4, "fn": 3}, "cubic-dev": {"tp": 2, "fp": 28, "fn": 1}, "mesa": {"tp": 0, "fp": 19, "fn": 3}, "codeant": {"tp": 1, "fp": 17, "fn": 2}, "kodus": {"tp": 1, "fp": 17, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 1, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 2}, "macroscope": {"tp": 0, "fp": 5, "fn": 3}, "tusk": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1", "language": "Python", "pr_size": "medium", "domain": "API", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 3}, "claude": {"tp": 2, "fp": 7, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 1, "fp": 2, "fn": 3}, "copilot": {"tp": 3, "fp": 13, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 1, "fn": 1}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 0, "fn": 4}, "coderabbit": {"tp": 3, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 3, "fn": 3}, "kg": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 2}, "devin": {"tp": 3, "fp": 2, "fn": 1}, "mra-max": {"tp": 1, "fp": 5, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 5, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 3, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 4}, "mra-a": {"tp": 3, "fp": 0, "fn": 1}, "entelligence": {"tp": 2, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 3, "fn": 1}, "mesa": {"tp": 2, "fp": 8, "fn": 2}, "codeant": {"tp": 2, "fp": 20, "fn": 2}, "kodus": {"tp": 2, "fp": 1, "fn": 2}, "claude-code": {"tp": 2, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 6, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 2}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 2}, "tusk": {"tp": 3, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/97529", "language": "Go", "pr_size": "small", "domain": "concurrency", "change_type": "performance", "complexity": "complex", "difficulty": "very_subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 3, "fn": 1}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 2, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 8, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 4, "fn": 0}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 1, "fp": 3, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 4, "fn": 0}, "mesa": {"tp": 1, "fp": 11, "fn": 1}, "codeant": {"tp": 1, "fp": 10, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80168", "language": "Python", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 2, "fn": 0}, "baz": {"tp": 0, "fp": 0, "fn": 2}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 5, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 2, "fp": 3, "fn": 0}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/80528", "language": "Python", "pr_size": "small", "domain": "scheduling", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 0, "fp": 0, "fn": 2}, "copilot": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 3, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 0, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 2, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 1, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 1, "fn": 1}, "mra-a": {"tp": 1, "fp": 7, "fn": 1}, "entelligence": {"tp": 0, "fp": 4, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 1, "fn": 1}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/77754", "language": "Python", "pr_size": "medium", "domain": "serialization", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 3}, "claude": {"tp": 2, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 2, "fp": 2, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 3}, "copilot": {"tp": 2, "fp": 0, "fn": 2}, "propel": {"tp": 2, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 0, "fn": 3}, "baz": {"tp": 1, "fp": 0, "fn": 3}, "coderabbit": {"tp": 3, "fp": 2, "fn": 1}, "greptile": {"tp": 2, "fp": 0, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 2, "fn": 2}, "greptile-v4": {"tp": 3, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 4}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 3}, "mra-b": {"tp": 2, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 3}, "entelligence": {"tp": 1, "fp": 0, "fn": 3}, "sourcery": {"tp": 2, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 4, "fn": 3}, "codeant": {"tp": 2, "fp": 4, "fn": 2}, "kodus": {"tp": 1, "fp": 2, "fn": 3}, "claude-code": {"tp": 1, "fp": 0, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 0, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 0, "fn": 3}, "macroscope": {"tp": 1, "fp": 0, "fn": 3}, "tusk": {"tp": 3, "fp": 2, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/95633", "language": "Python", "pr_size": "medium", "domain": "concurrency", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 3, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "bugbot": {"tp": 0, "fp": 2, "fn": 3}, "copilot": {"tp": 0, "fp": 6, "fn": 3}, "propel": {"tp": 0, "fp": 1, "fn": 3}, "augment": {"tp": 0, "fp": 7, "fn": 3}, "baz": {"tp": 0, "fp": 3, "fn": 3}, "coderabbit": {"tp": 0, "fp": 7, "fn": 3}, "greptile": {"tp": 0, "fp": 1, "fn": 3}, "kg": {"tp": 0, "fp": 2, "fn": 3}, "qodo-v2": {"tp": 0, "fp": 4, "fn": 3}, "devin": {"tp": 0, "fp": 4, "fn": 3}, "mra-max": {"tp": 0, "fp": 5, "fn": 3}, "greptile-v4": {"tp": 0, "fp": 7, "fn": 3}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 3}, "mra-b": {"tp": 1, "fp": 3, "fn": 2}, "mra-a": {"tp": 1, "fp": 7, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 3}, "mesa": {"tp": 0, "fp": 6, "fn": 3}, "codeant": {"tp": 1, "fp": 13, "fn": 2}, "kodus": {"tp": 1, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 3}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 0, "fp": 7, "fn": 3}, "macroscope": {"tp": 0, "fp": 2, "fn": 3}, "tusk": {"tp": 1, "fp": 7, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2", "language": "Python", "pr_size": "medium", "domain": "data_processing", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 3}, "claude": {"tp": 2, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 8, "fn": 1}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 3, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 3, "fn": 1}, "baz": {"tp": 3, "fp": 1, "fn": 0}, "coderabbit": {"tp": 0, "fp": 0, "fn": 3}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 0}, "devin": {"tp": 0, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 12, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 6, "fn": 0}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 13, "fn": 1}, "mra-b": {"tp": 3, "fp": 8, "fn": 0}, "mra-a": {"tp": 3, "fp": 5, "fn": 0}, "entelligence": {"tp": 3, "fp": 1, "fn": 0}, "sourcery": {"tp": 2, "fp": 10, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 14, "fn": 3}, "codeant": {"tp": 2, "fp": 8, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 5, "fn": 1}, "qodo-v2-2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 5, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 1}, "tusk": {"tp": 3, "fp": 2, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3", "language": "Python", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 4, "fn": 3}, "claude": {"tp": 1, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 2, "fp": 4, "fn": 1}, "bugbot": {"tp": 0, "fp": 3, "fn": 3}, "copilot": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 7, "fn": 3}, "coderabbit": {"tp": 2, "fp": 3, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 13, "fn": 3}, "greptile-v4": {"tp": 1, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 1}, "mra-b": {"tp": 2, "fp": 4, "fn": 1}, "mra-a": {"tp": 1, "fp": 7, "fn": 2}, "entelligence": {"tp": 1, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 6, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 9, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 2}, "codeant": {"tp": 1, "fp": 12, "fn": 2}, "kodus": {"tp": 1, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "linearb": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 3, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 3, "fn": 2}}}, {"url": "https://github.com/grafana/grafana/pull/103633", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 6, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 6, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 3, "fn": 2}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 9, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 10, "fn": 2}, "mra-b": {"tp": 0, "fp": 10, "fn": 2}, "mra-a": {"tp": 0, "fp": 9, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 1, "fp": 7, "fn": 1}, "kodus": {"tp": 1, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 3, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 5, "fn": 1}, "macroscope": {"tp": 0, "fp": 3, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/getsentry/sentry/pull/67876", "language": "Python", "pr_size": "small", "domain": "authentication", "change_type": "security_patch", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 2}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 7, "fn": 2}, "bugbot": {"tp": 1, "fp": 2, "fn": 2}, "copilot": {"tp": 1, "fp": 4, "fn": 2}, "augment": {"tp": 1, "fp": 2, "fn": 2}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 1, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 5, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 2}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 2}, "mra-b": {"tp": 1, "fp": 7, "fn": 2}, "mra-a": {"tp": 0, "fp": 1, "fn": 3}, "entelligence": {"tp": 1, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 9, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 1, "fn": 0}, "mesa": {"tp": 1, "fp": 3, "fn": 2}, "codeant": {"tp": 1, "fp": 9, "fn": 2}, "kodus": {"tp": 0, "fp": 2, "fn": 3}, "claude-code": {"tp": 2, "fp": 2, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 2}, "tusk": {"tp": 2, "fp": 3, "fn": 1}}}, {"url": "https://github.com/keycloak/keycloak/pull/32918", "language": "Java", "pr_size": "small", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 0, "fp": 2, "fn": 2}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 0, "fp": 1, "fn": 2}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 6, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 2}, "mra-b": {"tp": 0, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 4, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/94942", "language": "Go", "pr_size": "small", "domain": "data_processing", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 2, "fp": 9, "fn": 0}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 1, "fp": 2, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v2": {"tp": 2, "fp": 0, "fn": 0}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 9, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "mra-a": {"tp": 2, "fp": 2, "fn": 0}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 2, "fp": 1, "fn": 0}, "codeant": {"tp": 2, "fp": 7, "fn": 0}, "kodus": {"tp": 2, "fp": 0, "fn": 0}, "claude-code": {"tp": 1, "fp": 3, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/90939", "language": "Go", "pr_size": "small", "domain": "caching", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 0, "fn": 1}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 0, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "augment": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 1, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 8, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 0, "fn": 1}, "mra-nano": {"tp": 0, "fp": 2, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 1, "fn": 1}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 0, "fp": 3, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 2, "fn": 1}, "codeant": {"tp": 1, "fp": 3, "fn": 1}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 0, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 0, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/80329", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "performance", "complexity": "simple", "difficulty": "obvious", "risk": "low", "context": "local", "concern": "maintainability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 0}, "claude": {"tp": 1, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 1, "fp": 4, "fn": 0}, "copilot": {"tp": 1, "fp": 4, "fn": 0}, "propel": {"tp": 0, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 0}, "baz": {"tp": 0, "fp": 1, "fn": 1}, "coderabbit": {"tp": 1, "fp": 6, "fn": 0}, "greptile": {"tp": 1, "fp": 2, "fn": 0}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 6, "fn": 0}, "devin": {"tp": 1, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 6, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 0}, "mra-nano": {"tp": 0, "fp": 6, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 5, "fn": 1}, "mra-a": {"tp": 0, "fp": 7, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 1}, "sourcery": {"tp": 1, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 1, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 0}, "codeant": {"tp": 1, "fp": 9, "fn": 0}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 1, "fp": 7, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 1, "fp": 3, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 0}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 4, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/90045", "language": "Go", "pr_size": "medium", "domain": "logging", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 5, "fn": 0}, "claude": {"tp": 3, "fp": 4, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 4, "fn": 0}, "bugbot": {"tp": 3, "fp": 2, "fn": 0}, "copilot": {"tp": 3, "fp": 8, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 2}, "augment": {"tp": 3, "fp": 5, "fn": 0}, "baz": {"tp": 2, "fp": 2, "fn": 1}, "coderabbit": {"tp": 3, "fp": 3, "fn": 0}, "greptile": {"tp": 3, "fp": 1, "fn": 0}, "kg": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 6, "fn": 1}, "devin": {"tp": 3, "fp": 2, "fn": 0}, "mra-max": {"tp": 0, "fp": 5, "fn": 3}, "greptile-v4": {"tp": 3, "fp": 4, "fn": 0}, "mra-nano": {"tp": 0, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 3}, "mra-b": {"tp": 1, "fp": 2, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 3}, "entelligence": {"tp": 0, "fp": 4, "fn": 3}, "sourcery": {"tp": 3, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 3, "fp": 2, "fn": 0}, "mesa": {"tp": 3, "fp": 6, "fn": 0}, "codeant": {"tp": 3, "fp": 15, "fn": 0}, "kodus": {"tp": 2, "fp": 4, "fn": 1}, "claude-code": {"tp": 3, "fp": 5, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 2, "fp": 6, "fn": 1}, "qodo-extended": {"tp": 3, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 0}, "macroscope": {"tp": 3, "fp": 0, "fn": 0}, "tusk": {"tp": 3, "fp": 1, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/106778", "language": "Go", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 0, "fp": 9, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 2, "fp": 0, "fn": 0}, "copilot": {"tp": 1, "fp": 1, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 1, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 7, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 1}, "devin": {"tp": 2, "fp": 1, "fn": 0}, "mra-max": {"tp": 0, "fp": 23, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 10, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 8, "fn": 2}, "mra-b": {"tp": 0, "fp": 7, "fn": 2}, "mra-a": {"tp": 0, "fp": 10, "fn": 2}, "entelligence": {"tp": 0, "fp": 2, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 7, "fn": 0}, "mesa": {"tp": 0, "fp": 8, "fn": 2}, "codeant": {"tp": 0, "fp": 14, "fn": 2}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 5, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/grafana/grafana/pull/107534", "language": "Go", "pr_size": "small", "domain": "testing", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "low", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 0, "fp": 3, "fn": 1}, "bugbot": {"tp": 0, "fp": 2, "fn": 1}, "copilot": {"tp": 0, "fp": 4, "fn": 1}, "augment": {"tp": 0, "fp": 0, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 2, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 2, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 0}, "mra-nano": {"tp": 0, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 0}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 5, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 1}, "sourcery": {"tp": 0, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 1}, "codeant": {"tp": 1, "fp": 5, "fn": 0}, "kodus": {"tp": 0, "fp": 3, "fn": 1}, "claude-code": {"tp": 0, "fp": 7, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 0, "fp": 3, "fn": 1}, "macroscope": {"tp": 0, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 5, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/79265", "language": "Go", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 4}, "claude": {"tp": 1, "fp": 2, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 2, "fp": 1, "fn": 3}, "copilot": {"tp": 3, "fp": 6, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 5}, "augment": {"tp": 1, "fp": 1, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 4}, "baz": {"tp": 0, "fp": 1, "fn": 5}, "coderabbit": {"tp": 4, "fp": 5, "fn": 1}, "greptile": {"tp": 1, "fp": 2, "fn": 4}, "kg": {"tp": 1, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 4}, "devin": {"tp": 1, "fp": 1, "fn": 4}, "mra-max": {"tp": 1, "fp": 11, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 3}, "mra-nano": {"tp": 2, "fp": 3, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 15, "fn": 3}, "mra-b": {"tp": 2, "fp": 5, "fn": 3}, "mra-a": {"tp": 2, "fp": 4, "fn": 3}, "entelligence": {"tp": 0, "fp": 0, "fn": 5}, "sourcery": {"tp": 3, "fp": 3, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 6, "fn": 2}, "mesa": {"tp": 2, "fp": 9, "fn": 3}, "codeant": {"tp": 4, "fp": 9, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 4}, "claude-code": {"tp": 0, "fp": 4, "fn": 5}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 2, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 5}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 4}, "qodo-extended": {"tp": 1, "fp": 5, "fn": 4}, "cubic-v2": {"tp": 3, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 2, "fn": 5}, "tusk": {"tp": 3, "fp": 1, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9", "language": "Ruby", "pr_size": "small", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 0, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 1, "fn": 1}, "bugbot": {"tp": 1, "fp": 0, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 1}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 3, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 4, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 1, "fn": 2}, "mra-max": {"tp": 1, "fp": 17, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 9, "fn": 1}, "mra-b": {"tp": 1, "fp": 6, "fn": 1}, "mra-a": {"tp": 0, "fp": 4, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 0, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 1, "fp": 6, "fn": 1}, "codeant": {"tp": 0, "fp": 10, "fn": 2}, "kodus": {"tp": 0, "fp": 4, "fn": 2}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 0, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 3, "fn": 1}}}, {"url": "https://github.com/grafana/grafana/pull/76186", "language": "Go", "pr_size": "small", "domain": "logging", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 3, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 3, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "copilot": {"tp": 0, "fp": 0, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 1, "fp": 0, "fn": 1}, "greptile": {"tp": 0, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 1, "fp": 8, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 3, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 3, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 6, "fn": 2}, "sourcery": {"tp": 0, "fp": 6, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 0}, "mesa": {"tp": 0, "fp": 4, "fn": 2}, "codeant": {"tp": 1, "fp": 12, "fn": 1}, "kodus": {"tp": 0, "fp": 3, "fn": 2}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 2, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 0, "fp": 1, "fn": 2}, "tusk": {"tp": 1, "fp": 1, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10", "language": "Ruby", "pr_size": "large", "domain": "configuration", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 10, "fn": 3}, "claude": {"tp": 0, "fp": 3, "fn": 4}, "qodo": {"tp": 2, "fp": 4, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "bugbot": {"tp": 3, "fp": 4, "fn": 1}, "copilot": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 3, "fp": 1, "fn": 1}, "augment": {"tp": 3, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 0, "fn": 2}, "coderabbit": {"tp": 2, "fp": 15, "fn": 2}, "greptile": {"tp": 2, "fp": 3, "fn": 2}, "kg": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v2": {"tp": 2, "fp": 8, "fn": 2}, "devin": {"tp": 0, "fp": 2, "fn": 4}, "mra-max": {"tp": 0, "fp": 25, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 6, "fn": 2}, "mra-nano": {"tp": 0, "fp": 16, "fn": 4}, "mra-ultra": {"tp": 2, "fp": 11, "fn": 2}, "mra-b": {"tp": 1, "fp": 6, "fn": 3}, "mra-a": {"tp": 2, "fp": 6, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 4}, "sourcery": {"tp": 4, "fp": 11, "fn": 0}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 4}, "mesa": {"tp": 2, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 16, "fn": 2}, "kodus": {"tp": 3, "fp": 20, "fn": 1}, "claude-code": {"tp": 0, "fp": 6, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 2, "fp": 10, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 7, "fn": 2}, "macroscope": {"tp": 2, "fp": 5, "fn": 2}, "tusk": {"tp": 2, "fp": 4, "fn": 2}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "bug_fix", "complexity": "simple", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 2, "fn": 0}, "claude": {"tp": 3, "fp": 5, "fn": 0}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 0, "fp": 3, "fn": 3}, "bugbot": {"tp": 3, "fp": 1, "fn": 0}, "copilot": {"tp": 3, "fp": 1, "fn": 0}, "augment": {"tp": 3, "fp": 2, "fn": 0}, "propel": {"tp": 2, "fp": 1, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 2, "fn": 3}, "greptile": {"tp": 2, "fp": 4, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 3, "fp": 1, "fn": 0}, "devin": {"tp": 3, "fp": 3, "fn": 0}, "mra-max": {"tp": 0, "fp": 20, "fn": 3}, "greptile-v4": {"tp": 2, "fp": 2, "fn": 1}, "mra-nano": {"tp": 0, "fp": 0, "fn": 3}, "mra-ultra": {"tp": 0, "fp": 5, "fn": 3}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 2, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 3}, "sourcery": {"tp": 1, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 2, "fp": 3, "fn": 1}, "mesa": {"tp": 2, "fp": 5, "fn": 1}, "codeant": {"tp": 2, "fp": 6, "fn": 1}, "kodus": {"tp": 0, "fp": 5, "fn": 3}, "claude-code": {"tp": 3, "fp": 4, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-v2-2": {"tp": 2, "fp": 2, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 3}, "qodo-extended": {"tp": 2, "fp": 1, "fn": 1}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 3, "fp": 1, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 2}, "tusk": {"tp": 3, "fp": 2, "fn": 0}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8", "language": "Ruby", "pr_size": "medium", "domain": "API", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 2}, "claude": {"tp": 2, "fp": 5, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 1, "fp": 6, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 2}, "copilot": {"tp": 1, "fp": 7, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 1}, "propel": {"tp": 1, "fp": 4, "fn": 2}, "baz": {"tp": 1, "fp": 2, "fn": 2}, "coderabbit": {"tp": 2, "fp": 8, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 2}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 1, "fp": 2, "fn": 2}, "devin": {"tp": 1, "fp": 0, "fn": 2}, "mra-max": {"tp": 1, "fp": 16, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 1}, "mra-nano": {"tp": 1, "fp": 7, "fn": 2}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 0}, "mra-b": {"tp": 1, "fp": 5, "fn": 2}, "mra-a": {"tp": 1, "fp": 4, "fn": 2}, "entelligence": {"tp": 2, "fp": 4, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 15, "fn": 0}, "mesa": {"tp": 2, "fp": 8, "fn": 1}, "codeant": {"tp": 0, "fp": 11, "fn": 3}, "kodus": {"tp": 3, "fp": 8, "fn": 0}, "claude-code": {"tp": 1, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 1, "fp": 5, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 1, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 2}, "macroscope": {"tp": 1, "fp": 3, "fn": 2}, "tusk": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3", "language": "Ruby", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 4, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 5, "fn": 1}, "copilot": {"tp": 0, "fp": 2, "fn": 2}, "bugbot": {"tp": 0, "fp": 2, "fn": 2}, "augment": {"tp": 1, "fp": 5, "fn": 1}, "propel": {"tp": 1, "fp": 1, "fn": 1}, "baz": {"tp": 0, "fp": 4, "fn": 2}, "coderabbit": {"tp": 1, "fp": 9, "fn": 1}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 5, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 0, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 6, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 9, "fn": 1}, "mra-b": {"tp": 0, "fp": 0, "fn": 2}, "entelligence": {"tp": 0, "fp": 1, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 2, "fn": 0}, "mesa": {"tp": 1, "fp": 5, "fn": 1}, "codeant": {"tp": 2, "fp": 15, "fn": 0}, "kodus": {"tp": 2, "fp": 5, "fn": 0}, "claude-code": {"tp": 1, "fp": 6, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 1, "fp": 3, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 6, "fn": 0}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 1, "fp": 2, "fn": 1}, "macroscope": {"tp": 2, "fp": 0, "fn": 0}, "tusk": {"tp": 1, "fp": 2, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5", "language": "Ruby", "pr_size": "small", "domain": "UI", "change_type": "performance", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 1, "fn": 2}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 6, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 0, "fp": 3, "fn": 2}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 1, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 1, "fn": 2}, "greptile": {"tp": 0, "fp": 2, "fn": 2}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 0, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 13, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 3, "fn": 1}, "mra-nano": {"tp": 0, "fp": 8, "fn": 2}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 1, "fp": 5, "fn": 1}, "mra-a": {"tp": 0, "fp": 10, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 5, "fn": 1}, "mesa": {"tp": 0, "fp": 10, "fn": 2}, "codeant": {"tp": 0, "fp": 7, "fn": 2}, "kodus": {"tp": 0, "fp": 0, "fn": 2}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 1, "fn": 2}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 0, "fp": 0, "fn": 2}, "tusk": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6", "language": "Ruby", "pr_size": "small", "domain": "serialization", "change_type": "feature", "complexity": "simple", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 1}, "claude": {"tp": 0, "fp": 0, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 1}, "qodo": {"tp": 1, "fp": 5, "fn": 0}, "bugbot": {"tp": 0, "fp": 3, "fn": 1}, "copilot": {"tp": 1, "fp": 2, "fn": 0}, "augment": {"tp": 0, "fp": 3, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 1}, "baz": {"tp": 0, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 6, "fn": 1}, "greptile": {"tp": 0, "fp": 3, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 0, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 3, "fn": 1}, "mra-max": {"tp": 0, "fp": 11, "fn": 1}, "greptile-v4": {"tp": 0, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 3, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 4, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 1}, "mra-a": {"tp": 0, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 3, "fn": 1}, "sourcery": {"tp": 0, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 0, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 3, "fn": 1}, "codeant": {"tp": 0, "fp": 15, "fn": 1}, "kodus": {"tp": 0, "fp": 2, "fn": 1}, "claude-code": {"tp": 0, "fp": 3, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 1}, "sentry": {"tp": 0, "fp": 0, "fn": 1}, "bito": {"tp": 0, "fp": 0, "fn": 1}, "linearb": {"tp": 0, "fp": 1, "fn": 1}, "kodus-v2": {"tp": 0, "fp": 4, "fn": 1}, "qodo-v2-2": {"tp": 0, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 0, "fp": 7, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 5, "fn": 1}, "macroscope": {"tp": 0, "fp": 2, "fn": 1}, "tusk": {"tp": 0, "fp": 3, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4", "language": "Ruby", "pr_size": "large", "domain": "API", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 7, "fn": 3}, "claude": {"tp": 2, "fp": 5, "fn": 4}, "graphite": {"tp": 0, "fp": 0, "fn": 6}, "qodo": {"tp": 3, "fp": 3, "fn": 3}, "bugbot": {"tp": 1, "fp": 4, "fn": 5}, "copilot": {"tp": 1, "fp": 5, "fn": 5}, "augment": {"tp": 2, "fp": 3, "fn": 4}, "propel": {"tp": 1, "fp": 1, "fn": 5}, "baz": {"tp": 1, "fp": 4, "fn": 5}, "coderabbit": {"tp": 3, "fp": 25, "fn": 3}, "greptile": {"tp": 2, "fp": 5, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 6}, "qodo-v2": {"tp": 3, "fp": 14, "fn": 3}, "devin": {"tp": 1, "fp": 3, "fn": 5}, "mra-max": {"tp": 1, "fp": 17, "fn": 5}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 4}, "mra-nano": {"tp": 1, "fp": 5, "fn": 5}, "mra-ultra": {"tp": 1, "fp": 7, "fn": 5}, "mra-b": {"tp": 2, "fp": 10, "fn": 4}, "mra-a": {"tp": 1, "fp": 6, "fn": 5}, "entelligence": {"tp": 1, "fp": 1, "fn": 5}, "sourcery": {"tp": 1, "fp": 4, "fn": 5}, "cubic-dev": {"tp": 3, "fp": 29, "fn": 3}, "mesa": {"tp": 1, "fp": 12, "fn": 5}, "codeant": {"tp": 3, "fp": 12, "fn": 3}, "kodus": {"tp": 2, "fp": 20, "fn": 4}, "claude-code": {"tp": 2, "fp": 4, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 6}, "bito": {"tp": 0, "fp": 0, "fn": 6}, "sentry": {"tp": 0, "fp": 0, "fn": 6}, "linearb": {"tp": 0, "fp": 1, "fn": 6}, "kodus-v2": {"tp": 2, "fp": 3, "fn": 4}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 3}, "qodo-v22": {"tp": 2, "fp": 4, "fn": 4}, "qodo-extended": {"tp": 4, "fp": 7, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 4, "fn": 5}, "qodo-extended-summary": {"tp": 4, "fp": 5, "fn": 2}, "macroscope": {"tp": 1, "fp": 5, "fn": 5}, "tusk": {"tp": 2, "fp": 4, "fn": 4}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1", "language": "Ruby", "pr_size": "medium", "domain": "file_io", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 3}, "qodo": {"tp": 3, "fp": 3, "fn": 0}, "bugbot": {"tp": 2, "fp": 2, "fn": 1}, "copilot": {"tp": 2, "fp": 7, "fn": 1}, "augment": {"tp": 2, "fp": 4, "fn": 1}, "propel": {"tp": 2, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 1}, "coderabbit": {"tp": 2, "fp": 5, "fn": 1}, "greptile": {"tp": 2, "fp": 2, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 2, "fp": 4, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 1}, "mra-max": {"tp": 2, "fp": 17, "fn": 1}, "greptile-v4": {"tp": 3, "fp": 3, "fn": 0}, "mra-nano": {"tp": 2, "fp": 1, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 6, "fn": 1}, "mra-b": {"tp": 1, "fp": 4, "fn": 2}, "mra-a": {"tp": 1, "fp": 5, "fn": 2}, "entelligence": {"tp": 2, "fp": 2, "fn": 1}, "sourcery": {"tp": 2, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 3, "fp": 5, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 1}, "codeant": {"tp": 2, "fp": 11, "fn": 1}, "kodus": {"tp": 2, "fp": 2, "fn": 1}, "claude-code": {"tp": 3, "fp": 3, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 3}, "bito": {"tp": 0, "fp": 0, "fn": 3}, "sentry": {"tp": 0, "fp": 0, "fn": 3}, "linearb": {"tp": 0, "fp": 1, "fn": 3}, "kodus-v2": {"tp": 2, "fp": 3, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 2}, "tusk": {"tp": 2, "fp": 4, "fn": 1}}}, {"url": "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2", "language": "Ruby", "pr_size": "medium", "domain": "UI", "change_type": "feature", "complexity": "moderate", "difficulty": "moderate", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 0, "fn": 0}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 2, "fp": 3, "fn": 0}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 2, "fp": 3, "fn": 0}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 1, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 0, "fn": 1}, "coderabbit": {"tp": 2, "fp": 7, "fn": 0}, "greptile": {"tp": 2, "fp": 1, "fn": 0}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 15, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 4, "fn": 0}, "mra-nano": {"tp": 2, "fp": 11, "fn": 0}, "mra-ultra": {"tp": 0, "fp": 8, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 1, "fp": 8, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 5, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 8, "fn": 0}, "mesa": {"tp": 2, "fp": 2, "fn": 0}, "codeant": {"tp": 2, "fp": 11, "fn": 0}, "kodus": {"tp": 1, "fp": 5, "fn": 1}, "claude-code": {"tp": 0, "fp": 5, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 3, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 2, "fn": 1}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 4, "fn": 1}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/22532", "language": "TypeScript", "pr_size": "medium", "domain": "caching", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 1}, "claude": {"tp": 0, "fp": 1, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 5, "fn": 1}, "bugbot": {"tp": 0, "fp": 1, "fn": 2}, "copilot": {"tp": 2, "fp": 11, "fn": 0}, "augment": {"tp": 2, "fp": 4, "fn": 0}, "propel": {"tp": 1, "fp": 3, "fn": 1}, "baz": {"tp": 1, "fp": 2, "fn": 1}, "coderabbit": {"tp": 1, "fp": 10, "fn": 1}, "greptile": {"tp": 1, "fp": 1, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 1}, "devin": {"tp": 1, "fp": 1, "fn": 1}, "mra-max": {"tp": 2, "fp": 16, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 7, "fn": 1}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 8, "fn": 1}, "mra-a": {"tp": 1, "fp": 4, "fn": 1}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 3, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 0, "fp": 6, "fn": 2}, "codeant": {"tp": 1, "fp": 9, "fn": 1}, "kodus": {"tp": 2, "fp": 4, "fn": 0}, "claude-code": {"tp": 1, "fp": 4, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 5, "fn": 1}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 1, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 1, "fp": 5, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/8330", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "bug_fix", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 2, "fp": 4, "fn": 0}, "claude": {"tp": 2, "fp": 0, "fn": 0}, "graphite": {"tp": 2, "fp": 0, "fn": 0}, "qodo": {"tp": 2, "fp": 4, "fn": 0}, "bugbot": {"tp": 2, "fp": 1, "fn": 0}, "copilot": {"tp": 2, "fp": 4, "fn": 0}, "augment": {"tp": 2, "fp": 3, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 1}, "baz": {"tp": 2, "fp": 1, "fn": 0}, "coderabbit": {"tp": 2, "fp": 2, "fn": 0}, "greptile": {"tp": 2, "fp": 0, "fn": 0}, "kg": {"tp": 1, "fp": 2, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 2, "fn": 0}, "devin": {"tp": 2, "fp": 0, "fn": 0}, "mra-max": {"tp": 2, "fp": 11, "fn": 0}, "greptile-v4": {"tp": 2, "fp": 3, "fn": 0}, "mra-nano": {"tp": 1, "fp": 4, "fn": 1}, "mra-ultra": {"tp": 0, "fp": 6, "fn": 2}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 3, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 1}, "sourcery": {"tp": 2, "fp": 3, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 1, "fn": 0}, "mesa": {"tp": 2, "fp": 4, "fn": 0}, "codeant": {"tp": 2, "fp": 5, "fn": 0}, "kodus": {"tp": 2, "fp": 3, "fn": 0}, "claude-code": {"tp": 2, "fp": 2, "fn": 0}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 2, "fp": 1, "fn": 0}, "qodo-v2-2": {"tp": 1, "fp": 4, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 5, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 4, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 2, "fp": 1, "fn": 0}, "tusk": {"tp": 2, "fp": 2, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14943", "language": "TypeScript", "pr_size": "small", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 2, "fn": 2}, "claude": {"tp": 1, "fp": 0, "fn": 1}, "graphite": {"tp": 1, "fp": 0, "fn": 1}, "qodo": {"tp": 2, "fp": 1, "fn": 0}, "bugbot": {"tp": 1, "fp": 2, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 2, "fp": 1, "fn": 0}, "propel": {"tp": 2, "fp": 0, "fn": 0}, "baz": {"tp": 1, "fp": 3, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 2, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 2, "fp": 1, "fn": 0}, "devin": {"tp": 1, "fp": 0, "fn": 1}, "mra-max": {"tp": 0, "fp": 7, "fn": 2}, "greptile-v4": {"tp": 2, "fp": 1, "fn": 0}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 1, "fp": 2, "fn": 1}, "mra-b": {"tp": 1, "fp": 0, "fn": 1}, "mra-a": {"tp": 1, "fp": 2, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 2, "fp": 1, "fn": 0}, "cubic-dev": {"tp": 2, "fp": 0, "fn": 0}, "mesa": {"tp": 2, "fp": 3, "fn": 0}, "codeant": {"tp": 2, "fp": 6, "fn": 0}, "kodus": {"tp": 1, "fp": 1, "fn": 1}, "claude-code": {"tp": 0, "fp": 2, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v2-2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 2, "fp": 1, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 2, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 2, "fn": 0}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 0, "fp": 0, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/22345", "language": "TypeScript", "pr_size": "small", "domain": "database", "change_type": "migration", "complexity": "moderate", "difficulty": "subtle", "risk": "medium", "context": "file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 0, "fp": 5, "fn": 2}, "claude": {"tp": 0, "fp": 0, "fn": 2}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 0, "fp": 5, "fn": 2}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 0, "fp": 5, "fn": 2}, "augment": {"tp": 1, "fp": 3, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 3, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 0, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 1, "fp": 3, "fn": 1}, "devin": {"tp": 0, "fp": 0, "fn": 2}, "mra-max": {"tp": 0, "fp": 10, "fn": 2}, "greptile-v4": {"tp": 1, "fp": 12, "fn": 1}, "mra-nano": {"tp": 0, "fp": 3, "fn": 2}, "mra-ultra": {"tp": 1, "fp": 5, "fn": 1}, "mra-b": {"tp": 0, "fp": 3, "fn": 2}, "mra-a": {"tp": 0, "fp": 2, "fn": 2}, "entelligence": {"tp": 0, "fp": 0, "fn": 2}, "sourcery": {"tp": 1, "fp": 1, "fn": 1}, "cubic-dev": {"tp": 1, "fp": 1, "fn": 1}, "mesa": {"tp": 0, "fp": 5, "fn": 2}, "codeant": {"tp": 0, "fp": 8, "fn": 2}, "kodus": {"tp": 0, "fp": 2, "fn": 2}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 2}, "qodo-v22": {"tp": 0, "fp": 2, "fn": 2}, "qodo-extended": {"tp": 0, "fp": 6, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 1, "fn": 1}, "qodo-extended-summary": {"tp": 0, "fp": 2, "fn": 2}, "macroscope": {"tp": 1, "fp": 0, "fn": 1}, "tusk": {"tp": 1, "fp": 4, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/11059", "language": "TypeScript", "pr_size": "large", "domain": "authentication", "change_type": "feature", "complexity": "complex", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 3, "fp": 3, "fn": 2}, "claude": {"tp": 3, "fp": 3, "fn": 2}, "graphite": {"tp": 1, "fp": 1, "fn": 4}, "qodo": {"tp": 3, "fp": 9, "fn": 2}, "bugbot": {"tp": 4, "fp": 4, "fn": 1}, "copilot": {"tp": 5, "fp": 12, "fn": 0}, "augment": {"tp": 5, "fp": 8, "fn": 0}, "propel": {"tp": 1, "fp": 2, "fn": 4}, "baz": {"tp": 4, "fp": 1, "fn": 1}, "coderabbit": {"tp": 5, "fp": 14, "fn": 0}, "greptile": {"tp": 4, "fp": 7, "fn": 1}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 5, "fp": 6, "fn": 0}, "devin": {"tp": 4, "fp": 1, "fn": 1}, "mra-max": {"tp": 3, "fp": 13, "fn": 2}, "greptile-v4": {"tp": 4, "fp": 8, "fn": 1}, "mra-nano": {"tp": 2, "fp": 10, "fn": 3}, "mra-ultra": {"tp": 3, "fp": 4, "fn": 2}, "mra-b": {"tp": 4, "fp": 4, "fn": 1}, "mra-a": {"tp": 4, "fp": 6, "fn": 1}, "entelligence": {"tp": 0, "fp": 2, "fn": 5}, "sourcery": {"tp": 5, "fp": 0, "fn": 0}, "cubic-dev": {"tp": 5, "fp": 19, "fn": 0}, "mesa": {"tp": 2, "fp": 18, "fn": 3}, "codeant": {"tp": 4, "fp": 17, "fn": 1}, "kodus": {"tp": 3, "fp": 8, "fn": 2}, "claude-code": {"tp": 2, "fp": 2, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 3, "fp": 6, "fn": 2}, "qodo-v2-2": {"tp": 3, "fp": 4, "fn": 2}, "qodo-v22": {"tp": 5, "fp": 3, "fn": 0}, "qodo-extended": {"tp": 5, "fp": 8, "fn": 0}, "cubic-v2": {"tp": 3, "fp": 0, "fn": 2}, "qodo-extended-summary": {"tp": 5, "fp": 6, "fn": 0}, "macroscope": {"tp": 4, "fp": 6, "fn": 1}, "tusk": {"tp": 4, "fp": 2, "fn": 1}}}, {"url": "https://github.com/calcom/cal.com/pull/7232", "language": "TypeScript", "pr_size": "medium", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 2, "fn": 1}, "claude": {"tp": 1, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 3, "fn": 1}, "bugbot": {"tp": 1, "fp": 3, "fn": 1}, "copilot": {"tp": 2, "fp": 6, "fn": 0}, "augment": {"tp": 1, "fp": 2, "fn": 1}, "propel": {"tp": 0, "fp": 2, "fn": 2}, "baz": {"tp": 1, "fp": 1, "fn": 1}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 4, "fn": 1}, "kg": {"tp": 0, "fp": 1, "fn": 2}, "qodo-v2": {"tp": 2, "fp": 3, "fn": 0}, "devin": {"tp": 1, "fp": 4, "fn": 1}, "mra-max": {"tp": 2, "fp": 27, "fn": 0}, "greptile-v4": {"tp": 1, "fp": 8, "fn": 1}, "mra-nano": {"tp": 2, "fp": 5, "fn": 0}, "mra-ultra": {"tp": 2, "fp": 3, "fn": 0}, "mra-b": {"tp": 2, "fp": 1, "fn": 0}, "mra-a": {"tp": 1, "fp": 5, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 2, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 6, "fn": 0}, "mesa": {"tp": 0, "fp": 10, "fn": 2}, "codeant": {"tp": 1, "fp": 15, "fn": 1}, "kodus": {"tp": 2, "fp": 6, "fn": 0}, "claude-code": {"tp": 1, "fp": 5, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 1, "fp": 5, "fn": 1}, "qodo-v2-2": {"tp": 2, "fp": 0, "fn": 0}, "qodo-v22": {"tp": 2, "fp": 2, "fn": 0}, "qodo-extended": {"tp": 2, "fp": 3, "fn": 0}, "cubic-v2": {"tp": 2, "fp": 3, "fn": 0}, "qodo-extended-summary": {"tp": 2, "fp": 3, "fn": 0}, "macroscope": {"tp": 1, "fp": 2, "fn": 1}, "tusk": {"tp": 2, "fp": 3, "fn": 0}}}, {"url": "https://github.com/calcom/cal.com/pull/14740", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 3, "fn": 4}, "claude": {"tp": 2, "fp": 0, "fn": 3}, "graphite": {"tp": 1, "fp": 1, "fn": 4}, "qodo": {"tp": 3, "fp": 2, "fn": 2}, "bugbot": {"tp": 3, "fp": 3, "fn": 2}, "copilot": {"tp": 3, "fp": 8, "fn": 2}, "augment": {"tp": 4, "fp": 3, "fn": 1}, "propel": {"tp": 4, "fp": 0, "fn": 1}, "baz": {"tp": 1, "fp": 1, "fn": 4}, "coderabbit": {"tp": 4, "fp": 6, "fn": 1}, "greptile": {"tp": 4, "fp": 2, "fn": 1}, "kg": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2": {"tp": 4, "fp": 4, "fn": 1}, "devin": {"tp": 2, "fp": 0, "fn": 3}, "mra-max": {"tp": 2, "fp": 11, "fn": 3}, "greptile-v4": {"tp": 4, "fp": 7, "fn": 1}, "mra-nano": {"tp": 0, "fp": 7, "fn": 5}, "mra-ultra": {"tp": 1, "fp": 3, "fn": 4}, "mra-b": {"tp": 1, "fp": 2, "fn": 4}, "mra-a": {"tp": 1, "fp": 5, "fn": 4}, "entelligence": {"tp": 0, "fp": 3, "fn": 5}, "sourcery": {"tp": 3, "fp": 2, "fn": 2}, "cubic-dev": {"tp": 4, "fp": 11, "fn": 1}, "mesa": {"tp": 3, "fp": 5, "fn": 2}, "codeant": {"tp": 2, "fp": 13, "fn": 3}, "kodus": {"tp": 2, "fp": 6, "fn": 3}, "claude-code": {"tp": 1, "fp": 3, "fn": 4}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 0, "fn": 3}, "qodo-v2-2": {"tp": 4, "fp": 1, "fn": 1}, "qodo-v22": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended": {"tp": 4, "fp": 4, "fn": 1}, "cubic-v2": {"tp": 4, "fp": 2, "fn": 1}, "qodo-extended-summary": {"tp": 4, "fp": 1, "fn": 1}, "macroscope": {"tp": 2, "fp": 4, "fn": 3}, "tusk": {"tp": 3, "fp": 2, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/10600", "language": "TypeScript", "pr_size": "medium", "domain": "authentication", "change_type": "feature", "complexity": "moderate", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "security", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 5, "fn": 3}, "claude": {"tp": 1, "fp": 12, "fn": 3}, "graphite": {"tp": 0, "fp": 0, "fn": 4}, "qodo": {"tp": 0, "fp": 5, "fn": 4}, "bugbot": {"tp": 1, "fp": 0, "fn": 3}, "copilot": {"tp": 1, "fp": 9, "fn": 3}, "augment": {"tp": 1, "fp": 4, "fn": 3}, "propel": {"tp": 1, "fp": 1, "fn": 3}, "baz": {"tp": 0, "fp": 1, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 4}, "greptile": {"tp": 0, "fp": 4, "fn": 4}, "kg": {"tp": 0, "fp": 0, "fn": 4}, "qodo-v2": {"tp": 1, "fp": 6, "fn": 3}, "devin": {"tp": 1, "fp": 0, "fn": 3}, "mra-max": {"tp": 0, "fp": 14, "fn": 4}, "greptile-v4": {"tp": 2, "fp": 9, "fn": 2}, "mra-nano": {"tp": 1, "fp": 8, "fn": 3}, "mra-ultra": {"tp": 2, "fp": 10, "fn": 2}, "mra-b": {"tp": 2, "fp": 3, "fn": 2}, "mra-a": {"tp": 2, "fp": 10, "fn": 2}, "entelligence": {"tp": 1, "fp": 1, "fn": 3}, "sourcery": {"tp": 2, "fp": 5, "fn": 2}, "cubic-dev": {"tp": 3, "fp": 13, "fn": 1}, "mesa": {"tp": 1, "fp": 7, "fn": 3}, "codeant": {"tp": 2, "fp": 14, "fn": 2}, "kodus": {"tp": 0, "fp": 11, "fn": 4}, "claude-code": {"tp": 1, "fp": 5, "fn": 3}, "vercel": {"tp": 0, "fp": 0, "fn": 4}, "sentry": {"tp": 0, "fp": 0, "fn": 4}, "bito": {"tp": 0, "fp": 0, "fn": 4}, "linearb": {"tp": 0, "fp": 1, "fn": 4}, "kodus-v2": {"tp": 1, "fp": 1, "fn": 3}, "qodo-v2-2": {"tp": 0, "fp": 2, "fn": 4}, "qodo-v22": {"tp": 1, "fp": 4, "fn": 3}, "qodo-extended": {"tp": 1, "fp": 6, "fn": 3}, "cubic-v2": {"tp": 2, "fp": 2, "fn": 2}, "qodo-extended-summary": {"tp": 2, "fp": 4, "fn": 2}, "macroscope": {"tp": 0, "fp": 3, "fn": 4}, "tusk": {"tp": 0, "fp": 4, "fn": 4}}}, {"url": "https://github.com/calcom/cal.com/pull/10967", "language": "TypeScript", "pr_size": "large", "domain": "scheduling", "change_type": "bug_fix", "complexity": "complex", "difficulty": "subtle", "risk": "high", "context": "cross_file", "concern": "correctness", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 8, "fn": 4}, "claude": {"tp": 4, "fp": 3, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 5}, "qodo": {"tp": 0, "fp": 3, "fn": 5}, "bugbot": {"tp": 2, "fp": 4, "fn": 3}, "copilot": {"tp": 3, "fp": 8, "fn": 2}, "augment": {"tp": 2, "fp": 3, "fn": 3}, "propel": {"tp": 1, "fp": 0, "fn": 4}, "baz": {"tp": 1, "fp": 2, "fn": 4}, "coderabbit": {"tp": 0, "fp": 0, "fn": 5}, "greptile": {"tp": 2, "fp": 2, "fn": 3}, "kg": {"tp": 0, "fp": 0, "fn": 5}, "qodo-v2": {"tp": 1, "fp": 4, "fn": 4}, "devin": {"tp": 3, "fp": 2, "fn": 2}, "mra-max": {"tp": 0, "fp": 14, "fn": 5}, "greptile-v4": {"tp": 4, "fp": 5, "fn": 1}, "mra-nano": {"tp": 0, "fp": 15, "fn": 5}, "mra-ultra": {"tp": 3, "fp": 7, "fn": 2}, "mra-b": {"tp": 3, "fp": 15, "fn": 2}, "mra-a": {"tp": 2, "fp": 15, "fn": 3}, "entelligence": {"tp": 3, "fp": 4, "fn": 2}, "sourcery": {"tp": 2, "fp": 2, "fn": 3}, "cubic-dev": {"tp": 0, "fp": 0, "fn": 5}, "mesa": {"tp": 4, "fp": 11, "fn": 1}, "codeant": {"tp": 3, "fp": 7, "fn": 2}, "kodus": {"tp": 3, "fp": 6, "fn": 2}, "claude-code": {"tp": 3, "fp": 3, "fn": 2}, "vercel": {"tp": 0, "fp": 0, "fn": 5}, "bito": {"tp": 0, "fp": 0, "fn": 5}, "sentry": {"tp": 0, "fp": 0, "fn": 5}, "linearb": {"tp": 0, "fp": 1, "fn": 5}, "kodus-v2": {"tp": 2, "fp": 4, "fn": 3}, "qodo-v2-2": {"tp": 3, "fp": 5, "fn": 2}, "qodo-v22": {"tp": 3, "fp": 6, "fn": 2}, "qodo-extended": {"tp": 3, "fp": 5, "fn": 2}, "cubic-v2": {"tp": 1, "fp": 2, "fn": 4}, "qodo-extended-summary": {"tp": 4, "fp": 3, "fn": 1}, "macroscope": {"tp": 1, "fp": 4, "fn": 4}, "tusk": {"tp": 3, "fp": 2, "fn": 2}}}, {"url": "https://github.com/calcom/cal.com/pull/8087", "language": "TypeScript", "pr_size": "medium", "domain": "concurrency", "change_type": "refactoring", "complexity": "moderate", "difficulty": "subtle", "risk": "critical", "context": "cross_file", "concern": "reliability", "summary": "", "tool_metrics": {"gemini": {"tp": 1, "fp": 1, "fn": 1}, "claude": {"tp": 1, "fp": 1, "fn": 1}, "graphite": {"tp": 0, "fp": 0, "fn": 2}, "qodo": {"tp": 1, "fp": 2, "fn": 1}, "bugbot": {"tp": 1, "fp": 1, "fn": 1}, "copilot": {"tp": 1, "fp": 3, "fn": 1}, "augment": {"tp": 1, "fp": 1, "fn": 1}, "propel": {"tp": 0, "fp": 0, "fn": 2}, "baz": {"tp": 0, "fp": 1, "fn": 2}, "coderabbit": {"tp": 0, "fp": 0, "fn": 2}, "greptile": {"tp": 1, "fp": 3, "fn": 1}, "kg": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v2": {"tp": 1, "fp": 1, "fn": 1}, "devin": {"tp": 1, "fp": 6, "fn": 1}, "mra-max": {"tp": 1, "fp": 9, "fn": 1}, "greptile-v4": {"tp": 1, "fp": 4, "fn": 1}, "mra-nano": {"tp": 1, "fp": 3, "fn": 1}, "mra-ultra": {"tp": 2, "fp": 7, "fn": 0}, "mra-b": {"tp": 2, "fp": 3, "fn": 0}, "mra-a": {"tp": 1, "fp": 8, "fn": 1}, "entelligence": {"tp": 1, "fp": 0, "fn": 1}, "sourcery": {"tp": 1, "fp": 4, "fn": 1}, "cubic-dev": {"tp": 2, "fp": 11, "fn": 0}, "mesa": {"tp": 2, "fp": 7, "fn": 0}, "codeant": {"tp": 1, "fp": 8, "fn": 1}, "kodus": {"tp": 1, "fp": 3, "fn": 1}, "claude-code": {"tp": 1, "fp": 1, "fn": 1}, "vercel": {"tp": 0, "fp": 0, "fn": 2}, "sentry": {"tp": 0, "fp": 0, "fn": 2}, "bito": {"tp": 0, "fp": 0, "fn": 2}, "linearb": {"tp": 0, "fp": 1, "fn": 2}, "kodus-v2": {"tp": 0, "fp": 0, "fn": 2}, "qodo-v2-2": {"tp": 1, "fp": 0, "fn": 1}, "qodo-v22": {"tp": 1, "fp": 0, "fn": 1}, "qodo-extended": {"tp": 1, "fp": 3, "fn": 1}, "cubic-v2": {"tp": 2, "fp": 5, "fn": 0}, "qodo-extended-summary": {"tp": 1, "fp": 1, "fn": 1}, "macroscope": {"tp": 1, "fp": 1, "fn": 1}, "tusk": {"tp": 1, "fp": 9, "fn": 1}}}], "tools": ["augment", "baz", "bito", "bugbot", "claude", "claude-code", "codeant", "coderabbit", "copilot", "cubic-dev", "cubic-v2", "devin", "entelligence", "gemini", "graphite", "greptile", "greptile-v4", "kg", "kodus", "kodus-v2", "linearb", "macroscope", "mesa", "mra-a", "mra-b", "mra-max", "mra-nano", "mra-ultra", "propel", "qodo", "qodo-extended", "qodo-extended-summary", "qodo-v2", "qodo-v2-2", "qodo-v22", "sentry", "sourcery", "tusk", "vercel"], "dimensions": {"language": ["Go", "Java", "Python", "Ruby", "TypeScript"], "pr_size": ["small", "medium", "large"], "domain": ["API", "UI", "authentication", "caching", "concurrency", "configuration", "data_processing", "database", "file_io", "logging", "scheduling", "serialization", "testing"], "change_type": ["bug_fix", "feature", "migration", "performance", "refactoring", "security_patch"], "complexity": ["simple", "moderate", "complex"], "difficulty": ["obvious", "moderate", "subtle", "very_subtle"], "risk": ["low", "medium", "high", "critical"], "context": ["local", "file", "cross_file", "system"], "concern": ["correctness", "maintainability", "reliability", "security"]}, "overall_metrics": {"qodo": {"precision": 25.6, "recall": 44.5, "f1": 32.5, "tp": 61, "fp": 177, "fn": 76, "num_prs": 50}, "greptile": {"precision": 37.4, "recall": 38.0, "f1": 37.7, "tp": 52, "fp": 87, "fn": 85, "num_prs": 50}, "sentry": {"precision": 0.0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 1, "fn": 137, "num_prs": 50}, "kodus": {"precision": 21.0, "recall": 43.1, "f1": 28.2, "tp": 59, "fp": 222, "fn": 78, "num_prs": 50}, "copilot": {"precision": 24.7, "recall": 52.6, "f1": 33.6, "tp": 72, "fp": 220, "fn": 65, "num_prs": 50}, "macroscope": {"precision": 41.8, "recall": 40.9, "f1": 41.3, "tp": 56, "fp": 78, "fn": 81, "num_prs": 50}, "graphite": {"precision": 80.0, "recall": 8.8, "f1": 15.8, "tp": 12, "fp": 3, "fn": 125, "num_prs": 50}, "entelligence": {"precision": 27.7, "recall": 20.4, "f1": 23.5, "tp": 28, "fp": 73, "fn": 109, "num_prs": 50}, "sourcery": {"precision": 29.8, "recall": 53.3, "f1": 38.2, "tp": 73, "fp": 172, "fn": 64, "num_prs": 50}, "tusk": {"precision": 35.5, "recall": 56.2, "f1": 43.5, "tp": 77, "fp": 140, "fn": 60, "num_prs": 50}, "claude": {"precision": 32.3, "recall": 37.2, "f1": 34.6, "tp": 51, "fp": 107, "fn": 86, "num_prs": 50}, "baz": {"precision": 37.8, "recall": 27.0, "f1": 31.5, "tp": 37, "fp": 61, "fn": 100, "num_prs": 50}, "qodo-v22": {"precision": 38.4, "recall": 53.3, "f1": 44.6, "tp": 73, "fp": 117, "fn": 64, "num_prs": 50}, "codeant": {"precision": 11.9, "recall": 49.6, "f1": 19.2, "tp": 68, "fp": 504, "fn": 69, "num_prs": 50}, "linearb": {"precision": 8.0, "recall": 2.9, "f1": 4.3, "tp": 4, "fp": 46, "fn": 133, "num_prs": 50}, "vercel": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "coderabbit": {"precision": 21.3, "recall": 40.9, "f1": 28.0, "tp": 56, "fp": 207, "fn": 81, "num_prs": 50}, "mesa": {"precision": 15.5, "recall": 42.3, "f1": 22.7, "tp": 58, "fp": 315, "fn": 79, "num_prs": 50}, "gemini": {"precision": 26.3, "recall": 33.6, "f1": 29.5, "tp": 46, "fp": 129, "fn": 91, "num_prs": 50}, "qodo-extended-summary": {"precision": 38.1, "recall": 66.4, "f1": 48.4, "tp": 91, "fp": 148, "fn": 46, "num_prs": 50}, "cubic-dev": {"precision": 27.3, "recall": 73.0, "f1": 39.8, "tp": 100, "fp": 266, "fn": 37, "num_prs": 50}, "mra-ultra": {"precision": 15.3, "recall": 37.2, "f1": 21.7, "tp": 51, "fp": 282, "fn": 86, "num_prs": 50}, "qodo-extended": {"precision": 31.4, "recall": 59.9, "f1": 41.2, "tp": 82, "fp": 179, "fn": 55, "num_prs": 50}, "augment": {"precision": 41.9, "recall": 60.6, "f1": 49.6, "tp": 83, "fp": 115, "fn": 54, "num_prs": 50}, "mra-max": {"precision": 7.7, "recall": 31.4, "f1": 12.4, "tp": 43, "fp": 515, "fn": 94, "num_prs": 50}, "kodus-v2": {"precision": 35.9, "recall": 34.3, "f1": 35.1, "tp": 47, "fp": 84, "fn": 90, "num_prs": 50}, "kg": {"precision": 51.1, "recall": 16.8, "f1": 25.3, "tp": 23, "fp": 22, "fn": 114, "num_prs": 50}, "cubic-v2": {"precision": 53.6, "recall": 65.7, "f1": 59.0, "tp": 90, "fp": 78, "fn": 47, "num_prs": 50}, "devin": {"precision": 46.4, "recall": 37.2, "f1": 41.3, "tp": 51, "fp": 59, "fn": 86, "num_prs": 50}, "mra-b": {"precision": 19.9, "recall": 39.4, "f1": 26.5, "tp": 54, "fp": 217, "fn": 83, "num_prs": 50}, "propel": {"precision": 42.7, "recall": 36.5, "f1": 39.4, "tp": 50, "fp": 67, "fn": 87, "num_prs": 50}, "mra-a": {"precision": 17.6, "recall": 40.0, "f1": 24.4, "tp": 54, "fp": 253, "fn": 81, "num_prs": 49}, "claude-code": {"precision": 27.4, "recall": 41.6, "f1": 33.0, "tp": 57, "fp": 151, "fn": 80, "num_prs": 50}, "qodo-v2-2": {"precision": 41.5, "recall": 43.1, "f1": 42.3, "tp": 59, "fp": 83, "fn": 78, "num_prs": 50}, "greptile-v4": {"precision": 29.3, "recall": 58.4, "f1": 39.0, "tp": 80, "fp": 193, "fn": 57, "num_prs": 50}, "bito": {"precision": 0, "recall": 0.0, "f1": 0, "tp": 0, "fp": 0, "fn": 137, "num_prs": 50}, "mra-nano": {"precision": 10.8, "recall": 16.8, "f1": 13.1, "tp": 23, "fp": 190, "fn": 114, "num_prs": 50}, "bugbot": {"precision": 43.2, "recall": 43.8, "f1": 43.5, "tp": 60, "fp": 79, "fn": 77, "num_prs": 50}, "qodo-v2": {"precision": 36.2, "recall": 56.2, "f1": 44.0, "tp": 77, "fp": 136, "fn": 60, "num_prs": 50}}}};
+        const toolDisplayNames = {"graphite": "Graphite", "qodo": "Qodo", "gemini": "Gemini", "claude": "Claude Code", "augment": "Augment", "bugbot": "Cursor Bugbot", "coderabbit": "CodeRabbit", "propel": "Propel", "copilot": "GitHub Copilot", "baz": "Baz", "greptile": "Greptile", "kg": "KG", "entelligence": "Entelligence", "cubic-dev": "Cubic", "sourcery": "Sourcery", "mesa": "Mesa", "tusk": "Tusk"};
+        const toolColors = {"graphite": "#6366f1", "qodo": "#8b5cf6", "gemini": "#06b6d4", "claude": "#f59e0b", "augment": "#10b981", "bugbot": "#3b82f6", "coderabbit": "#ec4899", "propel": "#14b8a6", "copilot": "#6b7280", "baz": "#f97316", "greptile": "#22c55e", "kg": "#a855f7", "entelligence": "#0ea5e9", "cubic-dev": "#d946ef", "sourcery": "#84cc16", "mesa": "#f43f5e", "tusk": "#9333ea"};
+        const predefinedFilters = [{"id": "high_precision", "label": "Highest Precision", "filters": {}, "sort": "precision", "description": "Tools ranked by precision - fewer false positives, more reliable findings", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "graphite", "best_score": 100.0}, {"id": "tool_kodus-v2_domain_concurrency", "label": "Best for Concurrency (Precision)", "filters": {"domain": ["concurrency"]}, "description": "Threading and async operations.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "kodus-v2", "best_score": 100.0, "sort": "precision"}, {"id": "tool_kg_complexity_complex", "label": "Best for Complex Code (Precision)", "filters": {"complexity": ["complex"]}, "description": "Deep logic and dependencies.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "kg", "best_score": 100.0, "sort": "precision"}, {"id": "tool_copilot_change_type_bug_fix", "label": "Best for Bug Fixes (Recall)", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "copilot", "best_score": 82.4, "sort": "recall"}, {"id": "tool_mesa_change_type_bug_fix", "label": "Best for Bug Fixes (Recall)", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "mesa", "best_score": 82.4, "sort": "recall"}, {"id": "tool_greptile-v4_change_type_bug_fix", "label": "Best for Bug Fixes (Recall)", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "greptile-v4", "best_score": 82.4, "sort": "recall"}, {"id": "change_type_performance", "label": "Best for Performance Optimization", "filters": {"change_type": ["performance"]}, "description": "Performance optimization changes", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 81.1}, {"id": "tool_tusk_domain_ui", "label": "Best for Ui (Recall)", "filters": {"domain": ["UI"]}, "description": "User interface and frontend.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "tusk", "best_score": 76.9, "sort": "recall"}, {"id": "high_recall", "label": "Highest Recall", "filters": {}, "sort": "recall", "description": "Tools ranked by recall - catches more issues, may have more noise", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-dev", "best_score": 76.6}, {"id": "tool_qodo-v2-2_language_java_domain_authentication", "label": "Java + Authentication", "filters": {"language": ["Java"], "domain": ["authentication"]}, "description": "Java codebases with OOP patterns. Auth and access control.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v2-2", "best_score": 74.1}, {"id": "tool_mra-a_language_java_risk_high", "label": "Java + High Risk (Recall)", "filters": {"language": ["Java"], "risk": ["high"]}, "description": "Java codebases with OOP patterns. Significant impact, potential data loss.", "best_model": "openai_gpt-5.2", "best_tool": "mra-a", "best_score": 73.3, "sort": "recall"}, {"id": "tool_codeant_difficulty_moderate_context_file", "label": "Moderate Bugs + File Context (Recall)", "filters": {"difficulty": ["moderate"], "context": ["file"]}, "description": "Requires careful reading. Requires full file understanding.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "codeant", "best_score": 73.3, "sort": "recall"}, {"id": "ruby_medium", "label": "Best for Medium Ruby PRs", "filters": {"language": ["Ruby"], "pr_size": ["medium"]}, "description": "Ruby codebases with Rails patterns. Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "devin", "best_score": 72.7}, {"id": "tool_sourcery_language_typescript_concern_correctness", "label": "Typescript + Correctness", "filters": {"language": ["TypeScript"], "concern": ["correctness"]}, "description": "TypeScript codebases with frontend patterns. Logical correctness and expected behavior", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "sourcery", "best_score": 71.8}, {"id": "risk_critical", "label": "Best for Critical Risk", "filters": {"risk": ["critical"]}, "description": "Critical security or data corruption risk.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v22", "best_score": 69.8}, {"id": "tool_baz_language_java_domain_authentication", "label": "Java + Authentication", "filters": {"language": ["Java"], "domain": ["authentication"]}, "description": "Java codebases with OOP patterns. Auth and access control.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "baz", "best_score": 69.6}, {"id": "tool_qodo-extended_change_type_bug_fix_context_cross_file", "label": "Bug Fixes + Cross-File", "filters": {"change_type": ["bug_fix"], "context": ["cross_file"]}, "description": "Bug fixes and issue resolution Spans multiple files.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-extended", "best_score": 68.6}, {"id": "domain_ui", "label": "Best for Ui", "filters": {"domain": ["UI"]}, "description": "User interface and frontend.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "bugbot", "best_score": 66.7}, {"id": "tool_macroscope_context_file_concern_correctness", "label": "File Context + Correctness", "filters": {"context": ["file"], "concern": ["correctness"]}, "description": "Requires full file understanding. Logical correctness and expected behavior", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "macroscope", "best_score": 66.7}, {"id": "tool_qodo-v2_language_java_risk_high", "label": "Java + High Risk", "filters": {"language": ["Java"], "risk": ["high"]}, "description": "Java codebases with OOP patterns. Significant impact, potential data loss.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v2", "best_score": 66.7}, {"id": "tool_greptile_pr_size_medium_context_file", "label": "Medium PRs + File Context", "filters": {"pr_size": ["medium"], "context": ["file"]}, "description": "Medium PRs with 3-5 files, typical feature development Requires full file understanding.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "greptile", "best_score": 64.5}, {"id": "tool_propel_language_ruby_change_type_feature", "label": "Ruby + Features (Precision)", "filters": {"language": ["Ruby"], "change_type": ["feature"]}, "description": "Ruby codebases with Rails patterns. New feature implementation", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "propel", "best_score": 64.3, "sort": "precision"}, {"id": "complexity_complex", "label": "Best for Complex Code", "filters": {"complexity": ["complex"]}, "description": "Deep logic and dependencies.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "augment", "best_score": 62.9}, {"id": "tool_entelligence_change_type_performance_context_cross_file", "label": "Performance Optimization + Cross-File (Precision)", "filters": {"change_type": ["performance"], "context": ["cross_file"]}, "description": "Performance optimization changes Spans multiple files.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "entelligence", "best_score": 62.5, "sort": "precision"}, {"id": "java_medium", "label": "Best for Medium Java PRs", "filters": {"language": ["Java"], "pr_size": ["medium"]}, "description": "Java codebases with OOP patterns. Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-extended-summary", "best_score": 60.0}, {"id": "tool_claude_difficulty_moderate_context_file", "label": "Moderate Bugs + File Context", "filters": {"difficulty": ["moderate"], "context": ["file"]}, "description": "Requires careful reading. Requires full file understanding.", "best_model": "openai_gpt-5.2", "best_tool": "claude", "best_score": 57.1}, {"id": "tool_qodo_language_ruby_change_type_feature", "label": "Ruby + Features", "filters": {"language": ["Ruby"], "change_type": ["feature"]}, "description": "Ruby codebases with Rails patterns. New feature implementation", "best_model": "openai_gpt-5.2", "best_tool": "qodo", "best_score": 45.6}, {"id": "go_small", "label": "Best for Small Go PRs", "filters": {"language": ["Go"], "pr_size": ["small"]}, "description": "Go codebases with concurrency patterns. Small PRs with 1-2 files, easier to review thoroughly", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 77.4}, {"id": "language_go", "label": "Best for Go", "filters": {"language": ["Go"]}, "description": "Go codebases with concurrency patterns.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 75.5}, {"id": "pr_size_small", "label": "Best for Small PRs", "filters": {"pr_size": ["small"]}, "description": "Small PRs with 1-2 files, easier to review thoroughly", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 75.3}, {"id": "python_medium", "label": "Best for Medium Python PRs", "filters": {"language": ["Python"], "pr_size": ["medium"]}, "description": "Python codebases with dynamic typing. Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 71.0}, {"id": "language_python", "label": "Best for Python", "filters": {"language": ["Python"]}, "description": "Python codebases with dynamic typing.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 70.6}, {"id": "risk_high", "label": "Best for High Risk", "filters": {"risk": ["high"]}, "description": "Significant impact, potential data loss.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 70.2}, {"id": "change_type_bug_fix", "label": "Best for Bug Fixes", "filters": {"change_type": ["bug_fix"]}, "description": "Bug fixes and issue resolution", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 68.6}, {"id": "domain_scheduling", "label": "Best for Scheduling", "filters": {"domain": ["scheduling"]}, "description": "Task scheduling.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 68.4}, {"id": "context_file", "label": "Best for File Context", "filters": {"context": ["file"]}, "description": "Requires full file understanding.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 67.6}, {"id": "domain_concurrency", "label": "Best for Concurrency", "filters": {"domain": ["concurrency"]}, "description": "Threading and async operations.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-dev", "best_score": 64.5}, {"id": "difficulty_subtle", "label": "Best for Subtle Bugs", "filters": {"difficulty": ["subtle"]}, "description": "Non-obvious, needs domain knowledge.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 64.1}, {"id": "domain_caching", "label": "Best for Caching", "filters": {"domain": ["caching"]}, "description": "Cache and memoization.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 63.6}, {"id": "concern_reliability", "label": "Best for Reliability", "filters": {"concern": ["reliability"]}, "description": "Error handling and system stability", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v22", "best_score": 63.6}, {"id": "high_risk_auth", "label": "High Risk Auth", "filters": {"risk": ["high", "critical"], "domain": ["authentication"]}, "description": "Significant impact, potential data loss. Auth and access control.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 63.2}, {"id": "complex_subtle", "label": "Complex & Subtle", "filters": {"complexity": ["complex"], "difficulty": ["subtle", "very_subtle"]}, "description": "Deep logic and dependencies. Non-obvious, needs domain knowledge.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "augment", "best_score": 62.9}, {"id": "concern_correctness", "label": "Best for Correctness", "filters": {"concern": ["correctness"]}, "description": "Logical correctness and expected behavior", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 62.8}, {"id": "concern_security", "label": "Best for Security", "filters": {"concern": ["security"]}, "description": "Security vulnerabilities and attack vectors", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.9}, {"id": "security_critical", "label": "Security Critical", "filters": {"concern": ["security"], "risk": ["high", "critical"]}, "description": "Security vulnerabilities and attack vectors Significant impact, potential data loss.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.9}, {"id": "complexity_moderate", "label": "Best for Moderate Code", "filters": {"complexity": ["moderate"]}, "description": "Some abstraction.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.8}, {"id": "high_f1", "label": "Highest F1", "filters": {}, "sort": "f1", "description": "Tools ranked by F1 score - balanced precision and recall", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 61.8}, {"id": "language_typescript", "label": "Best for Typescript", "filters": {"language": ["TypeScript"]}, "description": "TypeScript codebases with frontend patterns.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "devin", "best_score": 61.8}, {"id": "language_java", "label": "Best for Java", "filters": {"language": ["Java"]}, "description": "Java codebases with OOP patterns.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "qodo-v22", "best_score": 61.5}, {"id": "language_ruby", "label": "Best for Ruby", "filters": {"language": ["Ruby"]}, "description": "Ruby codebases with Rails patterns.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 61.0}, {"id": "domain_authentication", "label": "Best for Authentication", "filters": {"domain": ["authentication"]}, "description": "Auth and access control.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 60.3}, {"id": "context_cross_file", "label": "Best for Cross-File", "filters": {"context": ["cross_file"]}, "description": "Spans multiple files.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 60.0}, {"id": "change_type_feature", "label": "Best for Features", "filters": {"change_type": ["feature"]}, "description": "New feature implementation", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 59.9}, {"id": "risk_medium", "label": "Best for Medium Risk", "filters": {"risk": ["medium"]}, "description": "Moderate user impact.", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 58.8}, {"id": "pr_size_large", "label": "Best for Large PRs", "filters": {"pr_size": ["large"]}, "description": "Large PRs with 6+ files, complex changes requiring careful review", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 58.3}, {"id": "pr_size_medium", "label": "Best for Medium PRs", "filters": {"pr_size": ["medium"]}, "description": "Medium PRs with 3-5 files, typical feature development", "best_model": "anthropic_claude-sonnet-4-5-20250929", "best_tool": "cubic-v2", "best_score": 56.7}, {"id": "difficulty_moderate", "label": "Best for Moderate Bugs", "filters": {"difficulty": ["moderate"]}, "description": "Requires careful reading.", "best_model": "anthropic_claude-opus-4-5-20251101", "best_tool": "cubic-v2", "best_score": 52.8}];
 
         let currentModel = 'anthropic_claude-opus-4-5-20251101';
         let currentFilters = {
diff --git a/offline/analysis/benchmark_dashboard.json b/offline/analysis/benchmark_dashboard.json
index 0a3ca36..3a595d2 100644
--- a/offline/analysis/benchmark_dashboard.json
+++ b/offline/analysis/benchmark_dashboard.json
@@ -204,6 +204,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -409,6 +414,11 @@
               "tp": 2,
               "fp": 2,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -614,6 +624,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -819,6 +834,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -1024,6 +1044,11 @@
               "tp": 2,
               "fp": 4,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 2
             }
           }
         },
@@ -1229,6 +1254,11 @@
               "tp": 2,
               "fp": 2,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 0
             }
           }
         },
@@ -1434,6 +1464,11 @@
               "tp": 1,
               "fp": 5,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -1639,6 +1674,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -1844,6 +1884,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -2049,6 +2094,11 @@
               "tp": 3,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 3
             }
           }
         },
@@ -2254,6 +2304,11 @@
               "tp": 0,
               "fp": 5,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -2459,6 +2514,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -2664,6 +2724,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -2869,6 +2934,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -3074,6 +3144,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 0,
+              "fn": 1
             }
           }
         },
@@ -3279,6 +3354,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -3484,6 +3564,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 2
             }
           }
         },
@@ -3689,6 +3774,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 0
             }
           }
         },
@@ -3894,6 +3984,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 2
             }
           }
         },
@@ -4099,6 +4194,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -4304,6 +4404,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -4509,6 +4614,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -4714,6 +4824,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 0,
+              "fn": 0
             }
           }
         },
@@ -4919,6 +5034,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 0,
+              "fn": 1
             }
           }
         },
@@ -5124,6 +5244,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -5329,6 +5454,11 @@
               "tp": 3,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 0
             }
           }
         },
@@ -5534,6 +5664,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -5739,6 +5874,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -5944,6 +6084,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 5
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 3
             }
           }
         },
@@ -6149,6 +6294,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -6354,6 +6504,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -6559,6 +6714,11 @@
               "tp": 3,
               "fp": 4,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 2
             }
           }
         },
@@ -6764,6 +6924,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -6969,6 +7134,11 @@
               "tp": 1,
               "fp": 3,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -7169,6 +7339,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -7374,6 +7549,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 0
             }
           }
         },
@@ -7579,6 +7759,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -7784,6 +7969,11 @@
               "tp": 1,
               "fp": 5,
               "fn": 5
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 4
             }
           }
         },
@@ -7989,6 +8179,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -8194,6 +8389,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -8399,6 +8599,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -8604,6 +8809,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -8809,6 +9019,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 0,
+              "fn": 2
             }
           }
         },
@@ -9014,6 +9229,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -9219,6 +9439,11 @@
               "tp": 4,
               "fp": 7,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 4,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -9424,6 +9649,11 @@
               "tp": 1,
               "fp": 2,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -9629,6 +9859,11 @@
               "tp": 2,
               "fp": 4,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 2
             }
           }
         },
@@ -9834,6 +10069,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 4,
+              "fn": 4
             }
           }
         },
@@ -10039,6 +10279,11 @@
               "tp": 2,
               "fp": 3,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 3
             }
           }
         },
@@ -10244,6 +10489,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         }
@@ -10286,6 +10536,7 @@
         "qodo-v22",
         "sentry",
         "sourcery",
+        "tusk",
         "vercel"
       ],
       "dimensions": {
@@ -10355,6 +10606,15 @@
         ]
       },
       "overall_metrics": {
+        "qodo": {
+          "precision": 31.8,
+          "recall": 44.5,
+          "f1": 37.1,
+          "tp": 61,
+          "fp": 131,
+          "fn": 76,
+          "num_prs": 50
+        },
         "greptile": {
           "precision": 41.5,
           "recall": 39.4,
@@ -10364,112 +10624,103 @@
           "fn": 83,
           "num_prs": 50
         },
-        "qodo-v2-2": {
-          "precision": 47.1,
-          "recall": 41.6,
-          "f1": 44.2,
-          "tp": 57,
-          "fp": 64,
-          "fn": 80,
-          "num_prs": 50
-        },
-        "qodo-v2": {
-          "precision": 42.9,
-          "recall": 55.5,
-          "f1": 48.4,
-          "tp": 76,
-          "fp": 101,
-          "fn": 61,
+        "sentry": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0,
+          "tp": 0,
+          "fp": 1,
+          "fn": 137,
           "num_prs": 50
         },
-        "claude-code": {
-          "precision": 34.8,
-          "recall": 40.9,
-          "f1": 37.6,
-          "tp": 56,
-          "fp": 105,
-          "fn": 81,
+        "kodus": {
+          "precision": 22.9,
+          "recall": 44.5,
+          "f1": 30.3,
+          "tp": 61,
+          "fp": 205,
+          "fn": 76,
           "num_prs": 50
         },
-        "cubic-v2": {
-          "precision": 56.3,
-          "recall": 68.6,
-          "f1": 61.8,
-          "tp": 94,
-          "fp": 73,
-          "fn": 43,
+        "copilot": {
+          "precision": 28.3,
+          "recall": 53.3,
+          "f1": 37.0,
+          "tp": 73,
+          "fp": 185,
+          "fn": 64,
           "num_prs": 50
         },
-        "augment": {
-          "precision": 47.5,
-          "recall": 61.3,
-          "f1": 53.5,
-          "tp": 84,
-          "fp": 93,
-          "fn": 53,
+        "macroscope": {
+          "precision": 48.4,
+          "recall": 43.8,
+          "f1": 46.0,
+          "tp": 60,
+          "fp": 64,
+          "fn": 77,
           "num_prs": 50
         },
-        "qodo-extended-summary": {
-          "precision": 40.2,
-          "recall": 67.2,
-          "f1": 50.3,
-          "tp": 92,
-          "fp": 137,
-          "fn": 45,
+        "graphite": {
+          "precision": 100.0,
+          "recall": 8.8,
+          "f1": 16.1,
+          "tp": 12,
+          "fp": 0,
+          "fn": 125,
           "num_prs": 50
         },
-        "claude": {
-          "precision": 34.8,
-          "recall": 35.8,
-          "f1": 35.3,
-          "tp": 49,
-          "fp": 92,
-          "fn": 88,
+        "entelligence": {
+          "precision": 30.1,
+          "recall": 20.4,
+          "f1": 24.3,
+          "tp": 28,
+          "fp": 65,
+          "fn": 109,
           "num_prs": 50
         },
-        "mra-b": {
-          "precision": 25.0,
-          "recall": 41.6,
-          "f1": 31.2,
-          "tp": 57,
-          "fp": 171,
-          "fn": 80,
+        "sourcery": {
+          "precision": 33.3,
+          "recall": 51.8,
+          "f1": 40.6,
+          "tp": 71,
+          "fp": 142,
+          "fn": 66,
           "num_prs": 50
         },
-        "greptile-v4": {
-          "precision": 33.1,
+        "tusk": {
+          "precision": 40.2,
           "recall": 56.9,
-          "f1": 41.8,
+          "f1": 47.1,
           "tp": 78,
-          "fp": 158,
+          "fp": 116,
           "fn": 59,
           "num_prs": 50
         },
-        "kodus-v2": {
-          "precision": 46.7,
+        "claude": {
+          "precision": 34.8,
           "recall": 35.8,
-          "f1": 40.5,
+          "f1": 35.3,
           "tp": 49,
-          "fp": 56,
+          "fp": 92,
           "fn": 88,
           "num_prs": 50
         },
-        "mra-max": {
-          "precision": 8.9,
-          "recall": 34.3,
-          "f1": 14.1,
-          "tp": 47,
-          "fp": 481,
-          "fn": 90,
+        "baz": {
+          "precision": 48.8,
+          "recall": 29.2,
+          "f1": 36.5,
+          "tp": 40,
+          "fp": 42,
+          "fn": 97,
           "num_prs": 50
         },
-        "gemini": {
-          "precision": 31.1,
-          "recall": 37.2,
-          "f1": 33.9,
-          "tp": 51,
-          "fp": 113,
-          "fn": 86,
+        "qodo-v22": {
+          "precision": 44.6,
+          "recall": 54.7,
+          "f1": 49.2,
+          "tp": 75,
+          "fp": 93,
+          "fn": 62,
           "num_prs": 50
         },
         "codeant": {
@@ -10481,40 +10732,13 @@
           "fn": 70,
           "num_prs": 50
         },
-        "bugbot": {
-          "precision": 47.2,
-          "recall": 43.8,
-          "f1": 45.5,
-          "tp": 60,
-          "fp": 67,
-          "fn": 77,
-          "num_prs": 50
-        },
-        "mra-a": {
-          "precision": 23.6,
-          "recall": 45.2,
-          "f1": 31.0,
-          "tp": 61,
-          "fp": 198,
-          "fn": 74,
-          "num_prs": 49
-        },
-        "sourcery": {
-          "precision": 33.3,
-          "recall": 51.8,
-          "f1": 40.6,
-          "tp": 71,
-          "fp": 142,
-          "fn": 66,
-          "num_prs": 50
-        },
-        "bito": {
-          "precision": 0,
-          "recall": 0.0,
-          "f1": 0,
-          "tp": 0,
-          "fp": 0,
-          "fn": 137,
+        "linearb": {
+          "precision": 8.3,
+          "recall": 2.9,
+          "f1": 4.3,
+          "tp": 4,
+          "fp": 44,
+          "fn": 133,
           "num_prs": 50
         },
         "vercel": {
@@ -10526,22 +10750,13 @@
           "fn": 137,
           "num_prs": 50
         },
-        "copilot": {
-          "precision": 28.3,
-          "recall": 53.3,
-          "f1": 37.0,
-          "tp": 73,
-          "fp": 185,
-          "fn": 64,
-          "num_prs": 50
-        },
-        "graphite": {
-          "precision": 100.0,
-          "recall": 8.8,
-          "f1": 16.1,
-          "tp": 12,
-          "fp": 0,
-          "fn": 125,
+        "coderabbit": {
+          "precision": 24.7,
+          "recall": 39.4,
+          "f1": 30.3,
+          "tp": 54,
+          "fp": 165,
+          "fn": 83,
           "num_prs": 50
         },
         "mesa": {
@@ -10553,67 +10768,40 @@
           "fn": 75,
           "num_prs": 50
         },
-        "kodus": {
-          "precision": 22.9,
-          "recall": 44.5,
-          "f1": 30.3,
-          "tp": 61,
-          "fp": 205,
-          "fn": 76,
-          "num_prs": 50
-        },
-        "kg": {
-          "precision": 50.0,
-          "recall": 16.8,
-          "f1": 25.1,
-          "tp": 23,
-          "fp": 23,
-          "fn": 114,
-          "num_prs": 50
-        },
-        "devin": {
-          "precision": 54.3,
+        "gemini": {
+          "precision": 31.1,
           "recall": 37.2,
-          "f1": 44.2,
+          "f1": 33.9,
           "tp": 51,
-          "fp": 43,
+          "fp": 113,
           "fn": 86,
           "num_prs": 50
         },
-        "baz": {
-          "precision": 48.8,
-          "recall": 29.2,
-          "f1": 36.5,
-          "tp": 40,
-          "fp": 42,
-          "fn": 97,
-          "num_prs": 50
-        },
-        "sentry": {
-          "precision": 0.0,
-          "recall": 0.0,
-          "f1": 0,
-          "tp": 0,
-          "fp": 1,
-          "fn": 137,
+        "qodo-extended-summary": {
+          "precision": 40.2,
+          "recall": 67.2,
+          "f1": 50.3,
+          "tp": 92,
+          "fp": 137,
+          "fn": 45,
           "num_prs": 50
         },
-        "macroscope": {
-          "precision": 48.4,
-          "recall": 43.8,
-          "f1": 46.0,
-          "tp": 60,
-          "fp": 64,
-          "fn": 77,
+        "cubic-dev": {
+          "precision": 30.1,
+          "recall": 75.9,
+          "f1": 43.2,
+          "tp": 104,
+          "fp": 241,
+          "fn": 33,
           "num_prs": 50
         },
-        "mra-nano": {
-          "precision": 12.6,
-          "recall": 17.5,
-          "f1": 14.7,
-          "tp": 24,
-          "fp": 166,
-          "fn": 113,
+        "mra-ultra": {
+          "precision": 18.3,
+          "recall": 37.2,
+          "f1": 24.6,
+          "tp": 51,
+          "fp": 227,
+          "fn": 86,
           "num_prs": 50
         },
         "qodo-extended": {
@@ -10625,76 +10813,148 @@
           "fn": 51,
           "num_prs": 50
         },
-        "cubic-dev": {
-          "precision": 30.1,
-          "recall": 75.9,
-          "f1": 43.2,
-          "tp": 104,
-          "fp": 241,
-          "fn": 33,
-          "num_prs": 50
-        },
-        "propel": {
-          "precision": 52.5,
-          "recall": 38.7,
+        "augment": {
+          "precision": 47.5,
+          "recall": 61.3,
+          "f1": 53.5,
+          "tp": 84,
+          "fp": 93,
+          "fn": 53,
+          "num_prs": 50
+        },
+        "mra-max": {
+          "precision": 8.9,
+          "recall": 34.3,
+          "f1": 14.1,
+          "tp": 47,
+          "fp": 481,
+          "fn": 90,
+          "num_prs": 50
+        },
+        "kodus-v2": {
+          "precision": 46.7,
+          "recall": 35.8,
+          "f1": 40.5,
+          "tp": 49,
+          "fp": 56,
+          "fn": 88,
+          "num_prs": 50
+        },
+        "kg": {
+          "precision": 50.0,
+          "recall": 16.8,
+          "f1": 25.1,
+          "tp": 23,
+          "fp": 23,
+          "fn": 114,
+          "num_prs": 50
+        },
+        "cubic-v2": {
+          "precision": 56.3,
+          "recall": 68.6,
+          "f1": 61.8,
+          "tp": 94,
+          "fp": 73,
+          "fn": 43,
+          "num_prs": 50
+        },
+        "devin": {
+          "precision": 54.3,
+          "recall": 37.2,
+          "f1": 44.2,
+          "tp": 51,
+          "fp": 43,
+          "fn": 86,
+          "num_prs": 50
+        },
+        "mra-b": {
+          "precision": 25.0,
+          "recall": 41.6,
+          "f1": 31.2,
+          "tp": 57,
+          "fp": 171,
+          "fn": 80,
+          "num_prs": 50
+        },
+        "propel": {
+          "precision": 52.5,
+          "recall": 38.7,
           "f1": 44.5,
           "tp": 53,
           "fp": 48,
           "fn": 84,
           "num_prs": 50
         },
-        "qodo-v22": {
-          "precision": 44.6,
-          "recall": 54.7,
-          "f1": 49.2,
-          "tp": 75,
-          "fp": 93,
-          "fn": 62,
+        "mra-a": {
+          "precision": 23.6,
+          "recall": 45.2,
+          "f1": 31.0,
+          "tp": 61,
+          "fp": 198,
+          "fn": 74,
+          "num_prs": 49
+        },
+        "claude-code": {
+          "precision": 34.8,
+          "recall": 40.9,
+          "f1": 37.6,
+          "tp": 56,
+          "fp": 105,
+          "fn": 81,
           "num_prs": 50
         },
-        "linearb": {
-          "precision": 8.3,
-          "recall": 2.9,
-          "f1": 4.3,
-          "tp": 4,
-          "fp": 44,
-          "fn": 133,
+        "qodo-v2-2": {
+          "precision": 47.1,
+          "recall": 41.6,
+          "f1": 44.2,
+          "tp": 57,
+          "fp": 64,
+          "fn": 80,
           "num_prs": 50
         },
-        "qodo": {
-          "precision": 31.8,
-          "recall": 44.5,
-          "f1": 37.1,
-          "tp": 61,
-          "fp": 131,
-          "fn": 76,
+        "greptile-v4": {
+          "precision": 33.1,
+          "recall": 56.9,
+          "f1": 41.8,
+          "tp": 78,
+          "fp": 158,
+          "fn": 59,
           "num_prs": 50
         },
-        "mra-ultra": {
-          "precision": 18.3,
-          "recall": 37.2,
-          "f1": 24.6,
-          "tp": 51,
-          "fp": 227,
-          "fn": 86,
+        "bito": {
+          "precision": 0,
+          "recall": 0.0,
+          "f1": 0,
+          "tp": 0,
+          "fp": 0,
+          "fn": 137,
           "num_prs": 50
         },
-        "entelligence": {
-          "precision": 30.1,
-          "recall": 20.4,
-          "f1": 24.3,
-          "tp": 28,
-          "fp": 65,
-          "fn": 109,
+        "mra-nano": {
+          "precision": 12.6,
+          "recall": 17.5,
+          "f1": 14.7,
+          "tp": 24,
+          "fp": 166,
+          "fn": 113,
           "num_prs": 50
         },
-        "coderabbit": {
-          "precision": 24.7,
-          "recall": 39.4,
-          "f1": 30.3,
-          "tp": 54,
-          "fp": 165,
-          "fn": 83,
+        "bugbot": {
+          "precision": 47.2,
+          "recall": 43.8,
+          "f1": 45.5,
+          "tp": 60,
+          "fp": 67,
+          "fn": 77,
+          "num_prs": 50
+        },
+        "qodo-v2": {
+          "precision": 42.9,
+          "recall": 55.5,
+          "f1": 48.4,
+          "tp": 76,
+          "fp": 101,
+          "fn": 61,
           "num_prs": 50
         }
       }
@@ -10903,6 +11163,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -11108,6 +11373,11 @@
               "tp": 2,
               "fp": 2,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -11313,6 +11583,11 @@
               "tp": 0,
               "fp": 3,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -11518,6 +11793,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -11723,6 +12003,11 @@
               "tp": 2,
               "fp": 4,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 2
             }
           }
         },
@@ -11928,6 +12213,11 @@
               "tp": 2,
               "fp": 2,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 0
             }
           }
         },
@@ -12133,6 +12423,11 @@
               "tp": 1,
               "fp": 5,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -12338,6 +12633,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -12543,6 +12843,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -12748,6 +13053,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 3
             }
           }
         },
@@ -12953,6 +13263,11 @@
               "tp": 0,
               "fp": 5,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -13158,6 +13473,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -13363,6 +13683,11 @@
               "tp": 1,
               "fp": 2,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -13568,6 +13893,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -13773,6 +14103,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 0,
+              "fn": 1
             }
           }
         },
@@ -13978,6 +14313,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -14183,6 +14523,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 8,
+              "fn": 2
             }
           }
         },
@@ -14388,6 +14733,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 0
             }
           }
         },
@@ -14593,6 +14943,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 2
             }
           }
         },
@@ -14798,6 +15153,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -15003,6 +15363,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -15208,6 +15573,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -15413,6 +15783,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -15618,6 +15993,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 0,
+              "fn": 1
             }
           }
         },
@@ -15823,6 +16203,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -16028,6 +16413,11 @@
               "tp": 3,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -16233,6 +16623,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -16438,6 +16833,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -16643,6 +17043,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 5
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 3
             }
           }
         },
@@ -16848,6 +17253,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -17053,6 +17463,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -17258,6 +17673,11 @@
               "tp": 3,
               "fp": 5,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 2
             }
           }
         },
@@ -17463,6 +17883,11 @@
               "tp": 3,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -17668,6 +18093,11 @@
               "tp": 1,
               "fp": 3,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -17868,6 +18298,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -18073,6 +18508,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -18278,6 +18718,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -18483,6 +18928,11 @@
               "tp": 1,
               "fp": 5,
               "fn": 5
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 4
             }
           }
         },
@@ -18688,6 +19138,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -18893,6 +19348,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -19098,6 +19558,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -19303,6 +19768,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -19508,6 +19978,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 0,
+              "fn": 2
             }
           }
         },
@@ -19713,6 +20188,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -19918,6 +20398,11 @@
               "tp": 4,
               "fp": 6,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 4,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -20123,6 +20608,11 @@
               "tp": 1,
               "fp": 2,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -20328,6 +20818,11 @@
               "tp": 2,
               "fp": 4,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 2
             }
           }
         },
@@ -20533,6 +21028,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 4,
+              "fn": 4
             }
           }
         },
@@ -20738,6 +21238,11 @@
               "tp": 2,
               "fp": 3,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 3
             }
           }
         },
@@ -20943,6 +21448,11 @@
               "tp": 1,
               "fp": 4,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         }
@@ -20985,6 +21495,7 @@
         "qodo-v22",
         "sentry",
         "sourcery",
+        "tusk",
         "vercel"
       ],
       "dimensions": {
@@ -21054,6 +21565,15 @@
         ]
       },
       "overall_metrics": {
+        "qodo": {
+          "precision": 27.9,
+          "recall": 45.3,
+          "f1": 34.5,
+          "tp": 62,
+          "fp": 160,
+          "fn": 75,
+          "num_prs": 50
+        },
         "greptile": {
           "precision": 43.7,
           "recall": 40.1,
@@ -21063,58 +21583,76 @@
           "fn": 82,
           "num_prs": 50
         },
-        "qodo-v2-2": {
-          "precision": 43.1,
-          "recall": 40.9,
-          "f1": 41.9,
-          "tp": 56,
-          "fp": 74,
-          "fn": 81,
+        "sentry": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0,
+          "tp": 0,
+          "fp": 1,
+          "fn": 137,
           "num_prs": 50
         },
-        "qodo-v2": {
-          "precision": 40.5,
-          "recall": 56.2,
-          "f1": 47.1,
-          "tp": 77,
-          "fp": 113,
-          "fn": 60,
+        "kodus": {
+          "precision": 21.9,
+          "recall": 44.5,
+          "f1": 29.3,
+          "tp": 61,
+          "fp": 218,
+          "fn": 76,
           "num_prs": 50
         },
-        "claude-code": {
-          "precision": 30.7,
-          "recall": 40.1,
-          "f1": 34.8,
-          "tp": 55,
-          "fp": 124,
-          "fn": 82,
+        "copilot": {
+          "precision": 27.2,
+          "recall": 51.1,
+          "f1": 35.5,
+          "tp": 70,
+          "fp": 187,
+          "fn": 67,
           "num_prs": 50
         },
-        "cubic-v2": {
-          "precision": 55.6,
-          "recall": 68.6,
-          "f1": 61.4,
-          "tp": 94,
-          "fp": 75,
-          "fn": 43,
+        "macroscope": {
+          "precision": 45.8,
+          "recall": 43.8,
+          "f1": 44.8,
+          "tp": 60,
+          "fp": 71,
+          "fn": 77,
           "num_prs": 50
         },
-        "augment": {
-          "precision": 46.0,
-          "recall": 63.5,
-          "f1": 53.4,
-          "tp": 87,
-          "fp": 102,
-          "fn": 50,
+        "graphite": {
+          "precision": 100.0,
+          "recall": 8.8,
+          "f1": 16.1,
+          "tp": 12,
+          "fp": 0,
+          "fn": 125,
           "num_prs": 50
         },
-        "qodo-extended-summary": {
-          "precision": 40.0,
-          "recall": 67.2,
-          "f1": 50.1,
-          "tp": 92,
-          "fp": 138,
-          "fn": 45,
+        "entelligence": {
+          "precision": 29.3,
+          "recall": 21.2,
+          "f1": 24.6,
+          "tp": 29,
+          "fp": 70,
+          "fn": 108,
+          "num_prs": 50
+        },
+        "sourcery": {
+          "precision": 33.3,
+          "recall": 51.8,
+          "f1": 40.6,
+          "tp": 71,
+          "fp": 142,
+          "fn": 66,
+          "num_prs": 50
+        },
+        "tusk": {
+          "precision": 37.6,
+          "recall": 56.2,
+          "f1": 45.0,
+          "tp": 77,
+          "fp": 128,
+          "fn": 60,
           "num_prs": 50
         },
         "claude": {
@@ -21126,49 +21664,22 @@
           "fn": 82,
           "num_prs": 50
         },
-        "mra-b": {
-          "precision": 22.0,
-          "recall": 40.9,
-          "f1": 28.6,
-          "tp": 56,
-          "fp": 198,
-          "fn": 81,
-          "num_prs": 50
-        },
-        "greptile-v4": {
-          "precision": 31.2,
-          "recall": 59.1,
-          "f1": 40.8,
-          "tp": 81,
-          "fp": 179,
-          "fn": 56,
-          "num_prs": 50
-        },
-        "kodus-v2": {
-          "precision": 44.9,
-          "recall": 35.0,
-          "f1": 39.3,
-          "tp": 48,
-          "fp": 59,
-          "fn": 89,
-          "num_prs": 50
-        },
-        "mra-max": {
-          "precision": 9.5,
-          "recall": 35.8,
-          "f1": 15.1,
-          "tp": 49,
-          "fp": 465,
-          "fn": 88,
+        "baz": {
+          "precision": 48.2,
+          "recall": 29.9,
+          "f1": 36.9,
+          "tp": 41,
+          "fp": 44,
+          "fn": 96,
           "num_prs": 50
         },
-        "gemini": {
-          "precision": 29.7,
-          "recall": 35.8,
-          "f1": 32.5,
-          "tp": 49,
-          "fp": 116,
-          "fn": 88,
+        "qodo-v22": {
+          "precision": 42.1,
+          "recall": 52.6,
+          "f1": 46.8,
+          "tp": 72,
+          "fp": 99,
+          "fn": 65,
           "num_prs": 50
         },
         "codeant": {
@@ -21180,40 +21691,13 @@
           "fn": 68,
           "num_prs": 50
         },
-        "bugbot": {
-          "precision": 45.4,
-          "recall": 43.1,
-          "f1": 44.2,
-          "tp": 59,
-          "fp": 71,
-          "fn": 78,
-          "num_prs": 50
-        },
-        "mra-a": {
-          "precision": 18.9,
-          "recall": 42.2,
-          "f1": 26.1,
-          "tp": 57,
-          "fp": 245,
-          "fn": 78,
-          "num_prs": 49
-        },
-        "sourcery": {
-          "precision": 33.3,
-          "recall": 51.8,
-          "f1": 40.6,
-          "tp": 71,
-          "fp": 142,
-          "fn": 66,
-          "num_prs": 50
-        },
-        "bito": {
-          "precision": 0,
-          "recall": 0.0,
-          "f1": 0,
-          "tp": 0,
-          "fp": 0,
-          "fn": 137,
+        "linearb": {
+          "precision": 8.0,
+          "recall": 2.9,
+          "f1": 4.3,
+          "tp": 4,
+          "fp": 46,
+          "fn": 133,
           "num_prs": 50
         },
         "vercel": {
@@ -21225,22 +21709,13 @@
           "fn": 137,
           "num_prs": 50
         },
-        "copilot": {
-          "precision": 27.2,
-          "recall": 51.1,
-          "f1": 35.5,
-          "tp": 70,
-          "fp": 187,
-          "fn": 67,
-          "num_prs": 50
-        },
-        "graphite": {
-          "precision": 100.0,
-          "recall": 8.8,
-          "f1": 16.1,
-          "tp": 12,
-          "fp": 0,
-          "fn": 125,
+        "coderabbit": {
+          "precision": 23.7,
+          "recall": 38.7,
+          "f1": 29.4,
+          "tp": 53,
+          "fp": 171,
+          "fn": 84,
           "num_prs": 50
         },
         "mesa": {
@@ -21252,13 +21727,76 @@
           "fn": 73,
           "num_prs": 50
         },
-        "kodus": {
-          "precision": 21.9,
-          "recall": 44.5,
-          "f1": 29.3,
-          "tp": 61,
-          "fp": 218,
-          "fn": 76,
+        "gemini": {
+          "precision": 29.7,
+          "recall": 35.8,
+          "f1": 32.5,
+          "tp": 49,
+          "fp": 116,
+          "fn": 88,
+          "num_prs": 50
+        },
+        "qodo-extended-summary": {
+          "precision": 40.0,
+          "recall": 67.2,
+          "f1": 50.1,
+          "tp": 92,
+          "fp": 138,
+          "fn": 45,
+          "num_prs": 50
+        },
+        "cubic-dev": {
+          "precision": 29.2,
+          "recall": 76.6,
+          "f1": 42.3,
+          "tp": 105,
+          "fp": 255,
+          "fn": 32,
+          "num_prs": 50
+        },
+        "mra-ultra": {
+          "precision": 17.1,
+          "recall": 38.7,
+          "f1": 23.7,
+          "tp": 53,
+          "fp": 257,
+          "fn": 84,
+          "num_prs": 50
+        },
+        "qodo-extended": {
+          "precision": 35.3,
+          "recall": 61.3,
+          "f1": 44.8,
+          "tp": 84,
+          "fp": 154,
+          "fn": 53,
+          "num_prs": 50
+        },
+        "augment": {
+          "precision": 46.0,
+          "recall": 63.5,
+          "f1": 53.4,
+          "tp": 87,
+          "fp": 102,
+          "fn": 50,
+          "num_prs": 50
+        },
+        "mra-max": {
+          "precision": 9.5,
+          "recall": 35.8,
+          "f1": 15.1,
+          "tp": 49,
+          "fp": 465,
+          "fn": 88,
+          "num_prs": 50
+        },
+        "kodus-v2": {
+          "precision": 44.9,
+          "recall": 35.0,
+          "f1": 39.3,
+          "tp": 48,
+          "fp": 59,
+          "fn": 89,
           "num_prs": 50
         },
         "kg": {
@@ -21270,6 +21808,15 @@
           "fn": 116,
           "num_prs": 50
         },
+        "cubic-v2": {
+          "precision": 55.6,
+          "recall": 68.6,
+          "f1": 61.4,
+          "tp": 94,
+          "fp": 75,
+          "fn": 43,
+          "num_prs": 50
+        },
         "devin": {
           "precision": 54.2,
           "recall": 38.0,
@@ -21279,33 +21826,69 @@
           "fn": 85,
           "num_prs": 50
         },
-        "baz": {
-          "precision": 48.2,
-          "recall": 29.9,
-          "f1": 36.9,
-          "tp": 41,
-          "fp": 44,
-          "fn": 96,
+        "mra-b": {
+          "precision": 22.0,
+          "recall": 40.9,
+          "f1": 28.6,
+          "tp": 56,
+          "fp": 198,
+          "fn": 81,
           "num_prs": 50
         },
-        "sentry": {
-          "precision": 0.0,
+        "propel": {
+          "precision": 49.5,
+          "recall": 36.5,
+          "f1": 42.0,
+          "tp": 50,
+          "fp": 51,
+          "fn": 87,
+          "num_prs": 50
+        },
+        "mra-a": {
+          "precision": 18.9,
+          "recall": 42.2,
+          "f1": 26.1,
+          "tp": 57,
+          "fp": 245,
+          "fn": 78,
+          "num_prs": 49
+        },
+        "claude-code": {
+          "precision": 30.7,
+          "recall": 40.1,
+          "f1": 34.8,
+          "tp": 55,
+          "fp": 124,
+          "fn": 82,
+          "num_prs": 50
+        },
+        "qodo-v2-2": {
+          "precision": 43.1,
+          "recall": 40.9,
+          "f1": 41.9,
+          "tp": 56,
+          "fp": 74,
+          "fn": 81,
+          "num_prs": 50
+        },
+        "greptile-v4": {
+          "precision": 31.2,
+          "recall": 59.1,
+          "f1": 40.8,
+          "tp": 81,
+          "fp": 179,
+          "fn": 56,
+          "num_prs": 50
+        },
+        "bito": {
+          "precision": 0,
           "recall": 0.0,
           "f1": 0,
           "tp": 0,
-          "fp": 1,
+          "fp": 0,
           "fn": 137,
           "num_prs": 50
         },
-        "macroscope": {
-          "precision": 45.8,
-          "recall": 43.8,
-          "f1": 44.8,
-          "tp": 60,
-          "fp": 71,
-          "fn": 77,
-          "num_prs": 50
-        },
         "mra-nano": {
           "precision": 10.4,
           "recall": 25.5,
@@ -21315,85 +21898,22 @@
           "fn": 102,
           "num_prs": 50
         },
-        "qodo-extended": {
-          "precision": 35.3,
-          "recall": 61.3,
-          "f1": 44.8,
-          "tp": 84,
-          "fp": 154,
-          "fn": 53,
+        "bugbot": {
+          "precision": 45.4,
+          "recall": 43.1,
+          "f1": 44.2,
+          "tp": 59,
+          "fp": 71,
+          "fn": 78,
           "num_prs": 50
         },
-        "cubic-dev": {
-          "precision": 29.2,
-          "recall": 76.6,
-          "f1": 42.3,
-          "tp": 105,
-          "fp": 255,
-          "fn": 32,
-          "num_prs": 50
-        },
-        "propel": {
-          "precision": 49.5,
-          "recall": 36.5,
-          "f1": 42.0,
-          "tp": 50,
-          "fp": 51,
-          "fn": 87,
-          "num_prs": 50
-        },
-        "qodo-v22": {
-          "precision": 42.1,
-          "recall": 52.6,
-          "f1": 46.8,
-          "tp": 72,
-          "fp": 99,
-          "fn": 65,
-          "num_prs": 50
-        },
-        "linearb": {
-          "precision": 8.0,
-          "recall": 2.9,
-          "f1": 4.3,
-          "tp": 4,
-          "fp": 46,
-          "fn": 133,
-          "num_prs": 50
-        },
-        "qodo": {
-          "precision": 27.9,
-          "recall": 45.3,
-          "f1": 34.5,
-          "tp": 62,
-          "fp": 160,
-          "fn": 75,
-          "num_prs": 50
-        },
-        "mra-ultra": {
-          "precision": 17.1,
-          "recall": 38.7,
-          "f1": 23.7,
-          "tp": 53,
-          "fp": 257,
-          "fn": 84,
-          "num_prs": 50
-        },
-        "entelligence": {
-          "precision": 29.3,
-          "recall": 21.2,
-          "f1": 24.6,
-          "tp": 29,
-          "fp": 70,
-          "fn": 108,
-          "num_prs": 50
-        },
-        "coderabbit": {
-          "precision": 23.7,
-          "recall": 38.7,
-          "f1": 29.4,
-          "tp": 53,
-          "fp": 171,
-          "fn": 84,
+        "qodo-v2": {
+          "precision": 40.5,
+          "recall": 56.2,
+          "f1": 47.1,
+          "tp": 77,
+          "fp": 113,
+          "fn": 60,
           "num_prs": 50
         }
       }
@@ -21602,6 +22122,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 2
             }
           }
         },
@@ -21807,6 +22332,11 @@
               "tp": 2,
               "fp": 2,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 0,
+              "fn": 2
             }
           }
         },
@@ -22012,6 +22542,11 @@
               "tp": 0,
               "fp": 4,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -22217,6 +22752,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -22422,6 +22962,11 @@
               "tp": 2,
               "fp": 4,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 2
             }
           }
         },
@@ -22627,6 +23172,11 @@
               "tp": 2,
               "fp": 2,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -22832,6 +23382,11 @@
               "tp": 1,
               "fp": 5,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -23037,6 +23592,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -23242,6 +23802,11 @@
               "tp": 2,
               "fp": 3,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -23447,6 +24012,11 @@
               "tp": 3,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 1,
+              "fn": 3
             }
           }
         },
@@ -23652,6 +24222,11 @@
               "tp": 0,
               "fp": 5,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 2
             }
           }
         },
@@ -23857,6 +24432,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -24062,6 +24642,11 @@
               "tp": 1,
               "fp": 2,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -24267,6 +24852,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -24472,6 +25062,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 0,
+              "fn": 1
             }
           }
         },
@@ -24677,6 +25272,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -24882,6 +25482,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 7,
+              "fn": 2
             }
           }
         },
@@ -25087,6 +25692,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -25292,6 +25902,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 2
             }
           }
         },
@@ -25497,6 +26112,11 @@
               "tp": 0,
               "fp": 3,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -25702,6 +26322,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -25907,6 +26532,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -26112,6 +26742,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -26317,6 +26952,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 0,
+              "fn": 1
             }
           }
         },
@@ -26522,6 +27162,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 0
             }
           }
         },
@@ -26727,6 +27372,11 @@
               "tp": 3,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 0
             }
           }
         },
@@ -26932,6 +27582,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -27137,6 +27792,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -27342,6 +28002,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 5
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 1,
+              "fn": 2
             }
           }
         },
@@ -27547,6 +28212,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -27752,6 +28422,11 @@
               "tp": 0,
               "fp": 1,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 1,
+              "fn": 1
             }
           }
         },
@@ -27957,6 +28632,11 @@
               "tp": 2,
               "fp": 5,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 2
             }
           }
         },
@@ -28162,6 +28842,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -28367,6 +29052,11 @@
               "tp": 1,
               "fp": 3,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -28567,6 +29257,11 @@
               "tp": 2,
               "fp": 0,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -28772,6 +29467,11 @@
               "tp": 0,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -28977,6 +29677,11 @@
               "tp": 0,
               "fp": 2,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 3,
+              "fn": 1
             }
           }
         },
@@ -29182,6 +29887,11 @@
               "tp": 1,
               "fp": 5,
               "fn": 5
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 4
             }
           }
         },
@@ -29387,6 +30097,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 2
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -29592,6 +30307,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -29797,6 +30517,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 5,
+              "fn": 1
             }
           }
         },
@@ -30002,6 +30727,11 @@
               "tp": 2,
               "fp": 1,
               "fn": 0
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 2,
+              "fn": 0
             }
           }
         },
@@ -30207,6 +30937,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 0,
+              "fn": 2
             }
           }
         },
@@ -30412,6 +31147,11 @@
               "tp": 1,
               "fp": 0,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 4,
+              "fn": 1
             }
           }
         },
@@ -30617,6 +31357,11 @@
               "tp": 4,
               "fp": 6,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 4,
+              "fp": 2,
+              "fn": 1
             }
           }
         },
@@ -30822,6 +31567,11 @@
               "tp": 1,
               "fp": 2,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 2,
+              "fp": 3,
+              "fn": 0
             }
           }
         },
@@ -31027,6 +31777,11 @@
               "tp": 2,
               "fp": 4,
               "fn": 3
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 2
             }
           }
         },
@@ -31232,6 +31987,11 @@
               "tp": 0,
               "fp": 3,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 0,
+              "fp": 4,
+              "fn": 4
             }
           }
         },
@@ -31437,6 +32197,11 @@
               "tp": 1,
               "fp": 4,
               "fn": 4
+            },
+            "tusk": {
+              "tp": 3,
+              "fp": 2,
+              "fn": 2
             }
           }
         },
@@ -31642,6 +32407,11 @@
               "tp": 1,
               "fp": 1,
               "fn": 1
+            },
+            "tusk": {
+              "tp": 1,
+              "fp": 9,
+              "fn": 1
             }
           }
         }
@@ -31684,6 +32454,7 @@
         "qodo-v22",
         "sentry",
         "sourcery",
+        "tusk",
         "vercel"
       ],
       "dimensions": {
@@ -31753,6 +32524,15 @@
         ]
       },
       "overall_metrics": {
+        "qodo": {
+          "precision": 25.6,
+          "recall": 44.5,
+          "f1": 32.5,
+          "tp": 61,
+          "fp": 177,
+          "fn": 76,
+          "num_prs": 50
+        },
         "greptile": {
           "precision": 37.4,
           "recall": 38.0,
@@ -31762,58 +32542,76 @@
           "fn": 85,
           "num_prs": 50
         },
-        "qodo-v2-2": {
-          "precision": 41.5,
+        "sentry": {
+          "precision": 0.0,
+          "recall": 0.0,
+          "f1": 0,
+          "tp": 0,
+          "fp": 1,
+          "fn": 137,
+          "num_prs": 50
+        },
+        "kodus": {
+          "precision": 21.0,
           "recall": 43.1,
-          "f1": 42.3,
+          "f1": 28.2,
           "tp": 59,
-          "fp": 83,
+          "fp": 222,
           "fn": 78,
           "num_prs": 50
         },
-        "qodo-v2": {
-          "precision": 36.2,
-          "recall": 56.2,
-          "f1": 44.0,
-          "tp": 77,
-          "fp": 136,
-          "fn": 60,
+        "copilot": {
+          "precision": 24.7,
+          "recall": 52.6,
+          "f1": 33.6,
+          "tp": 72,
+          "fp": 220,
+          "fn": 65,
           "num_prs": 50
         },
-        "claude-code": {
-          "precision": 27.4,
-          "recall": 41.6,
-          "f1": 33.0,
-          "tp": 57,
-          "fp": 151,
-          "fn": 80,
+        "macroscope": {
+          "precision": 41.8,
+          "recall": 40.9,
+          "f1": 41.3,
+          "tp": 56,
+          "fp": 78,
+          "fn": 81,
           "num_prs": 50
         },
-        "cubic-v2": {
-          "precision": 53.6,
-          "recall": 65.7,
-          "f1": 59.0,
-          "tp": 90,
-          "fp": 78,
-          "fn": 47,
+        "graphite": {
+          "precision": 80.0,
+          "recall": 8.8,
+          "f1": 15.8,
+          "tp": 12,
+          "fp": 3,
+          "fn": 125,
           "num_prs": 50
         },
-        "augment": {
-          "precision": 41.9,
-          "recall": 60.6,
-          "f1": 49.6,
-          "tp": 83,
-          "fp": 115,
-          "fn": 54,
+        "entelligence": {
+          "precision": 27.7,
+          "recall": 20.4,
+          "f1": 23.5,
+          "tp": 28,
+          "fp": 73,
+          "fn": 109,
           "num_prs": 50
         },
-        "qodo-extended-summary": {
-          "precision": 38.1,
-          "recall": 66.4,
-          "f1": 48.4,
-          "tp": 91,
-          "fp": 148,
-          "fn": 46,
+        "sourcery": {
+          "precision": 29.8,
+          "recall": 53.3,
+          "f1": 38.2,
+          "tp": 73,
+          "fp": 172,
+          "fn": 64,
+          "num_prs": 50
+        },
+        "tusk": {
+          "precision": 35.5,
+          "recall": 56.2,
+          "f1": 43.5,
+          "tp": 77,
+          "fp": 140,
+          "fn": 60,
           "num_prs": 50
         },
         "claude": {
@@ -31825,49 +32623,22 @@
           "fn": 86,
           "num_prs": 50
         },
-        "mra-b": {
-          "precision": 19.9,
-          "recall": 39.4,
-          "f1": 26.5,
-          "tp": 54,
-          "fp": 217,
-          "fn": 83,
-          "num_prs": 50
-        },
-        "greptile-v4": {
-          "precision": 29.3,
-          "recall": 58.4,
-          "f1": 39.0,
-          "tp": 80,
-          "fp": 193,
-          "fn": 57,
-          "num_prs": 50
-        },
-        "kodus-v2": {
-          "precision": 35.9,
-          "recall": 34.3,
-          "f1": 35.1,
-          "tp": 47,
-          "fp": 84,
-          "fn": 90,
-          "num_prs": 50
-        },
-        "mra-max": {
-          "precision": 7.7,
-          "recall": 31.4,
-          "f1": 12.4,
-          "tp": 43,
-          "fp": 515,
-          "fn": 94,
+        "baz": {
+          "precision": 37.8,
+          "recall": 27.0,
+          "f1": 31.5,
+          "tp": 37,
+          "fp": 61,
+          "fn": 100,
           "num_prs": 50
         },
-        "gemini": {
-          "precision": 26.3,
-          "recall": 33.6,
-          "f1": 29.5,
-          "tp": 46,
-          "fp": 129,
-          "fn": 91,
+        "qodo-v22": {
+          "precision": 38.4,
+          "recall": 53.3,
+          "f1": 44.6,
+          "tp": 73,
+          "fp": 117,
+          "fn": 64,
           "num_prs": 50
         },
         "codeant": {
@@ -31879,40 +32650,13 @@
           "fn": 69,
           "num_prs": 50
         },
-        "bugbot": {
-          "precision": 43.2,
-          "recall": 43.8,
-          "f1": 43.5,
-          "tp": 60,
-          "fp": 79,
-          "fn": 77,
-          "num_prs": 50
-        },
-        "mra-a": {
-          "precision": 17.6,
-          "recall": 40.0,
-          "f1": 24.4,
-          "tp": 54,
-          "fp": 253,
-          "fn": 81,
-          "num_prs": 49
-        },
-        "sourcery": {
-          "precision": 29.8,
-          "recall": 53.3,
-          "f1": 38.2,
-          "tp": 73,
-          "fp": 172,
-          "fn": 64,
-          "num_prs": 50
-        },
-        "bito": {
-          "precision": 0,
-          "recall": 0.0,
-          "f1": 0,
-          "tp": 0,
-          "fp": 0,
-          "fn": 137,
+        "linearb": {
+          "precision": 8.0,
+          "recall": 2.9,
+          "f1": 4.3,
+          "tp": 4,
+          "fp": 46,
+          "fn": 133,
           "num_prs": 50
         },
         "vercel": {
@@ -31924,22 +32668,13 @@
           "fn": 137,
           "num_prs": 50
         },
-        "copilot": {
-          "precision": 24.7,
-          "recall": 52.6,
-          "f1": 33.6,
-          "tp": 72,
-          "fp": 220,
-          "fn": 65,
-          "num_prs": 50
-        },
-        "graphite": {
-          "precision": 80.0,
-          "recall": 8.8,
-          "f1": 15.8,
-          "tp": 12,
-          "fp": 3,
-          "fn": 125,
+        "coderabbit": {
+          "precision": 21.3,
+          "recall": 40.9,
+          "f1": 28.0,
+          "tp": 56,
+          "fp": 207,
+          "fn": 81,
           "num_prs": 50
         },
         "mesa": {
@@ -31951,85 +32686,112 @@
           "fn": 79,
           "num_prs": 50
         },
-        "kodus": {
-          "precision": 21.0,
-          "recall": 43.1,
-          "f1": 28.2,
-          "tp": 59,
-          "fp": 222,
-          "fn": 78,
+        "gemini": {
+          "precision": 26.3,
+          "recall": 33.6,
+          "f1": 29.5,
+          "tp": 46,
+          "fp": 129,
+          "fn": 91,
           "num_prs": 50
         },
-        "kg": {
-          "precision": 51.1,
-          "recall": 16.8,
-          "f1": 25.3,
-          "tp": 23,
-          "fp": 22,
-          "fn": 114,
+        "qodo-extended-summary": {
+          "precision": 38.1,
+          "recall": 66.4,
+          "f1": 48.4,
+          "tp": 91,
+          "fp": 148,
+          "fn": 46,
           "num_prs": 50
         },
-        "devin": {
-          "precision": 46.4,
+        "cubic-dev": {
+          "precision": 27.3,
+          "recall": 73.0,
+          "f1": 39.8,
+          "tp": 100,
+          "fp": 266,
+          "fn": 37,
+          "num_prs": 50
+        },
+        "mra-ultra": {
+          "precision": 15.3,
           "recall": 37.2,
-          "f1": 41.3,
+          "f1": 21.7,
           "tp": 51,
-          "fp": 59,
+          "fp": 282,
           "fn": 86,
           "num_prs": 50
         },
-        "baz": {
-          "precision": 37.8,
-          "recall": 27.0,
-          "f1": 31.5,
-          "tp": 37,
-          "fp": 61,
-          "fn": 100,
+        "qodo-extended": {
+          "precision": 31.4,
+          "recall": 59.9,
+          "f1": 41.2,
+          "tp": 82,
+          "fp": 179,
+          "fn": 55,
           "num_prs": 50
         },
-        "sentry": {
-          "precision": 0.0,
-          "recall": 0.0,
-          "f1": 0,
-          "tp": 0,
-          "fp": 1,
-          "fn": 137,
+        "augment": {
+          "precision": 41.9,
+          "recall": 60.6,
+          "f1": 49.6,
+          "tp": 83,
+          "fp": 115,
+          "fn": 54,
           "num_prs": 50
         },
-        "macroscope": {
-          "precision": 41.8,
-          "recall": 40.9,
-          "f1": 41.3,
-          "tp": 56,
-          "fp": 78,
-          "fn": 81,
+        "mra-max": {
+          "precision": 7.7,
+          "recall": 31.4,
+          "f1": 12.4,
+          "tp": 43,
+          "fp": 515,
+          "fn": 94,
           "num_prs": 50
         },
-        "mra-nano": {
-          "precision": 10.8,
+        "kodus-v2": {
+          "precision": 35.9,
+          "recall": 34.3,
+          "f1": 35.1,
+          "tp": 47,
+          "fp": 84,
+          "fn": 90,
+          "num_prs": 50
+        },
+        "kg": {
+          "precision": 51.1,
           "recall": 16.8,
-          "f1": 13.1,
+          "f1": 25.3,
           "tp": 23,
-          "fp": 190,
+          "fp": 22,
           "fn": 114,
           "num_prs": 50
         },
-        "qodo-extended": {
-          "precision": 31.4,
-          "recall": 59.9,
-          "f1": 41.2,
-          "tp": 82,
-          "fp": 179,
-          "fn": 55,
+        "cubic-v2": {
+          "precision": 53.6,
+          "recall": 65.7,
+          "f1": 59.0,
+          "tp": 90,
+          "fp": 78,
+          "fn": 47,
           "num_prs": 50
         },
-        "cubic-dev": {
-          "precision": 27.3,
-          "recall": 73.0,
-          "f1": 39.8,
-          "tp": 100,
-          "fp": 266,
-          "fn": 37,
+        "devin": {
+          "precision": 46.4,
+          "recall": 37.2,
+          "f1": 41.3,
+          "tp": 51,
+          "fp": 59,
+          "fn": 86,
+          "num_prs": 50
+        },
+        "mra-b": {
+          "precision": 19.9,
+          "recall": 39.4,
+          "f1": 26.5,
+          "tp": 54,
+          "fp": 217,
+          "fn": 83,
           "num_prs": 50
         },
         "propel": {
@@ -32041,58 +32803,76 @@
           "fn": 87,
           "num_prs": 50
         },
-        "qodo-v22": {
-          "precision": 38.4,
-          "recall": 53.3,
-          "f1": 44.6,
-          "tp": 73,
-          "fp": 117,
-          "fn": 64,
+        "mra-a": {
+          "precision": 17.6,
+          "recall": 40.0,
+          "f1": 24.4,
+          "tp": 54,
+          "fp": 253,
+          "fn": 81,
+          "num_prs": 49
+        },
+        "claude-code": {
+          "precision": 27.4,
+          "recall": 41.6,
+          "f1": 33.0,
+          "tp": 57,
+          "fp": 151,
+          "fn": 80,
           "num_prs": 50
         },
-        "linearb": {
-          "precision": 8.0,
-          "recall": 2.9,
-          "f1": 4.3,
-          "tp": 4,
-          "fp": 46,
-          "fn": 133,
+        "qodo-v2-2": {
+          "precision": 41.5,
+          "recall": 43.1,
+          "f1": 42.3,
+          "tp": 59,
+          "fp": 83,
+          "fn": 78,
           "num_prs": 50
         },
-        "qodo": {
-          "precision": 25.6,
-          "recall": 44.5,
-          "f1": 32.5,
-          "tp": 61,
-          "fp": 177,
-          "fn": 76,
+        "greptile-v4": {
+          "precision": 29.3,
+          "recall": 58.4,
+          "f1": 39.0,
+          "tp": 80,
+          "fp": 193,
+          "fn": 57,
           "num_prs": 50
         },
-        "mra-ultra": {
-          "precision": 15.3,
-          "recall": 37.2,
-          "f1": 21.7,
-          "tp": 51,
-          "fp": 282,
-          "fn": 86,
+        "bito": {
+          "precision": 0,
+          "recall": 0.0,
+          "f1": 0,
+          "tp": 0,
+          "fp": 0,
+          "fn": 137,
           "num_prs": 50
         },
-        "entelligence": {
-          "precision": 27.7,
-          "recall": 20.4,
-          "f1": 23.5,
-          "tp": 28,
-          "fp": 73,
-          "fn": 109,
+        "mra-nano": {
+          "precision": 10.8,
+          "recall": 16.8,
+          "f1": 13.1,
+          "tp": 23,
+          "fp": 190,
+          "fn": 114,
           "num_prs": 50
         },
-        "coderabbit": {
-          "precision": 21.3,
-          "recall": 40.9,
-          "f1": 28.0,
-          "tp": 56,
-          "fp": 207,
-          "fn": 81,
+        "bugbot": {
+          "precision": 43.2,
+          "recall": 43.8,
+          "f1": 43.5,
+          "tp": 60,
+          "fp": 79,
+          "fn": 77,
+          "num_prs": 50
+        },
+        "qodo-v2": {
+          "precision": 36.2,
+          "recall": 56.2,
+          "f1": 44.0,
+          "tp": 77,
+          "fp": 136,
+          "fn": 60,
           "num_prs": 50
         }
       }
@@ -32138,7 +32918,7 @@
       "sort": "precision"
     },
     {
-      "id": "tool_greptile-v4_change_type_bug_fix",
+      "id": "tool_copilot_change_type_bug_fix",
       "label": "Best for Bug Fixes (Recall)",
       "filters": {
         "change_type": [
@@ -32146,13 +32926,13 @@
         ]
       },
       "description": "Bug fixes and issue resolution",
-      "best_model": "anthropic_claude-sonnet-4-5-20250929",
-      "best_tool": "greptile-v4",
+      "best_model": "anthropic_claude-opus-4-5-20251101",
+      "best_tool": "copilot",
       "best_score": 82.4,
       "sort": "recall"
     },
     {
-      "id": "tool_copilot_change_type_bug_fix",
+      "id": "tool_mesa_change_type_bug_fix",
       "label": "Best for Bug Fixes (Recall)",
       "filters": {
         "change_type": [
@@ -32160,13 +32940,13 @@
         ]
       },
       "description": "Bug fixes and issue resolution",
-      "best_model": "anthropic_claude-opus-4-5-20251101",
-      "best_tool": "copilot",
+      "best_model": "anthropic_claude-sonnet-4-5-20250929",
+      "best_tool": "mesa",
       "best_score": 82.4,
       "sort": "recall"
     },
     {
-      "id": "tool_mesa_change_type_bug_fix",
+      "id": "tool_greptile-v4_change_type_bug_fix",
       "label": "Best for Bug Fixes (Recall)",
       "filters": {
         "change_type": [
@@ -32175,7 +32955,7 @@
       },
       "description": "Bug fixes and issue resolution",
       "best_model": "anthropic_claude-sonnet-4-5-20250929",
-      "best_tool": "mesa",
+      "best_tool": "greptile-v4",
       "best_score": 82.4,
       "sort": "recall"
     },
@@ -32192,6 +32972,20 @@
       "best_tool": "cubic-v2",
       "best_score": 81.1
     },
+    {
+      "id": "tool_tusk_domain_ui",
+      "label": "Best for Ui (Recall)",
+      "filters": {
+        "domain": [
+          "UI"
+        ]
+      },
+      "description": "User interface and frontend.",
+      "best_model": "anthropic_claude-opus-4-5-20251101",
+      "best_tool": "tusk",
+      "best_score": 76.9,
+      "sort": "recall"
+    },
     {
       "id": "high_recall",
       "label": "Highest Recall",
@@ -32218,23 +33012,6 @@
       "best_tool": "qodo-v2-2",
       "best_score": 74.1
     },
-    {
-      "id": "tool_codeant_difficulty_moderate_context_file",
-      "label": "Moderate Bugs + File Context (Recall)",
-      "filters": {
-        "difficulty": [
-          "moderate"
-        ],
-        "context": [
-          "file"
-        ]
-      },
-      "description": "Requires careful reading. Requires full file understanding.",
-      "best_model": "anthropic_claude-opus-4-5-20251101",
-      "best_tool": "codeant",
-      "best_score": 73.3,
-      "sort": "recall"
-    },
     {
       "id": "tool_mra-a_language_java_risk_high",
       "label": "Java + High Risk (Recall)",
@@ -32252,6 +33029,23 @@
       "best_score": 73.3,
       "sort": "recall"
     },
+    {
+      "id": "tool_codeant_difficulty_moderate_context_file",
+      "label": "Moderate Bugs + File Context (Recall)",
+      "filters": {
+        "difficulty": [
+          "moderate"
+        ],
+        "context": [
+          "file"
+        ]
+      },
+      "description": "Requires careful reading. Requires full file understanding.",
+      "best_model": "anthropic_claude-opus-4-5-20251101",
+      "best_tool": "codeant",
+      "best_score": 73.3,
+      "sort": "recall"
+    },
     {
       "id": "ruby_medium",
       "label": "Best for Medium Ruby PRs",
@@ -32342,22 +33136,6 @@
       "best_tool": "bugbot",
       "best_score": 66.7
     },
-    {
-      "id": "tool_qodo-v2_language_java_risk_high",
-      "label": "Java + High Risk",
-      "filters": {
-        "language": [
-          "Java"
-        ],
-        "risk": [
-          "high"
-        ]
-      },
-      "description": "Java codebases with OOP patterns. Significant impact, potential data loss.",
-      "best_model": "anthropic_claude-opus-4-5-20251101",
-      "best_tool": "qodo-v2",
-      "best_score": 66.7
-    },
     {
       "id": "tool_macroscope_context_file_concern_correctness",
       "label": "File Context + Correctness",
@@ -32374,6 +33152,22 @@
       "best_tool": "macroscope",
       "best_score": 66.7
     },
+    {
+      "id": "tool_qodo-v2_language_java_risk_high",
+      "label": "Java + High Risk",
+      "filters": {
+        "language": [
+          "Java"
+        ],
+        "risk": [
+          "high"
+        ]
+      },
+      "description": "Java codebases with OOP patterns. Significant impact, potential data loss.",
+      "best_model": "anthropic_claude-opus-4-5-20251101",
+      "best_tool": "qodo-v2",
+      "best_score": 66.7
+    },
     {
       "id": "tool_greptile_pr_size_medium_context_file",
       "label": "Medium PRs + File Context",
@@ -32907,7 +33701,8 @@
     "entelligence": "Entelligence",
     "cubic-dev": "Cubic",
     "sourcery": "Sourcery",
-    "mesa": "Mesa"
+    "mesa": "Mesa",
+    "tusk": "Tusk"
   },
   "tool_colors": {
     "graphite": "#6366f1",
@@ -32925,7 +33720,8 @@
     "entelligence": "#0ea5e9",
     "cubic-dev": "#d946ef",
     "sourcery": "#84cc16",
-    "mesa": "#f43f5e"
+    "mesa": "#f43f5e",
+    "tusk": "#9333ea"
   },
   "default_model": "anthropic_claude-opus-4-5-20251101",
   "min_prs_threshold": 5
diff --git a/offline/analysis/benchmark_dashboard.py b/offline/analysis/benchmark_dashboard.py
index 6b3c696..42332f0 100644
--- a/offline/analysis/benchmark_dashboard.py
+++ b/offline/analysis/benchmark_dashboard.py
@@ -31,6 +31,7 @@
     "cubic-dev": "Cubic",
     "sourcery": "Sourcery",
     "mesa": "Mesa",
+    "tusk": "Tusk",
 }
 
 TOOL_COLORS = {
@@ -50,6 +51,7 @@
     "cubic-dev": "#d946ef",
     "sourcery": "#84cc16",
     "mesa": "#f43f5e",
+    "tusk": "#9333ea",
 }
 
 
diff --git a/offline/results/anthropic_claude-opus-4-5-20251101/candidates.json b/offline/results/anthropic_claude-opus-4-5-20251101/candidates.json
index 9445024..ec15bba 100644
--- a/offline/results/anthropic_claude-opus-4-5-20251101/candidates.json
+++ b/offline/results/anthropic_claude-opus-4-5-20251101/candidates.json
@@ -859,6 +859,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, meaning a user with only VIEW permission on groups can pass canManage() checks. This gates group creation via GroupsResource.addTopLevelGroup() which calls auth.groups().requireManage(). The per-group canManage(GroupModel) correctly checks only MANAGE. This is a copy-paste from canView() where checking both scopes is correct.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getGroupIdsWithViewPermission() uses groupResource.getId() (the authorization resource server's internal UUID) instead of groupResource.getName() (which stores the actual group model ID). This causes hasPermission() at line 122 to call resourceStore.findByName(server, <resource-internal-id>) which won't find the resource by name, so it always falls through to the 'all-groups' check, making per-group permission evaluation broken.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getGroupIdsWithViewPermission() adds resource UUIDs via granted.add(groupResource.getId()), but downstream in JpaUserProvider.groupsWithPermissionsSubquery() they're compared against the groupId column in the user-group membership table, which stores group model IDs \u2014 so the query filter will never match.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The interface docstring for canManage() states 'Also if it has permission to VIEW or MANAGE groups', which codifies the privilege escalation as intentional. If canManage() is fixed to only check MANAGE, this docstring is misleading and should be updated. If the docstring is 'correct', then canManage() is intentionally granting manage privileges to view-only users, which is a design-level security issue.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In the test @BeforeEach method, ApiUtil.handleCreatedResponse(response) is called on a response that's wrapped in a try-with-resources. Since handleCreatedResponse was refactored to also use try-with-resources internally (closing the response), the response is asserted for status 201 twice and closed twice. This pattern is used at lines 73-77, 195-199, and 261-265.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/33832": {
@@ -1746,6 +1778,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures, meaning bugs specific to handling full-size ES384/ES512 r/s values would not be caught.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name, producing a malformed log message.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/40940": {
@@ -2478,6 +2536,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getSubGroupsStream(String, Integer, Integer) at line 256 calls modelSupplier.get().getSubGroupsStream(...) without a null check, which will throw an NPE under concurrent-deletion scenario. The same applies to the overloads at lines 262 and 268. All three methods should guard against modelSupplier.get() returning null.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The reader thread is never joined before the assertion. After deletedAll.set(true), the test asserts caughtExceptions is empty, but the reader thread may still be in the middle of a groups() call. If that in-flight call throws an exception, it will be added to caughtExceptions after the assertion has already passed, causing the test to silently miss failures.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1": {
@@ -3204,6 +3276,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the method signature in parent class UsernamePasswordForm was changed to `isConditionalPasskeysEnabled(UserModel user)`, requiring one parameter. The call should be `isConditionalPasskeysEnabled(context.getUser())`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Regression: passkeys no longer set up during initial login. In `authenticate()`, the old code called `fillContextForm` when `context.getUser() == null` (initial login case), but the new code calls it only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This breaks passkey conditional UI on the initial login page.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null` (initial login error recovery), but the new code calls it only when `user != null`. This prevents passkey UI elements from being re-rendered on the error page after an invalid username entry.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Potential NullPointerException in `fillContextForm`: The refactored code passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. The method is `protected` and designed for override, so a future subclass could override it to return `true` without a null-user guard, causing NPE where `user` would be null.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/93824": {
@@ -3977,6 +4075,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Inconsistent metrics tag key: 'shards' (plural) is used here, while all other metrics in the same function use 'shard' (singular) as the tag key, causing this metric to be silently missing from dashboards or alerts that filter on the 'shard' tag",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When the deadline is exceeded during join(), the break on line 341 exits the loop entirely, skipping .terminate() for all remaining processes. With multiple processes, if process 0 takes the entire timeout, processes 1..N never get terminated and will remain alive as zombies until the parent exits",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Each sub-buffer's get_memory_info() calls iter_cluster_memory_usage(self.client), which queries Redis INFO on ALL nodes in the cluster. Since all sub-buffers share the same Redis cluster, calling this N times makes N redundant Redis INFO roundtrips, adding unnecessary latency to each submit() call on the hot path",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5": {
@@ -5217,6 +5335,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list, silently pairing wrong error IDs with wrong event data",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds) are compared directly against replay segment event timestamps which use Unix epoch milliseconds, causing all error events to always appear before any replay breadcrumbs rather than being interleaved in chronological order",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1": {
@@ -6316,6 +6448,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The method should convert datetime to float via float(value.strftime('%s.%f')) before calling math.floor(), like DateTimePaginator.get_item_key does",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), queryset[start_offset:stop] will raise AssertionError('Negative indexing is not supported.'). The comment stating Django ORM handles negative slicing is incorrect",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter, producing incorrect results or a database type error",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "organization_context.member can be None when the user has no membership in the org. Accessing .has_global_access on None raises AttributeError. This executes on every request to the endpoint because the enable_advanced check runs unconditionally before the use_optimized guard",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BasePaginator allows negative start_offset when cursor.is_prev is True without clamping. A malicious or malformed cursor string with a negative offset (e.g., '0:-3:1') will cause Django's AssertionError on queryset slicing",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/97529": {
@@ -6988,6 +7152,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Race condition from lock scope reduction: The old code held cacheMu.Lock() for the entire BuildIndex function, serializing all concurrent calls. The new code only locks the final cache write. Two concurrent callers can simultaneously build an index for the same key, causing bleve.New(dir, mapper) to fail for file-based indexes when the directory already exists, or duplicate work and silent overwrites for memory indexes.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Data race on shared err variable in concurrent goroutines: The err variable from stats, err := s.storage.GetResourceStats(...) is captured by reference in all group.Go closures. Multiple goroutines concurrently write to this same err via _, _, err = s.build(...) and then read it via return err, causing both a data race and correctness bug where a goroutine may return another goroutine's error or nil instead of its own.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Data race on totalBatchesIndexed: This variable is incremented from multiple goroutines launched via errgroup without any synchronization. With initWorkers > 1, concurrent increments are a data race.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/80168": {
@@ -7636,7 +7820,33 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "MetricAlertDetectorHandler extends StatefulDetectorHandler but implements none of its abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), causing TypeError when instantiated",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In TestEvaluateGroupKeyValue.test_dedupe, expected occurrence is built with group_key='val1' but evaluate_group_key_value is called with group_key='group_key', producing mismatched fingerprints that only pass because IssueOccurrence.__eq__ compares solely by id",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In test_state_results_multi_group, expected occurrence for group_2 is built with value=6 but actual data packet has group_2=10, passing only because mock function doesn't vary output based on value and IssueOccurrence.__eq__ compares by id",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In test_state_results, detector.detector_handler creates a new handler instance each time the property is accessed, so the handler used in process_detectors differs from the one used to build expected occurrence, making the test fragile if handler-instance-specific state is ever used",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/getsentry/sentry/pull/80528": {
     "gemini": [
@@ -8097,6 +8307,14 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In `get_monitor_environment_context`, line 168 returns `monitor_environment.monitor.config` (the original, unmodified config) instead of the local `config` variable, causing the `schedule_type` override made on line 162 to be silently discarded and the monitor context to always contain the raw `schedule_type` value instead of the human-readable display value",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/77754": {
@@ -8660,6 +8878,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` will share the same timestamp. Fix is `queued: datetime = field(default_factory=timezone.now)`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `assignment_source` parameter was added to the abstract `sync_status_outbound` method but status sync is unrelated to assignee sync cycles. `ExampleIntegration.sync_status_outbound` does not accept `**kwargs`, so any caller passing `assignment_source` to it will get a TypeError",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array - misleading name",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test method name `test_from_dict_inalid_data` has a typo - missing 'v' in 'invalid'",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/95633": {
@@ -9381,6 +9625,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "close() joins the commit thread BEFORE shutting down the worker pool. After shutdown_event is set, the commit thread does one final commit and exits. Then queue_pool.shutdown() drains remaining items from worker queues (queue.shutdown(immediate=False) lets workers process remaining items). These items are processed (side effects occur) but the commit thread is already dead, so their offsets are never committed. On restart, these items will be re-consumed and processed again, causing duplicate side effects.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_thread_queue_parallel_error_handling's docstring says 'Test that errors in processing don't block offset commits for other messages' but its assertion checks the OPPOSITE: asserts no commits happened. The code always marks offsets as complete via the finally block in OrderedQueueWorker.run(), so offsets WILL be committed for both the failed and successful message. The test only passes due to timing: the commit loop waits 1 second between iterations and the test only waits 0.2s after processing completes, so the commit hasn't fired yet. This test is inherently flaky.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' and 'Only commits offsets after successful processing', but neither is true. The queues are created as queue.Queue() without a maxsize parameter, so they are unbounded and never 'fill up' - there is no backpressure mechanism. If workers are slow and the producer is fast, memory will grow without bound.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The finally block in OrderedQueueWorker.run() always calls complete_offset regardless of whether processing succeeded or raised an exception, contradicting the claim that it 'Only commits offsets after successful processing'.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OffsetTracker._get_partition_lock accesses self.partition_locks without synchronization - two threads calling it with the same new partition simultaneously will both create a Lock() but setdefault ensures only one is stored. This is safe under CPython's GIL but not guaranteed by the Python language spec.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "get_committable_offsets iterates list(self.all_offsets.keys()) without holding any lock, while worker threads may be modifying all_offsets via defaultdict auto-creation. This relies on CPython GIL atomicity guarantees that are not part of the language specification.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2": {
@@ -10584,6 +10866,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. When used with order_by=\"-datetime\" on AuditLogEntry, getattr(item, self.key) returns a datetime that needs to be converted to float first like DateTimePaginator does via float(value.strftime(\"%s.%f\")) * self.multiplier",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing and raises AssertionError(\"Negative indexing is not supported.\"). The comment claiming Django handles negative slicing automatically is incorrect, causing 500 errors with crafted cursors",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The safety check that skipped segments exceeding max_segment_spans during flush was removed. While the Lua script caps spans at 1000 during insert, the flush-side guard protected against edge cases like race conditions or spans inserted before eviction logic was deployed. The max_segment_spans attribute is now dead code",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3": {
@@ -11593,6 +11895,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Python's built-in hash() is randomized per-process (via PYTHONHASHSEED), causing different Sentry worker processes to compute different hash values for the same project IDs, which defeats the caching optimization since cross-process cache lookups will always miss",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The invalidate_upsampling_cache function uses the same non-deterministic hash() in its cache key construction, so invalidation from one process would not clear the cache entry written by a different process",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The cache key doesn't include dataset or request context, so the same cache entry is reused across different query types for the same org+projects, potentially returning inconsistent results when cached True is combined with different query contexts",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The behavior was changed from any(project_id in allowlist) to all(project_id in allowlist), which could silently return incorrect underestimated data for mixed-project queries where some projects are allowlisted but not all",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The validation that 'timestamp' was not in raw_groupby was removed during refactoring, allowing users to now group by 'timestamp' which was previously explicitly disallowed and could cause unexpected query behavior",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/103633": {
@@ -12340,6 +12674,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. The comment says \"Allow access to the dashboard to prove this is not checked\" but the permission cache value is set to `false`, which would also deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix overlaps with the permission cache key prefix. If a denial key accidentally matches a permission key, `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"`, suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates cache miss vs. cache hit with stale data into the same metric label, making monitoring difficult.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/67876": {
@@ -13106,6 +13466,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The OAuth state parameter used to prevent CSRF is pipeline.signature, which is a deterministic MD5 hash of the pipeline view class names. This value is identical for every user and session, allowing an attacker who knows the pipeline configuration to compute and forge the state parameter, defeating CSRF protection. OAuth state should include a cryptographic random nonce tied to the user's session.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Accessing integration.metadata[\"sender\"][\"login\"] will raise a KeyError if the sender key is not present in the integration's metadata. The sender field is only set when the integration is created via a GitHub webhook installation event, so integrations created before this code change or through other paths will cause a 500 error instead of the intended user-facing error page.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When the Integration does not exist yet (Integration.DoesNotExist), the code calls pipeline.next_step() without performing the authenticated user check. This means the first installation of a GitHub app completely bypasses the sender/user validation, allowing an attacker to install a GitHub app on a victim's Sentry org without being the GitHub user who initiated the installation.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The redirect_uri is included in the authorization request but is missing from the token exchange request. Per OAuth2 spec (RFC 6749, Section 4.1.3), if a redirect_uri was included in the authorization request, the same redirect_uri MUST be included in the access token request. GitHub may reject the token exchange or behave unexpectedly when redirect_uri is omitted.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The error() helper function receives org (which is self.active_organization) and calls org.organization, but self.active_organization can be None. If the active_organization is a different type than expected, accessing org.organization would crash.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/32918": {
@@ -13612,6 +14004,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Test cleanup registers removal for a literal \"alias\" instead of the actual IDP alias (e.g., \"idp-alias-\" + i), meaning all 20 IDPs created in the loop are never cleaned up after the test, potentially causing state leakage to other tests in the suite. The same bug exists at line 425 for idp-alias-20.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The getForLogin cached path returns results from a HashSet<IdentityProviderModel> which does not preserve any ordering from the delegate's query, causing inconsistent behavior between cached and non-cached paths for any future caller relying on delegate ordering.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In registerIDPLoginInvalidationOnUpdate, the original model is fetched via getById() which wraps it with createOrganizationAwareIdentityProviderModel overriding isEnabled(), while the updated model is raw (not wrapped). This causes unnecessary cache invalidation when comparing with getLoginPredicate() because isEnabled() may yield different results for original vs updated even if the IDP's own enabled field hasn't changed.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/94942": {
@@ -14246,6 +14658,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code. Both code paths return `false`, so the convoluted logic with a meaningless feature flag check is misleading and should be simplified to just `return false`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The stub `DB.QueryFramesInto` method always returns an error ('not implemented'), but the caller in `sql_command.go:Execute` treats this error as a query failure and sets `rsp.Error = err`, then continues to build a result with an empty frame and appends it to `rsp.Values`. This is a latent issue if the guard is ever removed.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/90939": {
@@ -14651,6 +15077,14 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether entryPointAssetsCache was already populated by another goroutine, causing redundant file reads and a thundering herd problem at startup with many concurrent requests",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/80329": {
@@ -15497,6 +15931,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The cleanup service ticker was changed from 10 minutes to 1 minute, appearing to be an accidental debug leftover. The clean() method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running, and will increase database load significantly in production.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "All diagnostic log lines in the cleanup path use r.log.Error() instead of r.log.Debug() or r.log.Info(). These informational messages fire on every successful cleanup cycle, which will flood error-level logs and trigger monitoring/alerting systems in production.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The SQLite parameter limit guard checks r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit but should check len(ids) > sqliteParameterLimit. The guard should reflect the actual parameter count, not the configured ceiling, to avoid unnecessarily building inline SQL strings instead of using safer parameterized queries.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test helper createTestAnnotations hardcodes annotation IDs as int64(i + 1) via InsertMulti. Since tests share a single db.InitTestDB(t) and SQLite's auto-increment counter is not reset by DELETE FROM annotation, auto-generated IDs may collide with IDs from a previous createTestAnnotations call if any test case inserts without explicit IDs.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/90045": {
@@ -16741,6 +17201,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In Create's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. The storage operation failed, but the metric is recorded against the legacy histogram, corrupting both metrics.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In Update's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. Same issue as Create \u2014 the storage call failed but the latency is attributed to legacy.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In DeleteCollection's goroutine, `recordStorageDuration` is called instead of `recordLegacyDuration` for the legacy write. This is the inverse mistake \u2014 legacy latency is recorded in the storage histogram.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In Delete's success path, the third argument to `recordStorageDuration` is `name` (the object name) instead of `options.Kind`. This will create high-cardinality metric labels (one time series per object name), which can cause performance issues in Prometheus.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In the Delete method, `klog.NewContext(ctx, d.Log)` uses `d.Log` (the base logger) instead of the enriched `log` variable that includes the `name`, `kind`, and `method` values.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/106778": {
@@ -17761,6 +18253,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Missing React `key` prop on GrafanaRuleListItem in the filter view's list rendering - the old GrafanaRuleLoader had `key={key}`, but the replacement GrafanaRuleListItem omits it, causing React reconciliation issues when the list updates",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Delete menu item will never render in the new list view because AlertRuleMenu guards the delete menu item with `canDelete && rulerRule && (...)`, but rulerRule is always undefined when coming from GrafanaRuleListItem since the new list view no longer fetches ruler data",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Pause menu item will never render in the new list view because `canPause && rulerRuleType.grafana.rule(rulerRule)` requires rulerRule to be a grafana ruler rule, but rulerRule is undefined when coming from the new list view",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Silence drawer won't open in the new list view because the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` which is false when rule (RulerRuleDTO) is undefined from GrafanaRuleListItem",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Type mismatch in useGrafanaPromRuleAbility hook - called with promRule typed as Rule but hook expects GrafanaPromRuleDTO | SkipToken, and prometheusRuleType.grafana.rule() returns boolean not a type predicate, potentially causing incorrect permission checks due to missing folderUid field",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/107534": {
@@ -18291,7 +18815,27 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "Double interpolation of queries in the shard splitting path: runShardSplitQuery() calls applyTemplateVariables() at line 52, then runSplitQuery() also calls applyTemplateVariables() at line 299, causing ad-hoc filters to be added to the query expression twice since addAdHocFilters() unconditionally appends label matchers",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test mock for applyTemplateVariables mutates the input query object directly instead of returning a new object as the real implementation does, hiding potential bugs where code relies on the original targets remaining unmodified",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Filter ordering change in runShardSplitQuery may cause behavioral difference: the new code filters before interpolation, so a query with expr as a template variable that resolves to empty string will pass the filter but result in an empty expression being sent to splitQueriesByStreamShard",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/grafana/grafana/pull/79265": {
     "gemini": [
@@ -19256,6 +19800,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "TOCTOU race in device limit enforcement. CreateOrUpdateDevice first counts devices, then conditionally calls updateDevice. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The updateDevice method returns ErrDeviceLimitReached when rowsAffected == 0, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its updated_at falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with ErrDeviceLimitReached instead of being re-created or handled differently.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The AnonymousDeviceLimit field in FrontendSettingsDTO is int64 with json:\"anonymousDeviceLimit\" \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as number | undefined. The Go DTO has no omitempty tag, so the field is never undefined in practice. Any frontend code checking config.anonymousDeviceLimit === undefined to detect 'no limit configured' will never match; it will see 0 instead.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9": {
@@ -19904,7 +20474,27 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "LRU translation cache key does not include SiteSetting.default_locale, causing stale cached translations to be served when admin changes the site's default locale since there is no I18n.reload! triggered on default_locale change",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The translate method only loads config.locale but with I18n::Backend::Fallbacks the backend iterates through the fallback chain - if fallback locale YML files haven't been loaded, lookups return nil and fallbacks silently fail in code paths outside controllers (Sidekiq jobs, rake tasks, console)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The ensure_loaded! method initializes @loaded_locales ||= [] outside the LOAD_MUTEX, causing potential race condition in multi-threaded servers where two threads could overwrite each other's array, leading to redundant load_locale calls or skipped locale loading",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/grafana/grafana/pull/76186": {
     "gemini": [
@@ -20437,6 +21027,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(), causing any test using TestLogger to verify logged output to silently lose all log messages since the .Info() call goes to a throwaway instance. The method should return the receiver (f) to preserve log capture.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic if the middleware is used outside the Decorator (e.g., in tests), unlike LoggerMiddleware which has explicit nil request guard clauses.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10": {
@@ -21978,6 +22582,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist, the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote will break the SQL or allow injection.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist. In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value, `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Fabricator file contents are swapped. `category_fabricator.rb` contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators. The file names don't match their contents.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7": {
@@ -22846,6 +23488,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Light-theme lightness value changed from 30% to 70% in desktop/topic-post.scss for .embedded-posts link text - values appear to be accidentally swapped with dark theme",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Light-theme lightness value changed from 70% to 30% in mobile/modal.scss for .custom-message-length - opposite of desktop version, making character count text inverted between light and dark themes on mobile",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Light-theme lightness for h3 changed from 20% to 50% in mobile/topic-post.scss - makes headings significantly more faded/grey on light themes, dark-theme value should have been 80% to complement 20%",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Light-theme lightness for .name changed from 30% to 50% in both desktop and mobile user stylesheets - makes user names more washed out, inconsistent with nearby .username a in common/base/user.scss which correctly preserves 30%",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8": {
@@ -24047,6 +24715,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Breaking change in the public GroupsController#members endpoint response format - old code returned a flat JSON array, new code returns { members: [...], meta: { total, limit, offset } }, breaking existing consumers expecting a flat array",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "GroupsController#members endpoint now always paginates with default limit of 50, whereas old code had no limit/offset for non-automatic groups, causing large groups to silently return truncated results to existing callers that don't pass pagination params",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Off-by-one in totalPages calculation - Math.floor(user_count / limit) + 1 yields one extra page when user_count is exactly divisible by limit (e.g., 100 users with limit 50 gives 3 pages instead of 2), should use Math.ceil(user_count / limit)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test for 'cannot remove members from automatic groups' uses xhr :put but the route is defined as delete, so the test doesn't validate the actual HTTP method and route misconfiguration bugs would go undetected",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In update action, group.visible = params[:visible] == 'true' is executed unconditionally, so if a caller sends only partial params without visible parameter, visible gets set to false, causing partial PUT requests from API consumers to reset visibility",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "New groups will always be created with visible: false regardless of checkbox state because asJSON() sends visible: !!this.get('visible') which evaluates false for a new group where visible is undefined/null, and the visible property is never initialized to true in the Group model defaults",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3": {
@@ -24829,6 +25535,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation, which can run multiple times per user action, inflating blocked-email statistics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "record.save inside should_block? triggers BlockedEmail's own validations including a uniqueness DB query, making validation unnecessarily expensive",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The JS code adds email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures, causing valid emails belonging to other users to be permanently marked as rejected in the client",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "rejectedEmails is initialized as Em.A([]) directly in the class definition hash passed to .extend(), causing the array to be shared across all instances as a prototype property instead of instance property",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BlockedEmail.should_block? calls record.save but doesn't check the return value, silently ignoring failures and potentially returning blocking decisions based on stale in-memory state with dirty incremented match_count",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5": {
@@ -25571,7 +26309,45 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "The `order` mixin passes the raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values while the standard `order` property is 0-based. The correct value for `-webkit-box-ordinal-group` should be `$val + 1`, otherwise elements using `@include order(1)` would be indistinguishable from unordered elements in old WebKit browsers",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `.small-action`, the parent is now `display: flex` but child `.topic-avatar` still has `float: left`, indicating an incomplete migration - the float should be removed for clarity",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that created spacing from the avatar, causing the gap between avatar and description text to shrink significantly on narrow viewports and may overlap or appear too tight",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`-ms-align-items` is not a valid CSS property and will be silently ignored by IE10/11 - the correct IE prefix for `align-items` is only `-ms-flex-align`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`-ms-flex-align` uses different value keywords (`start`/`end` instead of `flex-start`/`flex-end`), so the mixin does not translate values and future callers passing `flex-start` or `flex-end` would silently break in IE",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `.badge-wrapper.bullet` rule had `margin-top: 5px` removed in the header's `.extra-info-wrapper` context but the same selector in `#topic-title .title-wrapper` retains `margin-top: 5px`, creating inconsistent bullet badge styling between contexts",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6": {
     "gemini": [
@@ -26285,6 +27061,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized even when `website` is nil/blank.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`website_name` is not included in the `untrusted_attributes` list but `website` is. Since `website_name` bypasses the `restrict_user_fields?` check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `Discourse.stubs(:current_hostname)` call happens inside the `it` block after `json` is already defined as a lazy let. If `json` were ever evaluated before the stub, the stub would not be in effect. The stub should be in a `before` block before `json` is accessed.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4": {
@@ -28086,6 +28882,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression. This will raise a SyntaxError when the best.html.erb template is rendered.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a substring of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`contents << \"\\n<hr>...\"` uses the `<<` operator which mutates the string in place, modifying the caller's original string. If `import` is called twice with the same string, the imported-from footer is appended multiple times. When `import` is called for an update path, the SHA1 was computed after mutation, so re-importing the same content will always differ from the stored SHA1.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1": {
@@ -29203,6 +30031,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The 413 error handler also hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2": {
@@ -30165,6 +31019,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "NullPointerError crash: TopicUser.find_by returns nil when no TopicUser record exists, and the next line calls tu.notification_level on the nil result, raising NoMethodError",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The unsubscribe_link i18n string requires an %{unsubscribe_url} interpolation variable, but existing email paths using add_unsubscribe_link: true without passing unsubscribe_url will raise I18n::MissingInterpolationArgument error or render a broken link",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS vulnerability: title value is injected directly into HTML via string concatenation without escaping, allowing user-controlled content with script tags to be rendered as raw HTML",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS vulnerability in unsubscribe template: triple-stache {{{ }}} renders raw HTML and the computed property interpolates model.fancyTitle which can contain malicious HTML that would be rendered unescaped",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The html_part method only substitutes %{respond_instructions} and %{unsubscribe_link} placeholders when add_unsubscribe_link is true, but the template always contains these placeholders, causing literal placeholder strings to appear in emails when add_unsubscribe_link is false",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/22532": {
@@ -31231,6 +32117,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `deleteCache` handler only checks `userId: user.id` on the credential but does not account for delegation credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `deleteCache` handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` is semantically incorrect \u2014 `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. `?? null` would be more semantically correct.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `CredentialActionsDropdown` component initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them, even when the component returns null.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/8330": {
@@ -32159,6 +33083,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range, it prevents checking whether the slot is valid for a different user's override. The combination of mutation and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/14943": {
@@ -33328,6 +34278,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Public methods getAuthorizationConditions() and getFilterConditions() return Prisma.Sql fragments containing OR operators without outer parentheses, causing SQL operator precedence bugs when composed with AND in raw queries, potentially leading to privilege escalation",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Constructor parameter type changed from InsightsBookingServiceOptions (discriminated union requiring teamId for scope='team') to InsightsBookingServicePublicOptions (flat type with optional teamId), removing compile-time checking when teamId is missing for team scope",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Falsy guard 'if (this.filters.eventTypeId)' skips the filter when eventTypeId is 0, and similarly 'if (this.filters.memberUserId)' skips memberUserId 0, causing silent unfiltered results instead of filtering by ID=0",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getBaseConditions method has unreachable dead code branches because getAuthorizationConditions() always returns a truthy Prisma.Sql object, and auth failures don't short-circuit filter condition computation",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Caching test suite was removed but caching logic still exists in the service, leaving no tests to catch regressions that could cause repeated database queries",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/11059": {
@@ -35197,6 +36179,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously googleCredentialSchema.parse() returned just the data. Now the DB will store {success: true, data: {access_token, refresh_token, expiry_date}} instead of {access_token, refresh_token, expiry_date}, corrupting the credential for all subsequent reads.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses res?.data (undefined on fetch Response \u2192 TypeError on token.access_token), Hubspot accesses .expiryDate/.expiresIn (undefined on Response), Zoho-bigin and Zohocrm access .data.error (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "minimumTokenResponseSchema uses [z.string().toString()] as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field exists in the response, making the 'minimum token validation' ineffective.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Salesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using credentialKey.access_token \u2014 the OLD, pre-refresh token. The newly refreshed token from accessTokenParsed.data is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/7232": {
@@ -36261,6 +37275,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record, causing the record to become permanently orphaned",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The CRON job first deletes ALL email reminders with a past `scheduledDate`, then queries for `cancelled: true` reminders to cancel via SendGrid. If a cancelled reminder's `scheduledDate` has already passed, the first `deleteMany` removes the DB record before the cancel loop processes it, so the email is never cancelled in SendGrid and gets delivered despite the booking being cancelled",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails, the exception aborts the loop and all remaining reminders are never cancelled or deleted, causing those emails to be sent despite their bookings being cancelled",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget and the surrounding try/catch will never catch rejections from these promises",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The sender name/sender ID input section was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}` but after the refactor is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/14740": {
@@ -37513,6 +38559,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When `booking.userId` is null, `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable if the user is an attendee of a booking whose organizer account was deleted.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. If the attendees array is ever empty, this would throw a runtime TypeError crashing the email send for the entire batch.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/10600": {
@@ -38639,6 +39717,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Password bypass via backup code + dummy totpCode. An attacker with a valid backup code can send a request with a dummy non-empty totpCode (e.g., 'x') alongside the stolen backupCode. The non-empty totpCode causes the password check at line 121 to be skipped, then the backup code path at line 131 succeeds without ever verifying the password, allowing login with only email + backup code.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Backup code login is blocked for non-CAL identity provider users. The guard at line 113 throws ThirdPartyIdentityProviderEnabled when !credentials.totpCode is true. When a Google/SAML user with 2FA tries to use a backup code (no totpCode provided), this check blocks the request before the backup code path is reached, making backup codes completely non-functional for these users.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Unhandled exception in API route returns unstructured error. When CALENDSO_ENCRYPTION_KEY is missing in the backup code path, throw new Error(ErrorCode.InternalServerError) causes Next.js to return a generic 500 HTML error page instead of a structured JSON response, causing the client code to fail when calling response.json().",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Object URL created for backup codes download is never revoked on component unmount or dialog close. URL.createObjectURL(textBlob) creates a blob URL that persists until explicitly revoked, but resetState (called on dialog close/cancel) does not revoke it and there is no cleanup effect, causing each 2FA setup flow to leak one blob URL.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/10967": {
@@ -40090,6 +41194,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In both `updateEvent` and `deleteEvent`, when `externalCalendarId` is falsy, the else-branch searches `event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId)` \u2014 but `externalCalendarId` is guaranteed falsy at that point, so the `.find()` will never match any element and always returns `undefined`. The old code correctly fell back to `event.destinationCalendar?.externalId`. This means calendar updates and deletes will use `undefined` as the calendar ID, likely falling back to the wrong calendar or failing.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When `evt.destinationCalendar` is `null` or an empty array, `const [mainHostDestinationCalendar] = evt.destinationCalendar ?? []` sets `mainHostDestinationCalendar` to `undefined`. The next line accesses `mainHostDestinationCalendar.integration` without optional chaining, which throws `TypeError: Cannot read property 'integration' of undefined`. Multiple callers (e.g., webhook handlers, payment handlers) set `destinationCalendar` to `[]` when no calendar is configured.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `updateAllCalendarEvents`, if `reference.credentialId` is truthy but the credential is not found in `this.calendarCredentials` AND the DB lookup fails (null result or no `app.slug`), `credential` remains `undefined`. It is then unconditionally passed to `updateEvent(credential, ...)` on line 542, which expects a `CredentialWithAppName`. This will crash when `updateEvent` calls `getCalendar(credential)` on the undefined value.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `handleNewBooking`, when neither `eventType.destinationCalendar` nor `organizerUser.destinationCalendar` exist, `evt.destinationCalendar` is set to `null` (line 1067). The subsequent `evt.destinationCalendar?.push(...teamDestinationCalendars)` (line 1078) is then a no-op due to optional chaining. For collective team events where the organizer lacks a calendar but team members have destination calendars, those team member calendars are silently dropped \u2014 the event will have no destination calendars despite team members having them configured.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The slug/requestedSlug logic was inverted from the original. Old code: `...(!IS_TEAM_BILLING_ENABLED && { slug })` set `slug` when billing is DISABLED (org gets slug immediately, no payment needed). New code: `...(IS_TEAM_BILLING_ENABLED ? { slug } : {})` sets `slug` when billing is ENABLED. Combined with the `requestedSlug` line (which also only fires when billing is enabled), organizations without billing now get neither `slug` nor `requestedSlug`, while organizations with billing get both \u2014 the opposite of the original intent.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/8087": {
@@ -41053,6 +42189,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "async callback inside forEach is fire-and-forget in packages/app-store/vital/lib/reschedule.ts - the forEach does not await the async callback, so the try/catch block on line 124 will never catch errors from the async operations, and deleteEvent/deleteMeeting calls are not awaited before the function proceeds to send reschedule emails on line 143",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "async-forEach fire-and-forget bug in packages/app-store/wipemycalother/lib/reschedule.ts - the async callback's promises are discarded, so calendar deleteEvent calls are unawaited, errors won't be caught and the function proceeds immediately to send reschedule emails",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "async-forEach fire-and-forget bug in packages/trpc/server/routers/viewer/bookings.tsx - the forEach on line 553 uses an async callback but forEach ignores the returned promises, the deleteEvent/deleteMeeting operations are unawaited, and any errors from the async operations are silently swallowed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "async-forEach fire-and-forget in handleCancelBooking at packages/features/bookings/lib/handleCancelBooking.ts - the forEach on line 460 uses an async callback but forEach discards the promises, causing the apiDeletes array to be empty when Promise.all(apiDeletes) is called later, silently skipping calendar deletions for recurring events",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getCalendarCredentials is a synchronous function that calls getCalendar (now async) inside flatMap on line 28 in packages/core/CalendarManager.ts - since getCalendar returns Promise<Calendar | null>, the calendar field is actually a Promise not a Calendar, breaking the type contract and causing any caller that doesn't await item.calendar to get a truthy Promise object instead of null, bypassing null checks",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/37429": {
@@ -42063,6 +43231,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian, affecting both account and login message files",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f), inconsistent with the rest of the file",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. This allows <br>, <p>, <strong>, <b> tags in translations even when the English source only has anchor tags, defeating the stated intent",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/37634": {
@@ -42886,6 +44074,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Copy-paste bug: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`, meaning `rawTokenId` is never null-checked and null values will be silently accepted, potentially causing NPEs later when the token ID is used (e.g., in `encodeTokenId()` which concatenates it)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Wrong substring indices in `isAccessTokenId` matcher: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType), should be `substring(4, 6)`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Inverted boolean logic in `isAccessTokenId` matcher: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match, the condition should be negated",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/38446": {
@@ -43667,6 +44875,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID because createFromValues() does not set an ID on the model and the stored ID is never propagated, causing removeStoredCredentialById() to fail when all codes are exhausted",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The test provider's isValid() for recovery codes accepts ANY code via anyMatch() instead of only validating against the NEXT sequential code like the production RecoveryAuthnCodesCredentialProvider.isValid() does",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "isValid() never consumes/removes the used recovery code, unlike the production provider which calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation, allowing the same code to be reused indefinitely",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "For federated users, when all recovery codes are exhausted, removeStoredCredentialById() delegates to local storage instead of federated storage, causing the exhausted credential to persist and potentially creating a loop of regenerating codes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "updateCredential() stores raw recovery codes JSON in credentialData field, but the standard RecoveryAuthnCodesCredentialModel uses credentialData for metadata and secretData for hashed codes, creating inconsistent usage that could break if the format is changed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/36882": {
@@ -44144,7 +45384,27 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons and never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode. The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand, but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/keycloak/keycloak/pull/36880": {
     "claude": [
@@ -44928,6 +46188,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Potential NullPointerException: getResourceTypeResource() can return null but the result is passed directly to findByResource(server, resource) without a null check, and PolicyStore.findByResource contract states the resource parameter cannot be null",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getClientsWithPermission is broken for V2: findByType(server, \"Clients\", ...) only finds the all-clients type resource, not individual per-client resources created via getOrCreateResource which never calls resource.setType(), causing per-client permissions to be invisible",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When all-clients permission exists in getClientsWithPermission, the type resource name \"Clients\" is added to the result set, and the caller AvailableRoleMappingResource.getRoleIdsWithPermissions then calls realm.getClientById(\"Clients\") which returns null, causing an NPE on .getRolesStream()",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "canMapRoles, canMapCompositeRoles, and canMapClientScopeRoles overrides in ClientPermissionsV2 do not check the MANAGE_CLIENTS admin role, unlike canManage(ClientModel) which checks root.hasOneAdminRole(AdminRoles.MANAGE_CLIENTS) first, meaning users with MANAGE_CLIENTS realm admin role cannot map roles without explicit fine-grained permissions",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Javadoc for requireView() says 'Returns {@code true} if canView() returns true' but the method is void and should throw ForbiddenException when the check fails - this is a copy-paste error from canView() doc",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Javadoc for requireViewClientScopes() has the same copy-paste error as requireView(), incorrectly stating it returns a boolean when it's a void method that should throw ForbiddenException",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   }
 }
\ No newline at end of file
diff --git a/offline/results/anthropic_claude-opus-4-5-20251101/evaluations.json b/offline/results/anthropic_claude-opus-4-5-20251101/evaluations.json
index 1c3ad4f..9b462b5 100644
--- a/offline/results/anthropic_claude-opus-4-5-20251101/evaluations.json
+++ b/offline/results/anthropic_claude-opus-4-5-20251101/evaluations.json
@@ -1823,6 +1823,52 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37429__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37429__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The translation is in Italian instead of Lithuanian. This should be translated to Lithuanian to match the file's locale (messages_lt.properties).",
+          "severity": "Medium",
+          "matched_candidate": "Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian, affecting both account and login message files",
+          "confidence": 0.99,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment - that the translation is in Italian instead of Lithuanian. Both comments point out that the text 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian when it should be Lithuanian to match the lt (Lithuanian) locale file."
+        },
+        {
+          "golden_comment": "The totpStep1 value uses Traditional Chinese terms in the Simplified Chinese file (zh_CN), which is likely incorrect for this locale. Please verify the locale\u2011appropriate translation.",
+          "severity": "Medium",
+          "matched_candidate": "Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f), inconsistent with the rest of the file",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: Traditional Chinese characters are being used in the Simplified Chinese (zh_CN) locale file. The golden comment specifically mentions totpStep1, while the candidate mentions 'account translation' and provides specific examples of Traditional vs Simplified characters (\u624b\u6a5f vs \u624b\u673a, etc.). Both are pointing out the same fundamental problem of incorrect character set usage for the zh_CN locale."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. This allows <br>, <p>, <strong>, <b> tags in translations even when the English source only has anchor tags, defeating the stated intent"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The anchor sanitization logic has a potential issue where it consumes English matcher groups without proper validation. If the translated text has more anchor tags than the English text, this could lead to incorrect validation results.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The method name 'santizeAnchors' should be 'sanitizeAnchors' (missing 'i').",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37429__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37429__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/37634": {
@@ -3564,6 +3610,52 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37634__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37634__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Wrong parameter in null check (grantType vs. rawTokenId)",
+          "severity": "Critical",
+          "matched_candidate": "Copy-paste bug: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`, meaning `rawTokenId` is never null-checked and null values will be silently accepted, potentially causing NPEs later when the token ID is used (e.g., in `encodeTokenId()` which concatenates it)",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same bug as the golden comment - that the null check is using the wrong parameter (grantType instead of rawTokenId). The candidate provides more detail about the consequences but points to the identical underlying issue."
+        },
+        {
+          "golden_comment": "In isAccessTokenId, the substring for the grant shortcut and the equality check look inverted: the grant shortcut occupies indices 4\u20135 (substring(4,6)), and a match should return true (combined with UUID check), not false.",
+          "severity": "High",
+          "matched_candidate": "Wrong substring indices in `isAccessTokenId` matcher: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType), should be `substring(4, 6)`",
+          "confidence": 0.85,
+          "reasoning": "The candidate issue correctly identifies the same substring index problem in isAccessTokenId - that the code uses substring(3, 5) but should use substring(4, 6) to extract the grant type shortcut. Both the golden comment and candidate issue point to the same bug about incorrect substring indices for extracting the grant shortcut. The candidate doesn't mention the inverted return value (true vs false) issue that the golden comment also mentions, but it does identify the core substring index problem."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Inverted boolean logic in `isAccessTokenId` matcher: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match, the condition should be negated"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Javadoc mentions \"usually like 3-letters shortcut\" but some implementations use 2-letter shortcuts (\"ac\", \"cc\", \"rt\", \"te\", \"pc\", \"ci\", \"ro\"). Consider updating documentation to reflect actual usage pattern.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": " Catching generic RuntimeException is too broad. The implementation throws IllegalArgumentException specifically - catch that instead for more precise testing.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37634__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37634__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/38446": {
@@ -4991,6 +5083,50 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR38446__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR38446__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "After creating the RecoveryAuthnCodesCredentialModel, consider setting its id from the stored credential (e.g., myUser.recoveryCodes.getId()); otherwise getId() will be null and downstream removal by id (e.g., removeStoredCredentialById in the authenticator flow) may not work.",
+          "severity": "Low",
+          "matched_candidate": "getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID because createFromValues() does not set an ID on the model and the stored ID is never propagated, causing removeStoredCredentialById() to fail when all codes are exhausted",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: the RecoveryAuthnCodesCredentialModel is created without setting its ID from the stored credential, which causes getId() to return null and subsequently causes removeStoredCredentialById() to fail. Both comments point to the same bug - the missing propagation of the stored credential's ID to the model."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The test provider's isValid() for recovery codes accepts ANY code via anyMatch() instead of only validating against the NEXT sequential code like the production RecoveryAuthnCodesCredentialProvider.isValid() does"
+        },
+        {
+          "candidate": "isValid() never consumes/removes the used recovery code, unlike the production provider which calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation, allowing the same code to be reused indefinitely"
+        },
+        {
+          "candidate": "For federated users, when all recovery codes are exhausted, removeStoredCredentialById() delegates to local storage instead of federated storage, causing the exhausted credential to persist and potentially creating a loop of regenerating codes"
+        },
+        {
+          "candidate": "updateCredential() stores raw recovery codes JSON in credentialData field, but the standard RecoveryAuthnCodesCredentialModel uses credentialData for metadata and secretData for hashed codes, creating inconsistent usage that could break if the format is changed"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Unsafe raw List deserialization without type safety. Calling Optional.get() directly on the Optional returned by RecoveryAuthnCodesUtils.getCredential(user) without checking isPresent() can lead to a NoSuchElementException if the Optional is empty.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR38446__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR38446__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/36882": {
@@ -6109,6 +6245,39 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR36882__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR36882__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons and never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments."
+        },
+        {
+          "candidate": "The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode. The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure."
+        },
+        {
+          "candidate": "`testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand, but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Incorrect method call for exit codes. The picocli.exit() method calls System.exit() directly, which is problematic:",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR36882__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36882__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/36880": {
@@ -7692,6 +7861,54 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR36880__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR36880__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In getClientsWithPermission(String scope), iterating resourceStore.findByType(server, AdminPermissionsSchema.CLIENTS_RESOURCE_TYPE) and returning resource.getName() will only ever consider the type-level 'Clients' resource (per-client resources have no type) and return its name, while AvailableRoleMappingResource#getRoleIdsWithPermissions expects actual client IDs to pass to realm.getClientById, which can lead to incorrect behavior or a null client and subsequent failures.",
+          "severity": "High",
+          "matched_candidate": "When all-clients permission exists in getClientsWithPermission, the type resource name \"Clients\" is added to the result set, and the caller AvailableRoleMappingResource.getRoleIdsWithPermissions then calls realm.getClientById(\"Clients\") which returns null, causing an NPE on .getRolesStream()",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: the getClientsWithPermission method returns the type-level resource name 'Clients' instead of actual client IDs, and when AvailableRoleMappingResource.getRoleIdsWithPermissions calls realm.getClientById with this value, it returns null, leading to failures (NPE on .getRolesStream()). Both comments point to the same bug where the wrong resource names are being returned and causing null client lookups."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Potential NullPointerException: getResourceTypeResource() can return null but the result is passed directly to findByResource(server, resource) without a null check, and PolicyStore.findByResource contract states the resource parameter cannot be null"
+        },
+        {
+          "candidate": "canMapRoles, canMapCompositeRoles, and canMapClientScopeRoles overrides in ClientPermissionsV2 do not check the MANAGE_CLIENTS admin role, unlike canManage(ClientModel) which checks root.hasOneAdminRole(AdminRoles.MANAGE_CLIENTS) first, meaning users with MANAGE_CLIENTS realm admin role cannot map roles without explicit fine-grained permissions"
+        },
+        {
+          "candidate": "Javadoc for requireView() says 'Returns {@code true} if canView() returns true' but the method is void and should throw ForbiddenException when the check fails - this is a copy-paste error from canView() doc"
+        },
+        {
+          "candidate": "Javadoc for requireViewClientScopes() has the same copy-paste error as requireView(), incorrectly stating it returns a boolean when it's a void method that should throw ForbiddenException"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Inconsistent feature flag bug causing orphaned permissions. The AdminPermissions event listener, responsible for cleaning up permissions upon role, client, or group removal, is incorrectly guarded by the ADMIN_FINE_GRAINED_AUTHZ (V1) feature flag. This is inconsistent with other methods in the class that use ADMIN_FINE_GRAINED_AUTHZ_V2. Consequently, if ADMIN_FINE_GRAINED_AUTHZ_V2 is enabled but V1 is not, the permission cleanup logic will not execute, leading to orphaned permission data. Cleanup should occur regardless of which fine-grained authorization version is enabled.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "In hasPermission(ClientModel client, String scope), the resource lookup uses findByName(server, client.getId(), server.getId()), but AdminPermissionsSchema.getOrCreateResource creates per-client resources with the owner set to resourceServer.getClientId(), so this lookup will never find those resources and will always fall back to the 'all-clients' resource, effectively ignoring client-specific permissions.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 4,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR36880__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36880__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/37038": {
@@ -9131,6 +9348,43 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37038__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37038__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Incorrect permission check in canManage() method",
+          "severity": "High",
+          "matched_candidate": "GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, meaning a user with only VIEW permission on groups can pass canManage() checks. This gates group creation via GroupsResource.addTopLevelGroup() which calls auth.groups().requireManage(). The per-group canManage(GroupModel) correctly checks only MANAGE. This is a copy-paste from canView() where checking both scopes is correct.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue directly identifies the same problem as the golden comment - an incorrect permission check in the canManage() method. The candidate provides more detail, explaining that the no-arg canManage() method incorrectly checks for both VIEW and MANAGE scopes (copy-pasted from canView()), when it should only check for MANAGE scope. This is exactly the 'incorrect permission check in canManage() method' that the golden comment references."
+        },
+        {
+          "golden_comment": "In getGroupIdsWithViewPermission, hasPermission is called with groupResource.getId() and the same groupResource.getId() is added to granted, but hasPermission resolves resources by name (treating the argument as a group id) and the GroupPermissionEvaluator contract says this method returns group IDs that are later used as UserModel.GROUPS and in getUsersCount group filters. This mismatch means per-group VIEW_MEMBERS/MANAGE_MEMBERS permissions may not yield the expected group IDs for filtering and counts, and evaluation may effectively only look at the type-level 'all-groups' resource; consider revisiting whether this should operate on the underlying group ids (resource names) instead so it aligns with the JPA queries and the interface contract.",
+          "severity": "High",
+          "matched_candidate": "getGroupIdsWithViewPermission() uses groupResource.getId() (the authorization resource server's internal UUID) instead of groupResource.getName() (which stores the actual group model ID). This causes hasPermission() at line 122 to call resourceStore.findByName(server, <resource-internal-id>) which won't find the resource by name, so it always falls through to the 'all-groups' check, making per-group permission evaluation broken.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: getGroupIdsWithViewPermission() incorrectly uses groupResource.getId() (the authorization resource's internal UUID) when it should use groupResource.getName() (the actual group model ID). Both explain that this causes hasPermission() to fail to find resources properly because it looks up by name, resulting in the code falling back to only checking the 'all-groups' type-level resource instead of per-group permissions. The candidate accurately captures the mismatch between resource IDs and names that the golden comment describes."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In the test @BeforeEach method, ApiUtil.handleCreatedResponse(response) is called on a response that's wrapped in a try-with-resources. Since handleCreatedResponse was refactored to also use try-with-resources internally (closing the response), the response is asserted for status 201 twice and closed twice. This pattern is used at lines 73-77, 195-199, and 261-265."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 1,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37038__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37038__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/33832": {
@@ -10596,6 +10850,47 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR33832__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR33832__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Dead code exists where ASN1Encoder instances are created and written to, but their results are immediately discarded. The actual encoding is performed by new ASN1Encoder instances created in the subsequent return statement, rendering the earlier operations useless.",
+          "severity": "Low",
+          "matched_candidate": "Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: ASN1Encoder instances are created and written to, but their results are immediately discarded because the actual encoding is performed by new ASN1Encoder instances in the return statement. The candidate even specifies the exact lines (114-115 for dead code, 117-121 for actual encoding) and correctly identifies this as dead/useless code."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking."
+        },
+        {
+          "candidate": "The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures, meaning bugs specific to handling full-size ES384/ES512 r/s values would not be caught."
+        },
+        {
+          "candidate": "In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name, producing a malformed log message."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Returns wrong provider (default keystore instead of BouncyCastle)",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR33832__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR33832__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/40940": {
@@ -11933,6 +12228,41 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR40940__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR40940__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The reader thread isn\u2019t waited for; flipping deletedAll to true and asserting immediately can race and miss exceptions added just after the flag change, making this test flaky.",
+          "severity": "Medium",
+          "matched_candidate": "The reader thread is never joined before the assertion. After deletedAll.set(true), the test asserts caughtExceptions is empty, but the reader thread may still be in the middle of a groups() call. If that in-flight call throws an exception, it will be added to caughtExceptions after the assertion has already passed, causing the test to silently miss failures.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: the reader thread is not properly waited for (joined) before the assertion. They both explain that setting deletedAll to true and immediately asserting can race with the reader thread, potentially missing exceptions that are added after the flag change but before the thread completes. The candidate issue provides the same diagnosis with slightly more detail about the specific scenario (in-flight groups() call), but it's fundamentally the same race condition and flaky test concern."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "getSubGroupsStream(String, Integer, Integer) at line 256 calls modelSupplier.get().getSubGroupsStream(...) without a null check, which will throw an NPE under concurrent-deletion scenario. The same applies to the overloads at lines 262 and 268. All three methods should guard against modelSupplier.get() returning null."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Returning null from getSubGroupsCount() violates the GroupModel contract (Javadoc says it never returns null) and may lead to NPEs in callers that expect a non-null count.",
+          "severity": "Critical"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR40940__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR40940__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1": {
@@ -13313,6 +13643,46 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak-greptile__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak-greptile__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "ConditionalPasskeysEnabled() called without UserModel parameter",
+          "severity": "Medium",
+          "matched_candidate": "Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the method signature in parent class UsernamePasswordForm was changed to `isConditionalPasskeysEnabled(UserModel user)`, requiring one parameter. The call should be `isConditionalPasskeysEnabled(context.getUser())`.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - that ConditionalPasskeysEnabled() (or isConditionalPasskeysEnabled()) is being called without the required UserModel parameter. The candidate provides more detail about the specific file, the method signature change, and the suggested fix, but it's fundamentally the same issue about the missing UserModel parameter."
+        },
+        {
+          "golden_comment": "With isConditionalPasskeysEnabled(UserModel user) requiring user != null, authenticate(...) will not call webauthnAuth.fillContextForm(context) on the initial login page where context.getUser() is still null, so conditional passkey UI will not be set up for first-time passkey login. Consider whether this should also be enabled when no user has been selected yet so ID-less passkey authentication on the initial login form continues to work.",
+          "severity": "Medium",
+          "matched_candidate": "Regression: passkeys no longer set up during initial login. In `authenticate()`, the old code called `fillContextForm` when `context.getUser() == null` (initial login case), but the new code calls it only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This breaks passkey conditional UI on the initial login page.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: the new code with isConditionalPasskeysEnabled(user) requiring user != null means fillContextForm won't be called when context.getUser() is null (initial login page), breaking the conditional passkey UI setup for first-time/ID-less passkey authentication. Both comments point to the same regression in the authenticate() method logic."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null` (initial login error recovery), but the new code calls it only when `user != null`. This prevents passkey UI elements from being re-rendered on the error page after an invalid username entry."
+        },
+        {
+          "candidate": "Potential NullPointerException in `fillContextForm`: The refactored code passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. The method is `protected` and designed for override, so a future subclass could override it to return `true` without a null-user guard, causing NPE where `user` would be null."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak-greptile__tusk__PR1__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak-greptile__tusk__PR1__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/93824": {
@@ -15216,6 +15586,56 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR93824__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR93824__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Inconsistent metric tagging with 'shard' and 'shards'",
+          "severity": "Medium",
+          "matched_candidate": "Inconsistent metrics tag key: 'shards' (plural) is used here, while all other metrics in the same function use 'shard' (singular) as the tag key, causing this metric to be silently missing from dashboards or alerts that filter on the 'shard' tag",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment - the inconsistency between using 'shard' (singular) and 'shards' (plural) as metric tag keys. The candidate provides more detail about the consequences but points to the identical underlying issue."
+        },
+        {
+          "golden_comment": "Breaking out of the loop when the deadline has elapsed can skip terminating remaining flusher processes, potentially leaving them running after shutdown; consider ensuring termination is attempted even if the deadline is exceeded.",
+          "severity": "Medium",
+          "matched_candidate": "When the deadline is exceeded during join(), the break on line 341 exits the loop entirely, skipping .terminate() for all remaining processes. With multiple processes, if process 0 takes the entire timeout, processes 1..N never get terminated and will remain alive as zombies until the parent exits",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: when the deadline is exceeded during the join() operation, the break statement exits the loop, which causes remaining flusher processes to not be terminated. Both comments point out that this can leave processes running after shutdown. The candidate provides more specific details (mentioning line 341, process numbering, and zombie processes) but describes the identical underlying bug."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Each sub-buffer's get_memory_info() calls iter_cluster_memory_usage(self.client), which queries Redis INFO on ALL nodes in the cluster. Since all sub-buffers share the same Redis cluster, calling this N times makes N redundant Redis INFO roundtrips, adding unnecessary latency to each submit() call on the hot path"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Fixed sleep in tests can be flaky; wait on condition instead",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Because flusher processes are created via multiprocessing.get_context('spawn').Process, they are instances of multiprocessing.context.SpawnProcess, which on POSIX is not a subclass of multiprocessing.Process, so this isinstance check will always be false and hung processes won't be killed here.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "Sleep in test_consumer.py won\u2019t actually wait because time.sleep was monkeypatched above; consider restoring sleep or using a different sync to ensure the flusher has time to process.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 1,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR93824__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR93824__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5": {
@@ -17026,6 +17446,45 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR5__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR5__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Using zip(error_ids, events.values()) assumes the get_multi result preserves the input order; dict value order is not guaranteed to match error_ids, so event data can be paired with the wrong ID (missing nodes also shift alignment).",
+          "severity": "Low",
+          "matched_candidate": "fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list, silently pairing wrong error IDs with wrong event data",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: using zip(error_ids, events.values()) is problematic because the dictionary order is not guaranteed to match the original error_ids order. The golden comment mentions dict value order not being guaranteed and missing nodes shifting alignment. The candidate comment provides more specific detail about why the order breaks (cache_items being appended via update()), but both identify the same fundamental bug - that event data can be paired with the wrong error ID due to dict ordering assumptions."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds) are compared directly against replay segment event timestamps which use Unix epoch milliseconds, causing all error events to always appear before any replay breadcrumbs rather than being interleaved in chronological order"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Breaking changes in error response format",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Detector validator uses wrong key when updating type",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR5__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR5__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1": {
@@ -18883,6 +19342,55 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Django querysets do not support negative slicing",
+          "severity": "High",
+          "matched_candidate": "Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), queryset[start_offset:stop] will raise AssertionError('Negative indexing is not supported.'). The comment stating Django ORM handles negative slicing is incorrect",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - that Django querysets do not support negative slicing/indexing. The candidate provides more detail about how this could be triggered (via negative cursor.offset) and mentions the specific error that would be raised, but it's fundamentally the same underlying issue about Django's lack of support for negative indexing on querysets."
+        },
+        {
+          "golden_comment": "When requests are authenticated with API keys or org auth tokens (which have user_id=None), organization_context.member is None. Line 71 attempts to access organization_context.member.has_global_access without checking if member is None, causing an AttributeError crash when optimized_pagination=true is used, even though the request passed all permission checks with valid org:write scope.",
+          "severity": "High",
+          "matched_candidate": "organization_context.member can be None when the user has no membership in the org. Accessing .has_global_access on None raises AttributeError. This executes on every request to the endpoint because the enable_advanced check runs unconditionally before the use_optimized guard",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: organization_context.member can be None, and accessing .has_global_access on it will raise an AttributeError. The golden comment specifically mentions this happens with API keys/org auth tokens (user_id=None), while the candidate mentions it happens when user has no membership. Both point to the same bug at line 71 where member.has_global_access is accessed without a None check."
+        },
+        {
+          "golden_comment": "get_item_key assumes a numeric key, but the paginator is used with order_by=-datetime in the audit logs endpoint; calling math.floor/ceil on a datetime will raise a TypeError.",
+          "severity": "High",
+          "matched_candidate": "OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The method should convert datetime to float via float(value.strftime('%s.%f')) before calling math.floor(), like DateTimePaginator.get_item_key does",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil on datetime objects, which will raise a TypeError. The candidate even provides additional context about how to fix it (converting datetime to float like DateTimePaginator does), but the core issue identified is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter, producing incorrect results or a database type error"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Importing non-existent OptimizedCursorPaginator",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 4,
+      "tp": 3,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.75,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR1__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR1__20260410/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/97529": {
@@ -20252,6 +20760,44 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR97529__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR97529__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "A race condition in BuildIndex allows multiple goroutines to concurrently build the same expensive index for the same key. This is caused by moving the b.cacheMu lock from protecting the entire function to only protecting the final cache assignment. ",
+          "severity": "High",
+          "matched_candidate": "Race condition from lock scope reduction: The old code held cacheMu.Lock() for the entire BuildIndex function, serializing all concurrent calls. The new code only locks the final cache write. Two concurrent callers can simultaneously build an index for the same key, causing bleve.New(dir, mapper) to fail for file-based indexes when the directory already exists, or duplicate work and silent overwrites for memory indexes.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: a race condition in BuildIndex caused by moving the cacheMu lock from protecting the entire function to only protecting the final cache assignment. Both describe how this allows multiple goroutines to concurrently build the same index for the same key, causing duplicate work. The candidate even provides additional detail about potential consequences (bleve.New failing, silent overwrites)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Data race on shared err variable in concurrent goroutines: The err variable from stats, err := s.storage.GetResourceStats(...) is captured by reference in all group.Go closures. Multiple goroutines concurrently write to this same err via _, _, err = s.build(...) and then read it via return err, causing both a data race and correctness bug where a goroutine may return another goroutine's error or nil instead of its own."
+        },
+        {
+          "candidate": "Data race on totalBatchesIndexed: This variable is incremented from multiple goroutines launched via errgroup without any synchronization. With initWorkers > 1, concurrent increments are a data race."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Calling s.search.TotalDocs() here may race with concurrent index creation: TotalDocs iterates b.cache without synchronization, and the event watcher goroutine started just above could trigger BuildIndex writes concurrently, potentially causing a concurrent map read/write panic.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR97529__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR97529__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/80168": {
@@ -21591,6 +22137,47 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR80168__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR80168__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "MetricAlertDetectorHandler inherits from StatefulDetectorHandler but only contains pass, failing to implement its required abstract methods: counter_names (property), get_dedupe_value(), get_group_key_values(), and build_occurrence_and_event_data(). This will cause a TypeError at runtime when the class is instantiated.",
+          "severity": "High",
+          "matched_candidate": "MetricAlertDetectorHandler extends StatefulDetectorHandler but implements none of its abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), causing TypeError when instantiated",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: MetricAlertDetectorHandler extends StatefulDetectorHandler but fails to implement the required abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), which will cause a TypeError at runtime when instantiated. The wording is slightly different but the underlying issue is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In TestEvaluateGroupKeyValue.test_dedupe, expected occurrence is built with group_key='val1' but evaluate_group_key_value is called with group_key='group_key', producing mismatched fingerprints that only pass because IssueOccurrence.__eq__ compares solely by id"
+        },
+        {
+          "candidate": "In test_state_results_multi_group, expected occurrence for group_2 is built with value=6 but actual data packet has group_2=10, passing only because mock function doesn't vary output based on value and IssueOccurrence.__eq__ compares by id"
+        },
+        {
+          "candidate": "In test_state_results, detector.detector_handler creates a new handler instance each time the property is accessed, so the handler used in process_detectors differs from the one used to build expected occurrence, making the test fragile if handler-instance-specific state is ever used"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Docstring says this returns a list of DetectorEvaluationResult, but the method now returns a dict keyed by DetectorGroupKey. Consider updating the docstring to match the new return type.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR80168__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80168__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/80528": {
@@ -22861,6 +23448,37 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR80528__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR80528__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The function modifies the config variable to include display values but then returns the original monitor.config instead of the modified version.",
+          "severity": "High",
+          "matched_candidate": "In `get_monitor_environment_context`, line 168 returns `monitor_environment.monitor.config` (the original, unmodified config) instead of the local `config` variable, causing the `schedule_type` override made on line 162 to be silently discarded and the monitor context to always contain the raw `schedule_type` value instead of the human-readable display value",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the function modifies a local 'config' variable (specifically the schedule_type to a display value on line 162), but then returns 'monitor_environment.monitor.config' (the original unmodified config) instead of the modified 'config' variable. Both comments point to the same bug where modifications are lost because the wrong variable is returned."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "The code fetches MonitorCheckIn objects by ID when the required data already exists in previous_checkins. This creates an unnecessary database query.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 1,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 0,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR80528__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80528__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/77754": {
@@ -24485,6 +25103,55 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR77754__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR77754__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Shared mutable default in dataclass timestamp",
+          "severity": "Medium",
+          "matched_candidate": "Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` will share the same timestamp. Fix is `queued: datetime = field(default_factory=timezone.now)`",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment - the shared mutable default in a dataclass where timezone.now() is evaluated once at class definition time rather than at instance creation. The candidate even provides the same fix using field(default_factory=timezone.now). This is a perfect match."
+        },
+        {
+          "golden_comment": "The method name has a typo: test_from_dict_inalid_data should be test_from_dict_invalid_data.",
+          "severity": "Low",
+          "matched_candidate": "Test method name `test_from_dict_inalid_data` has a typo - missing 'v' in 'invalid'",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - the test method name 'test_from_dict_inalid_data' has a typo where 'invalid' is misspelled as 'inalid' (missing the 'v'). Both comments point to the same typo in the same method name."
+        },
+        {
+          "golden_comment": "Method name says 'empty_array' but tests empty dict - consider renaming to 'test_from_dict_empty_dict' for clarity.",
+          "severity": "Low",
+          "matched_candidate": "Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array - misleading name",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: the test method is named 'test_from_dict_empty_array' but it actually tests an empty dict ({}), not an empty array. Both suggest this is a misleading/unclear name. The golden comment suggests renaming to 'test_from_dict_empty_dict' while the candidate just identifies it as misleading, but they're pointing to the same underlying naming inconsistency problem."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `assignment_source` parameter was added to the abstract `sync_status_outbound` method but status sync is unrelated to assignee sync cycles. `ExampleIntegration.sync_status_outbound` does not accept `**kwargs`, so any caller passing `assignment_source` to it will get a TypeError"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "to_dict() returns a datetime for queued; if this dict is passed in task kwargs (e.g., via apply_async), JSON serialization may fail depending on the serializer, which can cause enqueue errors.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 4,
+      "tp": 3,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.75,
+      "recall": 0.75,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR77754__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR77754__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/95633": {
@@ -26018,6 +26685,57 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR95633__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR95633__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The test test_thread_queue_parallel_error_handling has a docstring that doesn't match the test implementation.",
+          "severity": "Low",
+          "matched_candidate": "test_thread_queue_parallel_error_handling's docstring says 'Test that errors in processing don't block offset commits for other messages' but its assertion checks the OPPOSITE: asserts no commits happened. The code always marks offsets as complete via the finally block in OrderedQueueWorker.run(), so offsets WILL be committed for both the failed and successful message. The test only passes due to timing: the commit loop waits 1 second between iterations and the test only waits 0.2s after processing completes, so the commit hasn't fired yet. This test is inherently flaky.",
+          "confidence": 0.95,
+          "reasoning": "The golden comment states that the test's docstring doesn't match the test implementation. The candidate issue identifies the same problem - it explicitly points out that the docstring says 'errors in processing don't block offset commits for other messages' but the test assertion 'checks the OPPOSITE: asserts no commits happened.' This is exactly the mismatch between docstring and implementation that the golden comment refers to. The candidate provides more detail about why this mismatch exists and why the test is flaky, but the core issue identified is the same."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "close() joins the commit thread BEFORE shutting down the worker pool. After shutdown_event is set, the commit thread does one final commit and exits. Then queue_pool.shutdown() drains remaining items from worker queues (queue.shutdown(immediate=False) lets workers process remaining items). These items are processed (side effects occur) but the commit thread is already dead, so their offsets are never committed. On restart, these items will be re-consumed and processed again, causing duplicate side effects."
+        },
+        {
+          "candidate": "The SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' and 'Only commits offsets after successful processing', but neither is true. The queues are created as queue.Queue() without a maxsize parameter, so they are unbounded and never 'fill up' - there is no backpressure mechanism. If workers are slow and the producer is fast, memory will grow without bound."
+        },
+        {
+          "candidate": "The finally block in OrderedQueueWorker.run() always calls complete_offset regardless of whether processing succeeded or raised an exception, contradicting the claim that it 'Only commits offsets after successful processing'."
+        },
+        {
+          "candidate": "OffsetTracker._get_partition_lock accesses self.partition_locks without synchronization - two threads calling it with the same new partition simultaneously will both create a Lock() but setdefault ensures only one is stored. This is safe under CPython's GIL but not guaranteed by the Python language spec."
+        },
+        {
+          "candidate": "get_committable_offsets iterates list(self.all_offsets.keys()) without holding any lock, while worker threads may be modifying all_offsets via defaultdict auto-creation. This relies on CPython GIL atomicity guarantees that are not part of the language specification."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The queue.shutdown() method with 'immediate=False' parameter may not exist in the standard Python queue module. This could cause AttributeError at runtime. Verify the correct API or implement a custom shutdown mechanism.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The magic number 50 for max_wait is used repeatedly throughout the tests. Consider extracting this as a named constant to improve maintainability.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 5,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR95633__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR95633__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2": {
@@ -27837,6 +28555,50 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR2__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR2__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "OptimizedCursorPaginator negative-offset branch slices QuerySet with a negative start index",
+          "severity": "Critical",
+          "matched_candidate": "The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing and raises AssertionError(\"Negative indexing is not supported.\"). The comment claiming Django handles negative slicing automatically is incorrect, causing 500 errors with crafted cursors",
+          "confidence": 0.99,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - that the OptimizedCursorPaginator uses negative indexing on a QuerySet (queryset[negative_offset:stop]), which Django ORM does not support. Both comments point to the same bug: negative-offset branch slicing QuerySet with a negative start index."
+        },
+        {
+          "golden_comment": "BasePaginator negative-offset branch slices QuerySet with a negative start index",
+          "severity": "High",
+          "matched_candidate": "The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing and raises AssertionError(\"Negative indexing is not supported.\"). The comment claiming Django handles negative slicing automatically is incorrect, causing 500 errors with crafted cursors",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - that BasePaginator's negative offset branch attempts to slice a QuerySet with a negative start index, which Django ORM does not support. The candidate provides additional detail about the specific error (AssertionError) and consequences (500 errors), but it's fundamentally the same underlying issue about negative indexing not being supported in Django QuerySets."
+        },
+        {
+          "golden_comment": "OptimizedCursorPaginator.get_item_key uses floor/ceil on a datetime key (order_by='-datetime'), causing TypeError.",
+          "severity": "High",
+          "matched_candidate": "OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. When used with order_by=\"-datetime\" on AuditLogEntry, getattr(item, self.key) returns a datetime that needs to be converted to float first like DateTimePaginator does via float(value.strftime(\"%s.%f\")) * self.multiplier",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: OptimizedCursorPaginator.get_item_key uses floor/ceil on a datetime key, which causes a TypeError. The candidate even provides additional context about how DateTimePaginator handles this correctly by converting to float first."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The safety check that skipped segments exceeding max_segment_spans during flush was removed. While the Lua script caps spans at 1000 during insert, the flush-side guard protected against edge cases like race conditions or spans inserted before eviction logic was deployed. The max_segment_spans attribute is now dead code"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 1,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR2__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR2__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3": {
@@ -29530,6 +30292,51 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR3__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR3__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Using Python\u2019s built-in hash() to build cache keys is non-deterministic across processes (hash randomization), so keys won\u2019t match across workers and invalidate_upsampling_cache may fail to delete them. Use a deterministic serialization of project_ids for the cache key.",
+          "severity": "Low",
+          "matched_candidate": "Python's built-in hash() is randomized per-process (via PYTHONHASHSEED), causing different Sentry worker processes to compute different hash values for the same project IDs, which defeats the caching optimization since cross-process cache lookups will always miss",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: Python's built-in hash() function is randomized per-process (PYTHONHASHSEED), which causes different worker processes to compute different hash values for the same input. Both comments point out that this defeats the caching mechanism because cache keys won't match across workers. The golden comment also mentions that invalidate_upsampling_cache may fail to delete keys, while the candidate focuses on cache lookups always missing - both are consequences of the same underlying issue with non-deterministic hash values."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The cache key doesn't include dataset or request context, so the same cache entry is reused across different query types for the same org+projects, potentially returning inconsistent results when cached True is combined with different query contexts"
+        },
+        {
+          "candidate": "The behavior was changed from any(project_id in allowlist) to all(project_id in allowlist), which could silently return incorrect underestimated data for mixed-project queries where some projects are allowlisted but not all"
+        },
+        {
+          "candidate": "The validation that 'timestamp' was not in raw_groupby was removed during refactoring, allowing users to now group by 'timestamp' which was previously explicitly disallowed and could cause unexpected query behavior"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "sample_rate = 0.0 is falsy and skipped",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The upsampling eligibility check passes the outer dataset instead of the actual dataset used by scoped_dataset. In paths where the query ultimately runs against discover (e.g., dashboard split) while the original dataset is metrics, upsampling may be skipped even when all projects are allowlisted.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 3,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR3__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR3__20260410/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/103633": {
@@ -30945,6 +31752,47 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR103633__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR103633__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The test comment says the cached permissions 'allow access', but the map stores false for dashboards:uid:dash1, so checkPermission will still treat this scope as not allowed.",
+          "severity": "Low",
+          "matched_candidate": "Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. The comment says \"Allow access to the dashboard to prove this is not checked\" but the permission cache value is set to `false`, which would also deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: the test claims to 'allow access' to prove something isn't checked, but the permission cache actually stores `false` for the dashboard scope, which means access would be denied anyway. The golden comment points out the mismatch between the comment saying 'allow access' and the map storing false. The candidate issue makes the same observation - the permission cache value is set to `false` which would deny access, making the test not actually prove what it claims."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource."
+        },
+        {
+          "candidate": "The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix overlaps with the permission cache key prefix. If a denial key accidentally matches a permission key, `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs."
+        },
+        {
+          "candidate": "After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"`, suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates cache miss vs. cache hit with stale data into the same metric label, making monitoring difficult."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The Check operation exhibits asymmetric cache trust logic: cached permission grants are trusted and returned immediately, but cached denials from the same permission cache are ignored, leading to a fresh database lookup. This allows stale cached grants to provide access to revoked resources, posing a security risk. ",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR103633__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR103633__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/67876": {
@@ -32532,6 +33380,54 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR67876__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR67876__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "OAuth state uses pipeline.signature (static) instead of a per-request random value",
+          "severity": "Medium",
+          "matched_candidate": "The OAuth state parameter used to prevent CSRF is pipeline.signature, which is a deterministic MD5 hash of the pipeline view class names. This value is identical for every user and session, allowing an attacker who knows the pipeline configuration to compute and forge the state parameter, defeating CSRF protection. OAuth state should include a cryptographic random nonce tied to the user's session.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the OAuth state parameter uses pipeline.signature (which is static/deterministic based on pipeline configuration) instead of a per-request random value. The candidate provides more detail about why this is problematic (CSRF vulnerability, same for all users) but it's fundamentally the same security concern about using a static value instead of a random per-request nonce."
+        },
+        {
+          "golden_comment": "The code attempts to access integration.metadata[sender][login] without checking for the existence of the sender key. This causes a KeyError for integrations where the sender metadata was not set during creation",
+          "severity": "High",
+          "matched_candidate": "Accessing integration.metadata[\"sender\"][\"login\"] will raise a KeyError if the sender key is not present in the integration's metadata. The sender field is only set when the integration is created via a GitHub webhook installation event, so integrations created before this code change or through other paths will cause a 500 error instead of the intended user-facing error page.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - accessing integration.metadata['sender']['login'] without checking if the 'sender' key exists, which will cause a KeyError. The candidate even provides additional context about why this happens (integrations created before the code change or through other paths), but it's fundamentally the same underlying issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "When the Integration does not exist yet (Integration.DoesNotExist), the code calls pipeline.next_step() without performing the authenticated user check. This means the first installation of a GitHub app completely bypasses the sender/user validation, allowing an attacker to install a GitHub app on a victim's Sentry org without being the GitHub user who initiated the installation."
+        },
+        {
+          "candidate": "The redirect_uri is included in the authorization request but is missing from the token exchange request. Per OAuth2 spec (RFC 6749, Section 4.1.3), if a redirect_uri was included in the authorization request, the same redirect_uri MUST be included in the access token request. GitHub may reject the token exchange or behave unexpectedly when redirect_uri is omitted."
+        },
+        {
+          "candidate": "The error() helper function receives org (which is self.active_organization) and calls org.organization, but self.active_organization can be None. If the active_organization is a different type than expected, accessing org.organization would crash."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Null reference if github_authenticated_user state is missing",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR67876__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR67876__20260410/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/32918": {
@@ -33815,6 +34711,44 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR32918__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR32918__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Cleanup reference uses incorrect alias - should be 'idp-alias-' + i instead of 'alias'.",
+          "severity": "Medium",
+          "matched_candidate": "Test cleanup registers removal for a literal \"alias\" instead of the actual IDP alias (e.g., \"idp-alias-\" + i), meaning all 20 IDPs created in the loop are never cleaned up after the test, potentially causing state leakage to other tests in the suite. The same bug exists at line 425 for idp-alias-20.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment - that the cleanup reference uses the literal string 'alias' instead of the actual IDP alias ('idp-alias-' + i). The candidate provides additional context about the consequences (IDPs not being cleaned up, state leakage) and mentions the same bug exists at line 425, but the core issue identified is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The getForLogin cached path returns results from a HashSet<IdentityProviderModel> which does not preserve any ordering from the delegate's query, causing inconsistent behavior between cached and non-cached paths for any future caller relying on delegate ordering."
+        },
+        {
+          "candidate": "In registerIDPLoginInvalidationOnUpdate, the original model is fetched via getById() which wraps it with createOrganizationAwareIdentityProviderModel overriding isEnabled(), while the updated model is raw (not wrapped). This causes unnecessary cache invalidation when comparing with getLoginPredicate() because isEnabled() may yield different results for original vs updated even if the IDP's own enabled field hasn't changed."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Recursive caching call using session instead of delegate",
+          "severity": "Critical"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR32918__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR32918__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/94942": {
@@ -35119,6 +36053,39 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR94942__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR94942__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The enableSqlExpressions function has flawed logic that always returns false, effectively disabling SQL expressions unconditionally:",
+          "severity": "Critical",
+          "matched_candidate": "The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code. Both code paths return `false`, so the convoluted logic with a meaningless feature flag check is misleading and should be simplified to just `return false`",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - that the enableSqlExpressions function always returns false regardless of the feature flag value, making the feature flag check meaningless/dead code. Both comments point out that the function effectively disables SQL expressions unconditionally due to flawed logic where both code paths return false."
+        },
+        {
+          "golden_comment": "Several methods such as NewInMemoryDB().RunCommands and db.QueryFramesInto return 'not implemented'.",
+          "severity": "High",
+          "matched_candidate": "The stub `DB.QueryFramesInto` method always returns an error ('not implemented'), but the caller in `sql_command.go:Execute` treats this error as a query failure and sets `rsp.Error = err`, then continues to build a result with an empty frame and appends it to `rsp.Values`. This is a latent issue if the guard is ever removed.",
+          "confidence": 0.75,
+          "reasoning": "The golden comment points out that several methods including NewInMemoryDB().RunCommands and db.QueryFramesInto return 'not implemented'. The candidate issue specifically identifies that DB.QueryFramesInto returns 'not implemented' error and discusses its implications. While the candidate focuses on only one of the methods mentioned in the golden comment (QueryFramesInto) and adds analysis about the caller behavior, it does identify the same underlying issue - that QueryFramesInto returns 'not implemented'. This is a partial match since the golden comment mentions multiple methods but the candidate only addresses one of them."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 0,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR94942__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR94942__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/90939": {
@@ -36367,6 +37334,37 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR90939__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR90939__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The GetWebAssets function implements an incomplete double-checked locking pattern for caching web assets. The function first checks if the cache is populated using a read lock (RLock), and if the cache is empty, it acquires a write lock to populate it. However, it fails to re-check whether the cache was populated by another goroutine while waiting to acquire the write lock.",
+          "severity": "Medium",
+          "matched_candidate": "Double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether entryPointAssetsCache was already populated by another goroutine, causing redundant file reads and a thundering herd problem at startup with many concurrent requests",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the double-checked locking pattern is incomplete because after releasing the read lock and acquiring the write lock, the code fails to re-check whether the cache was already populated by another goroutine. Both comments point to the same bug - missing the second check in the double-checked locking pattern."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "In addition to the missing double-check, the function has a critical flaw in its error handling: it unconditionally assigns the fetch result to the cache (line 69: entryPointAssetsCache = result) regardless of whether the fetch succeeded or failed. When an error occurs during asset fetching, result is nil, and this nil value overwrites any previously valid cache entry.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 1,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 0,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR90939__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90939__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/80329": {
@@ -37647,6 +38645,42 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR80329__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR80329__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The code uses Error log level for what appears to be debugging information. This will pollute error logs in production. Consider using Debug or Info level instead.",
+          "severity": "Low",
+          "matched_candidate": "All diagnostic log lines in the cleanup path use r.log.Error() instead of r.log.Debug() or r.log.Info(). These informational messages fire on every successful cleanup cycle, which will flood error-level logs and trigger monitoring/alerting systems in production.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: using Error log level for what should be Debug or Info level logging. Both comments point out that this will pollute/flood error logs in production and suggest using Debug or Info level instead. The candidate provides additional context about it being in the cleanup path and mentions monitoring/alerting concerns, but it's fundamentally the same issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The cleanup service ticker was changed from 10 minutes to 1 minute, appearing to be an accidental debug leftover. The clean() method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running, and will increase database load significantly in production."
+        },
+        {
+          "candidate": "The SQLite parameter limit guard checks r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit but should check len(ids) > sqliteParameterLimit. The guard should reflect the actual parameter count, not the configured ceiling, to avoid unnecessarily building inline SQL strings instead of using safer parameterized queries."
+        },
+        {
+          "candidate": "Test helper createTestAnnotations hardcodes annotation IDs as int64(i + 1) via InsertMulti. Since tests share a single db.InitTestDB(t) and SQLite's auto-increment counter is not reset by DELETE FROM annotation, auto-generated IDs may collide with IDs from a previous createTestAnnotations call if any test case inserts without explicit IDs."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 1,
+      "tp": 1,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR80329__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR80329__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/90045": {
@@ -39432,6 +40466,50 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR90045__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR90045__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The context is being created with d.Log instead of the log variable that was initialized with additional context values (name, kind, method). This means those values won't be propagated to the logging context.",
+          "severity": "Medium",
+          "matched_candidate": "In the Delete method, `klog.NewContext(ctx, d.Log)` uses `d.Log` (the base logger) instead of the enriched `log` variable that includes the `name`, `kind`, and `method` values.",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: in the Delete method, the code uses d.Log (the base logger) instead of the enriched 'log' variable when creating the context with klog.NewContext. Both comments point out that this means the additional context values (name, kind, method) won't be propagated to the logging context."
+        },
+        {
+          "golden_comment": "Bug: calling recordLegacyDuration when storage operation fails should be recordStorageDuration.",
+          "severity": "High",
+          "matched_candidate": "In Update's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. Same issue as Create \u2014 the storage call failed but the latency is attributed to legacy.",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same bug as the golden comment - that recordLegacyDuration is being called when a storage operation fails, but it should be recordStorageDuration instead. The candidate even provides additional context by noting it's the same issue pattern as in the Create method."
+        },
+        {
+          "golden_comment": "Inconsistency: using name instead of options.Kind for metrics recording differs from other methods.",
+          "severity": "Medium",
+          "matched_candidate": "In Delete's success path, the third argument to `recordStorageDuration` is `name` (the object name) instead of `options.Kind`. This will create high-cardinality metric labels (one time series per object name), which can cause performance issues in Prometheus.",
+          "confidence": 0.95,
+          "reasoning": "The candidate issue identifies the same problem as the golden comment - that 'name' is being used instead of 'options.Kind' in the Delete method's recordStorageDuration call. The candidate provides additional context about why this is problematic (high-cardinality metrics), but it's pointing to the exact same inconsistency/bug."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In DeleteCollection's goroutine, `recordStorageDuration` is called instead of `recordLegacyDuration` for the legacy write. This is the inverse mistake \u2014 legacy latency is recorded in the storage histogram."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 1,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR90045__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90045__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/106778": {
@@ -40977,6 +42055,49 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR106778__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR106778__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The rendered GrafanaRuleListItem is missing the required key prop for React list items. This can cause rendering issues when the list order changes.",
+          "severity": "Medium",
+          "matched_candidate": "Missing React `key` prop on GrafanaRuleListItem in the filter view's list rendering - the old GrafanaRuleLoader had `key={key}`, but the replacement GrafanaRuleListItem omits it, causing React reconciliation issues when the list updates",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the GrafanaRuleListItem component is missing the required React 'key' prop that was present in the old GrafanaRuleLoader component. Both comments point out this will cause React reconciliation/rendering issues when the list updates or order changes."
+        },
+        {
+          "golden_comment": "RuleActionsButtons is invoked with only promRule, but SilenceGrafanaRuleDrawer inside RuleActionsButtons still depends on a Grafana Ruler rule being present, so for Grafana rules coming from list views the 'Silence notifications' menu entry (now driven by Grafana Prom abilities) will toggle showSilenceDrawer without ever rendering the drawer. This means clicking 'Silence notifications' for these rules has no visible effect, even when abilities indicate silencing is allowed.",
+          "severity": "High",
+          "matched_candidate": "Silence drawer won't open in the new list view because the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` which is false when rule (RulerRuleDTO) is undefined from GrafanaRuleListItem",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: the silence drawer won't render/open properly because SilenceGrafanaRuleDrawer depends on a Grafana Ruler rule being present, but in the list view context only promRule is passed (not the full RulerRuleDTO). The golden comment explains that clicking 'Silence notifications' will toggle showSilenceDrawer without rendering the drawer, and the candidate identifies the same problem - the drawer rendering is guarded by a check that fails when the ruler rule is undefined. They're describing the exact same bug from slightly different angles."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Delete menu item will never render in the new list view because AlertRuleMenu guards the delete menu item with `canDelete && rulerRule && (...)`, but rulerRule is always undefined when coming from GrafanaRuleListItem since the new list view no longer fetches ruler data"
+        },
+        {
+          "candidate": "Pause menu item will never render in the new list view because `canPause && rulerRuleType.grafana.rule(rulerRule)` requires rulerRule to be a grafana ruler rule, but rulerRule is undefined when coming from the new list view"
+        },
+        {
+          "candidate": "Type mismatch in useGrafanaPromRuleAbility hook - called with promRule typed as Rule but hook expects GrafanaPromRuleDTO | SkipToken, and prometheusRuleType.grafana.rule() returns boolean not a type predicate, potentially causing incorrect permission checks due to missing folderUid field"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR106778__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR106778__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/107534": {
@@ -42115,6 +43236,39 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR107534__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR107534__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Double interpolation of queries in the shard splitting path: runShardSplitQuery() calls applyTemplateVariables() at line 52, then runSplitQuery() also calls applyTemplateVariables() at line 299, causing ad-hoc filters to be added to the query expression twice since addAdHocFilters() unconditionally appends label matchers"
+        },
+        {
+          "candidate": "Test mock for applyTemplateVariables mutates the input query object directly instead of returning a new object as the real implementation does, hiding potential bugs where code relies on the original targets remaining unmodified"
+        },
+        {
+          "candidate": "Filter ordering change in runShardSplitQuery may cause behavioral difference: the new code filters before interpolation, so a query with expr as a template variable that resolves to empty string will pass the filter but result in an empty expression being sent to splitQueriesByStreamShard"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The applyTemplateVariables method is called with request.filters as the third parameter, but this parameter is not used in the corresponding test setup.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR107534__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR107534__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/79265": {
@@ -44101,6 +45255,59 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR79265__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR79265__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Race condition: Multiple concurrent requests could pass the device count check simultaneously and create devices beyond the limit. Consider using a database transaction or lock.",
+          "severity": "High",
+          "matched_candidate": "TOCTOU race in device limit enforcement. CreateOrUpdateDevice first counts devices, then conditionally calls updateDevice. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same race condition problem as the golden comment. Both describe: 1) Multiple concurrent requests can pass the device count check simultaneously, 2) This allows creating devices beyond the configured limit, 3) The issue is a TOCTOU (Time-of-Check-Time-of-Use) race condition between counting devices and creating/updating them. The candidate even provides more technical detail by naming it as a 'TOCTOU race' and explaining the specific flow (count check then update/insert). Both suggest the same underlying concern about concurrent access to the device limit enforcement."
+        },
+        {
+          "golden_comment": "Returning ErrDeviceLimitReached when no rows were updated is misleading; the device might not exist.",
+          "severity": "Low",
+          "matched_candidate": "The updateDevice method returns ErrDeviceLimitReached when rowsAffected == 0, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its updated_at falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with ErrDeviceLimitReached instead of being re-created or handled differently.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: returning ErrDeviceLimitReached when rowsAffected == 0 is misleading because it conflates multiple scenarios. The golden comment says the device 'might not exist', while the candidate expands on this by identifying two distinct cases where rows wouldn't be affected (device doesn't exist, or device exists but falls outside the time range). Both are pointing to the same fundamental problem - that ErrDeviceLimitReached is returned inappropriately when the actual cause could be something other than hitting the device limit."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout."
+        },
+        {
+          "candidate": "The AnonymousDeviceLimit field in FrontendSettingsDTO is int64 with json:\"anonymousDeviceLimit\" \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as number | undefined. The Go DTO has no omitempty tag, so the field is never undefined in practice. Any frontend code checking config.anonymousDeviceLimit === undefined to detect 'no limit configured' will never match; it will see 0 instead."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Anonymous authentication now fails entirely if anonDeviceService.TagDevice returns ErrDeviceLimitReached. Previously, device tagging was asynchronous and non-blocking. This change prevents anonymous users from authenticating when the device limit is reached.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "This call won\u2019t compile: dbSession.Exec(args...) is given a []interface{} where the first element is the query, but Exec\u2019s signature requires a first parameter of type string (not an interface{} splat).",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Time window calculation inconsistency: Using device.UpdatedAt.UTC().Add(-anonymousDeviceExpiration) as the lower bound but device.UpdatedAt as the current time may not match the intended logic. Consider using time.Now().UTC() consistently.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 2,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR79265__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR79265__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9": {
@@ -45459,6 +46666,44 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR9__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR9__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Thread-safety issue with lazy @loaded_locales",
+          "severity": "Low",
+          "matched_candidate": "The ensure_loaded! method initializes @loaded_locales ||= [] outside the LOAD_MUTEX, causing potential race condition in multi-threaded servers where two threads could overwrite each other's array, leading to redundant load_locale calls or skipped locale loading",
+          "confidence": 0.95,
+          "reasoning": "The candidate issue identifies the exact same thread-safety problem as the golden comment. Both point out that @loaded_locales has a race condition/thread-safety issue. The candidate specifically explains that '@loaded_locales ||= []' is initialized outside the LOAD_MUTEX, which is the lazy initialization thread-safety problem the golden comment refers to. The candidate provides more detail about the consequences (race condition, overwriting arrays, redundant loads) but it's fundamentally the same underlying issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "LRU translation cache key does not include SiteSetting.default_locale, causing stale cached translations to be served when admin changes the site's default locale since there is no I18n.reload! triggered on default_locale change"
+        },
+        {
+          "candidate": "The translate method only loads config.locale but with I18n::Backend::Fallbacks the backend iterates through the fallback chain - if fallback locale YML files haven't been loaded, lookups return nil and fallbacks silently fail in code paths outside controllers (Sidekiq jobs, rake tasks, console)"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Consider normalizing the input locale (e.g., to a symbol) when checking/loading here to avoid double-loading if the same locale is passed as a String vs Symbol (also applies to other locations in the PR).",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR9__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR9__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/76186": {
@@ -46771,6 +48016,41 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR76186__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR76186__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
+          "severity": "High",
+          "matched_candidate": "ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic if the middleware is used outside the Decorator (e.g., in tests), unlike LoggerMiddleware which has explicit nil request guard clauses.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without checking if req is nil first, which would cause a nil pointer dereference panic. Both comments point out that this is different from other middleware (LoggerMiddleware/previous middleware layers) that properly handle nil requests. The candidate even mentions the same methods and the same root cause (accessing req.PluginContext without nil-checking req)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(), causing any test using TestLogger to verify logged output to silently lose all log messages since the .Info() call goes to a throwaway instance. The method should return the receiver (f) to preserve log capture."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR76186__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR76186__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10": {
@@ -48898,6 +50178,61 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR10__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR10__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "NoMethodError before_validation in EmbeddableHost",
+          "severity": "Critical",
+          "matched_candidate": "`before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value, `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment - a NoMethodError occurring in the before_validation callback of EmbeddableHost when the host attribute is nil. The candidate provides more detail about the specific error (undefined method 'sub!' for nil:NilClass) but it's clearly the same underlying issue."
+        },
+        {
+          "golden_comment": "The update and destroy methods in Admin::EmbeddableHostsController do not validate the existence of the EmbeddableHost record retrieved by ID. If EmbeddableHost.where(id: params[:id]).first returns nil (i.e., the host does not exist), attempting to call methods on the nil object (e.g., save_host or destroy) will result in a NoMethodError.",
+          "severity": "Medium",
+          "matched_candidate": "Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist. In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: both point out that the update and destroy methods in Admin::EmbeddableHostsController don't validate the existence of the EmbeddableHost record, and that when EmbeddableHost.where(id: params[:id]).first returns nil, calling methods like destroy or save_host on the nil object will result in a NoMethodError. The candidate even provides the same specific details about which methods will crash."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist, the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached."
+        },
+        {
+          "candidate": "SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote will break the SQL or allow injection."
+        },
+        {
+          "candidate": "Fabricator file contents are swapped. `category_fabricator.rb` contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators. The file names don't match their contents."
+        },
+        {
+          "candidate": "JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "record_for_host compares lower(host) = ? but does not normalize the parameter\u2019s case, so mixed\u2011case referer hosts may fail to match even though comparison intends to be case\u2011insensitive.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Because this migration inserts embeddable_hosts rows with raw SQL, any existing embeddable_hosts values that include http:// or /https:// or path segments won\u2019t go through the EmbeddableHost model\u2019s normalization, so the new host lookup (which compares only the bare host) may fail for migrated data. Consider ensuring that migrated hosts are normalized to the same format as newly created EmbeddableHost records so existing embedding configurations keep working.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 4,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR10__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR10__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7": {
@@ -50540,6 +51875,53 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR7__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR7__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In .topic-meta-data h5 a, the original code had color: scale-color($primary, $lightness: 30%) but was changed to dark-light-choose(scale-color($primary, $lightness: 70%), scale-color($secondary, $lightness: 30%)). The lightness for the light theme changed from 30% to 70%, which is a dramatic inversion",
+          "severity": "Low",
+          "matched_candidate": "Light-theme lightness value changed from 30% to 70% in desktop/topic-post.scss for .embedded-posts link text - values appear to be accidentally swapped with dark theme",
+          "confidence": 0.85,
+          "reasoning": "The candidate issue identifies the same problem - a lightness value change from 30% to 70% in the light theme. While the candidate mentions '.embedded-posts link text' and the golden mentions '.topic-meta-data h5 a', both are pointing to the same underlying issue: the dramatic inversion of lightness values (30% to 70%) for the light theme, suggesting values were accidentally swapped. The core bug identified is the same."
+        },
+        {
+          "golden_comment": "This change for desktop/user.css changes $primary from 30% to 50% for the light theme; most other changes preserve the original $primary value and move the complement to $secondary for dark. Consider reviewing this (also applies to a similar .name change in the mobile variant).",
+          "severity": "Low",
+          "matched_candidate": "Light-theme lightness for .name changed from 30% to 50% in both desktop and mobile user stylesheets - makes user names more washed out, inconsistent with nearby .username a in common/base/user.scss which correctly preserves 30%",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: the light-theme lightness for .name was changed from 30% to 50% in the desktop/user.css (and mobile variant). The golden comment notes this change and suggests reviewing it, while the candidate issue explicitly calls out this same change in both desktop and mobile stylesheets, noting it makes user names more washed out and is inconsistent with nearby code. They're pointing to the exact same underlying concern about the 30% to 50% lightness change."
+        },
+        {
+          "golden_comment": "In topic-post.css the original code used $lightness: 70% but the replacement uses $lightness: 30% for the light theme. This makes the text significantly darker than intended.",
+          "severity": "Low",
+          "matched_candidate": "Light-theme lightness value changed from 30% to 70% in desktop/topic-post.scss for .embedded-posts link text - values appear to be accidentally swapped with dark theme",
+          "confidence": 0.9,
+          "reasoning": "The candidate identifies the same issue - a lightness value swap between light and dark themes in topic-post.scss. The candidate says '30% to 70%' while the golden says '70% to 30%', but they're describing the same problem from opposite perspectives (one describes what the code shows now vs what it should be, the other describes the change direction). Both identify that the light theme has the wrong lightness value due to an accidental swap with dark theme values."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Light-theme lightness value changed from 70% to 30% in mobile/modal.scss for .custom-message-length - opposite of desktop version, making character count text inverted between light and dark themes on mobile"
+        },
+        {
+          "candidate": "Light-theme lightness for h3 changed from 20% to 50% in mobile/topic-post.scss - makes headings significantly more faded/grey on light themes, dark-theme value should have been 80% to complement 20%"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.75,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR7__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR7__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8": {
@@ -52334,6 +53716,57 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR8__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR8__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In the next action, capping the next offset at user_count can produce an empty page (e.g., total equal to limit results in offset == total, showing 2/2 with no members). This can cause confusing UX on the last page.",
+          "severity": "Medium",
+          "matched_candidate": "Off-by-one in totalPages calculation - Math.floor(user_count / limit) + 1 yields one extra page when user_count is exactly divisible by limit (e.g., 100 users with limit 50 gives 3 pages instead of 2), should use Math.ceil(user_count / limit)",
+          "confidence": 0.7,
+          "reasoning": "The golden comment is about the next offset calculation causing an empty last page when offset equals total (e.g., when total equals limit). The candidate issue is about totalPages calculation using Math.floor + 1 instead of Math.ceil, causing an extra page when user_count is exactly divisible by limit. Both issues relate to pagination edge cases that result in an extra/empty page, but they point to different specific calculations - the golden focuses on the 'next' offset capping logic while the candidate focuses on the totalPages formula. However, both describe the same underlying symptom: getting an extra/empty page in certain divisibility conditions. These are closely related pagination bugs that could stem from the same root cause."
+        },
+        {
+          "golden_comment": "HTTP method mismatch in .remove_member - test uses PUT but remove_member action expects DELETE",
+          "severity": "Medium",
+          "matched_candidate": "Test for 'cannot remove members from automatic groups' uses xhr :put but the route is defined as delete, so the test doesn't validate the actual HTTP method and route misconfiguration bugs would go undetected",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment - that the test uses PUT method but the remove_member action expects DELETE. The candidate specifically mentions 'Test for cannot remove members from automatic groups uses xhr :put but the route is defined as delete', which is precisely the HTTP method mismatch issue described in the golden comment."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Breaking change in the public GroupsController#members endpoint response format - old code returned a flat JSON array, new code returns { members: [...], meta: { total, limit, offset } }, breaking existing consumers expecting a flat array"
+        },
+        {
+          "candidate": "GroupsController#members endpoint now always paginates with default limit of 50, whereas old code had no limit/offset for non-automatic groups, causing large groups to silently return truncated results to existing callers that don't pass pagination params"
+        },
+        {
+          "candidate": "In update action, group.visible = params[:visible] == 'true' is executed unconditionally, so if a caller sends only partial params without visible parameter, visible gets set to false, causing partial PUT requests from API consumers to reset visibility"
+        },
+        {
+          "candidate": "New groups will always be created with visible: false regardless of checkbox state because asJSON() sends visible: !!this.get('visible') which evaluates false for a new group where visible is undefined/null, and the visible property is never initialized to true in the Group model defaults"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": " The findMembers() call is now asynchronous and unhandled. The controller may not have member data immediately available, creating a race condition.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR8__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR8__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3": {
@@ -53720,6 +55153,44 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR3__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR3__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "BlockedEmail.should_block_email? method has side effects during a read operation - it updates statistics even when just checking if an email should be blocked. This could cause race conditions in concurrent environments and makes the method name misleading.",
+          "severity": "Medium",
+          "matched_candidate": "BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation, which can run multiple times per user action, inflating blocked-email statistics",
+          "confidence": 0.9,
+          "reasoning": "Both comments identify the same core issue: the BlockedEmail.should_block? method has a side effect of performing database writes (updating match_count and last_match_at statistics) during what should be a read/check operation. The golden comment emphasizes the misleading method name and race conditions, while the candidate emphasizes the inflation of statistics due to multiple validation runs. Both are pointing to the same fundamental problem - a query/check method that unexpectedly modifies data."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The JS code adds email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures, causing valid emails belonging to other users to be permanently marked as rejected in the client"
+        },
+        {
+          "candidate": "rejectedEmails is initialized as Em.A([]) directly in the class definition hash passed to .extend(), causing the array to be shared across all instances as a prototype property instead of instance property"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Regex pattern @(#{domains}) only matches domain suffixes, not full domains. evil.example.com would match whitelist entry example.com.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR3__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR3__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5": {
@@ -55110,6 +56581,52 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR5__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR5__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Mixing float: left with flexbox causes layout issues. Further this PR removes the float-based right alignment for .d-header .panel, which may cause the login panel in the non-Ember/noscript header (where .panel is nested inside .row and not a flex item) to stack under the title instead of remaining right-aligned.",
+          "severity": "Low",
+          "matched_candidate": "In `.small-action`, the parent is now `display: flex` but child `.topic-avatar` still has `float: left`, indicating an incomplete migration - the float should be removed for clarity",
+          "confidence": 0.75,
+          "reasoning": "Both comments identify the same underlying issue: mixing float properties with flexbox layout. The golden comment discusses float: left with flexbox causing layout issues in the header panel context, while the candidate identifies the same pattern (float: left with display: flex parent) in a different location (.small-action and .topic-avatar). They both point to the same type of bug - incomplete migration from float-based to flexbox layout - but in different specific locations in the code. The core concern is identical (float + flexbox mixing), though the specific elements mentioned differ."
+        },
+        {
+          "golden_comment": "-ms-align-items never existed in any version of IE/Edge; the correct legacy property is -ms-flex-align.",
+          "severity": "Low",
+          "matched_candidate": "`-ms-align-items` is not a valid CSS property and will be silently ignored by IE10/11 - the correct IE prefix for `align-items` is only `-ms-flex-align`",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: the use of `-ms-align-items` is incorrect/invalid, and the correct property for IE should be `-ms-flex-align`. The golden comment states it 'never existed' while the candidate says it's 'not a valid CSS property' - both convey the same meaning that this prefix is wrong and will not work in IE."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `order` mixin passes the raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values while the standard `order` property is 0-based. The correct value for `-webkit-box-ordinal-group` should be `$val + 1`, otherwise elements using `@include order(1)` would be indistinguishable from unordered elements in old WebKit browsers"
+        },
+        {
+          "candidate": "The padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that created spacing from the avatar, causing the gap between avatar and description text to shrink significantly on narrow viewports and may overlap or appear too tight"
+        },
+        {
+          "candidate": "`-ms-flex-align` uses different value keywords (`start`/`end` instead of `flex-start`/`flex-end`), so the mixin does not translate values and future callers passing `flex-start` or `flex-end` would silently break in IE"
+        },
+        {
+          "candidate": "The `.badge-wrapper.bullet` rule had `margin-top: 5px` removed in the header's `.extra-info-wrapper` context but the same selector in `#topic-title .title-wrapper` retains `margin-top: 5px`, creating inconsistent bullet badge styling between contexts"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 4,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR5__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR5__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6": {
@@ -56324,6 +57841,39 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR6__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR6__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The include_website_name method is missing the required ? suffix. Rails serializers expect include_ methods to end with ? for conditional attribute inclusion, a convention followed by other methods in this serializer. Without it, the website_name attribute may not be conditionally included as intended. Additionally, the '.' << website_host string concatenation should be replaced with '.' + website_host or '.#{website_host}' to avoid mutating string literals, which can lead to issues.",
+          "severity": "Medium",
+          "matched_candidate": "Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized even when `website` is nil/blank.",
+          "confidence": 0.85,
+          "reasoning": "The candidate issue correctly identifies the same primary problem as the golden comment - the missing '?' suffix on the include_website_name method. Both comments explain that Rails/ActiveModel::Serializer expects include_ methods to end with '?' for conditional attribute inclusion, and that without it, the conditional logic won't work as intended. However, the candidate issue does not mention the secondary concern about string concatenation ('.' << website_host) that the golden comment also raises. Despite missing this secondary issue, the candidate does identify the main underlying problem."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "`website_name` is not included in the `untrusted_attributes` list but `website` is. Since `website_name` bypasses the `restrict_user_fields?` check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden."
+        },
+        {
+          "candidate": "The `Discourse.stubs(:current_hostname)` call happens inside the `it` block after `json` is already defined as a lazy let. If `json` were ever evaluated before the stub, the stub would not be in effect. The stub should be in a `before` block before `json` is accessed."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 1,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR6__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR6__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4": {
@@ -58853,6 +60403,66 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR4__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR4__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The current origin validation using indexOf is insufficient and can be bypassed. An attacker could use a malicious domain like evil-discourseUrl.com to pass this check.",
+          "severity": "Medium",
+          "matched_candidate": "The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a substring of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same security vulnerability: the origin validation using indexOf is insufficient and can be bypassed. The golden comment mentions an attacker using 'evil-discourseUrl.com' while the candidate mentions 'a domain that is a prefix of the legitimate domain'. Both are describing the same substring matching vulnerability in postMessage origin validation, just with slightly different attack vector descriptions (suffix vs prefix). The core issue is identical - indexOf-based origin checking is insecure and can be bypassed by malicious domains."
+        },
+        {
+          "golden_comment": "The ERB block closes with end if, which is invalid Ruby/ERB and will raise at render; it should just be end to close the if block.",
+          "severity": "Medium",
+          "matched_candidate": "ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression. This will raise a SyntaxError when the best.html.erb template is rendered.",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the ERB block incorrectly uses 'end if' instead of just 'end' to close the if block, which is invalid Ruby/ERB syntax and will cause a SyntaxError at render time. Both comments point to the same bug with the same explanation and solution."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped."
+        },
+        {
+          "candidate": "In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`."
+        },
+        {
+          "candidate": "`contents << \"\\n<hr>...\"` uses the `<<` operator which mutates the string in place, modifying the caller's original string. If `import` is called twice with the same string, the imported-from footer is appended multiple times. When `import` is called for an update path, the SHA1 was computed after mutation, so re-importing the same content will always differ from the stored SHA1."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "SSRF vulnerability using open(url) without validation",
+          "severity": "Critical"
+        },
+        {
+          "golden_comment": "postMessage targetOrigin should be the origin (scheme+host+port), not the full referrer URL; using the full URL will cause the message to be dropped and prevent resizing.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "The code sets X-Frame-Options: ALLOWALL which completely disables clickjacking protection. The referer validation can be bypassed (referer headers are easily spoofed), and the fallback to empty string for nil referer masks validation failures.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "The TopicEmbed.import method is susceptible to a NoMethodError if the contents parameter is nil when attempting to append a string, and an XSS vulnerability due to unescaped url interpolation in the generated HTML.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 6,
+      "tp": 2,
+      "fp": 3,
+      "fn": 4,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR4__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR4__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1": {
@@ -60506,6 +62116,51 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The downsize method is defined twice. The second definition, which expects a single dimensions string parameter, overrides the first, which expected separate max_width and max_height parameters. This makes the original method unreachable and breaks existing code that calls it with separate width and height arguments.",
+          "severity": "Medium",
+          "matched_candidate": "Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same core issue: the downsize method is defined twice in Ruby, and the second definition overrides/replaces the first, making the original method unreachable. Both note this will break existing code that calls the method with the original parameter signature. The candidate provides more specific details (line numbers, exact argument counts, specific caller location) but identifies the exact same underlying bug."
+        },
+        {
+          "golden_comment": "Hardcoding maxSizeKB = 10 * 1024 ignores Discourse.SiteSettings['max_' + type + '_size_kb'], so the client-side limit can diverge from server-side and per-type settings (also applies to the 413 handler below).",
+          "severity": "Low",
+          "matched_candidate": "The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: the hardcoded maxSizeKB = 10 * 1024 ignores the dynamic Discourse.SiteSettings['max_' + type + '_size_kb'] lookup, causing client-side limits to diverge from server-side and per-type settings. The candidate provides more detail about the implications but addresses the same core issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The 413 error handler also hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message."
+        },
+        {
+          "candidate": "The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Passing 80% as the dimensions can fail for animated GIFs when allow_animated_thumbnails is true, since the animated path uses gifsicle --resize-fit which expects WxH geometry, not a percentage; downsizing would then silently fail.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR1__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR1__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2": {
@@ -62016,6 +63671,50 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR2__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR2__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "logic: Potential nil pointer exception - if no TopicUser record exists, tu will be nil and calling methods on it will crash",
+          "severity": "High",
+          "matched_candidate": "NullPointerError crash: TopicUser.find_by returns nil when no TopicUser record exists, and the next line calls tu.notification_level on the nil result, raising NoMethodError",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: TopicUser.find_by can return nil when no record exists, and subsequently calling methods on the nil result (tu.notification_level) will cause a crash/error. The golden comment calls it a 'nil pointer exception' while the candidate calls it 'NullPointerError crash' and 'NoMethodError', but they're describing the same underlying bug."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The unsubscribe_link i18n string requires an %{unsubscribe_url} interpolation variable, but existing email paths using add_unsubscribe_link: true without passing unsubscribe_url will raise I18n::MissingInterpolationArgument error or render a broken link"
+        },
+        {
+          "candidate": "XSS vulnerability: title value is injected directly into HTML via string concatenation without escaping, allowing user-controlled content with script tags to be rendered as raw HTML"
+        },
+        {
+          "candidate": "XSS vulnerability in unsubscribe template: triple-stache {{{ }}} renders raw HTML and the computed property interpolates model.fancyTitle which can contain malicious HTML that would be rendered unescaped"
+        },
+        {
+          "candidate": "The html_part method only substitutes %{respond_instructions} and %{unsubscribe_link} placeholders when add_unsubscribe_link is true, but the template always contains these placeholders, causing literal placeholder strings to appear in emails when add_unsubscribe_link is false"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Typo in property name: 'stopNotificiationsText' should be 'stopNotificationsText' (missing 'n' in 'Notifications')",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR2__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR2__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/22532": {
@@ -63591,6 +65290,53 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR22532__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR22532__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The updateManyByCredentialId call uses an empty data object, which prevents Prisma's @updatedAt decorator from updating the updatedAt timestamp. This results in inaccurate cache status tracking, as the timestamp isn't updated when the cache is refreshed. To fix this, explicitly set the updatedAt field.",
+          "severity": "Medium",
+          "matched_candidate": "Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: calling updateManyByCredentialId with an empty data object ({}) fails to update the updatedAt timestamp. The golden comment says the @updatedAt decorator won't work with empty data, while the candidate provides more technical detail explaining that updateMany doesn't trigger @updatedAt (only update() and upsert() do). Both point to the same bug - the updatedAt timestamp won't be updated when the cache is refreshed, leading to inaccurate tracking."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `deleteCache` handler only checks `userId: user.id` on the credential but does not account for delegation credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache."
+        },
+        {
+          "candidate": "The `deleteCache` handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials."
+        },
+        {
+          "candidate": "For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries."
+        },
+        {
+          "candidate": "Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` is semantically incorrect \u2014 `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. `?? null` would be more semantically correct."
+        },
+        {
+          "candidate": "The `CredentialActionsDropdown` component initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them, even when the component returns null."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "logic: macOS-specific sed syntax with empty string after -i flag will fail on Linux systems",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR22532__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22532__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/8330": {
@@ -65080,6 +66826,46 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR8330__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR8330__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Incorrect end time calculation using slotStartTime instead of slotEndTime",
+          "severity": "Medium",
+          "matched_candidate": "Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue directly identifies the same bug as the golden comment - that the end time calculation incorrectly uses slotStartTime instead of slotEndTime. The candidate provides additional detail about the consequences of this bug, but the core issue identified is exactly the same."
+        },
+        {
+          "golden_comment": "Using === for dayjs object comparison will always return false as it compares object references, not values. Use .isSame() method instead: dayjs(date.start).add(utcOffset, 'minutes').isSame(dayjs(date.end).add(utcOffset, minutes))",
+          "severity": "Medium",
+          "matched_candidate": "Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: using === to compare dayjs objects compares references instead of values, which will always return false. Both comments point to the same bug where dayjs object comparison needs to use .isSame() or equivalent value comparison instead of reference equality."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC."
+        },
+        {
+          "candidate": "The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range, it prevents checking whether the slot is valid for a different user's override. The combination of mutation and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR8330__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8330__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/14943": {
@@ -66411,6 +68197,33 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR14943__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR14943__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "Using retryCount: reminder.retryCount + 1 reads a possibly stale value and can lose increments under concurrency; consider an atomic increment via Prisma (increment: 1) to avoid race conditions (also applies to the similar update in the catch block).",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The deletion logic in scheduleSMSReminders.ts incorrectly deletes non-SMS workflow reminders (e.g., Email, WhatsApp) that have retryCount > 1. This occurs because the retryCount condition within the OR clause for deletion lacks a method: WorkflowMethods.SMS filter, causing it to apply to all reminder types instead of only SMS reminders, which is the intended scope of this function.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 0,
+      "total_golden": 2,
+      "tp": 0,
+      "fp": 0,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR14943__20260411",
+      "pr_url": null
     }
   },
   "https://github.com/calcom/cal.com/pull/22345": {
@@ -67716,6 +69529,50 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR22345__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR22345__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In getBaseConditions(), the else if (filterConditions) and final else branches are unreachable. This is because getAuthorizationConditions() always returns a non-null Prisma.Sql object, making authConditions always truthy, which means only the first two if/else if conditions are ever evaluated.",
+          "severity": "Low",
+          "matched_candidate": "getBaseConditions method has unreachable dead code branches because getAuthorizationConditions() always returns a truthy Prisma.Sql object, and auth failures don't short-circuit filter condition computation",
+          "confidence": 0.95,
+          "reasoning": "The candidate issue identifies the same problem as the golden comment - that getBaseConditions() has unreachable/dead code branches because getAuthorizationConditions() always returns a truthy Prisma.Sql object. Both comments point to the same underlying bug where certain else branches can never be executed due to authConditions always being truthy."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Public methods getAuthorizationConditions() and getFilterConditions() return Prisma.Sql fragments containing OR operators without outer parentheses, causing SQL operator precedence bugs when composed with AND in raw queries, potentially leading to privilege escalation"
+        },
+        {
+          "candidate": "Constructor parameter type changed from InsightsBookingServiceOptions (discriminated union requiring teamId for scope='team') to InsightsBookingServicePublicOptions (flat type with optional teamId), removing compile-time checking when teamId is missing for team scope"
+        },
+        {
+          "candidate": "Falsy guard 'if (this.filters.eventTypeId)' skips the filter when eventTypeId is 0, and similarly 'if (this.filters.memberUserId)' skips memberUserId 0, causing silent unfiltered results instead of filtering by ID=0"
+        },
+        {
+          "candidate": "Caching test suite was removed but caching logic still exists in the service, leaving no tests to catch regressions that could cause repeated database queries"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Fetching userIdsFromOrg only when teamsFromOrg.length > 0 can exclude org-level members for orgs without child teams; consider deriving from teamIds (which includes orgId) or removing the guard so org-only orgs still include member user bookings.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR22345__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22345__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/11059": {
@@ -70137,6 +71994,65 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR11059__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR11059__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Invalid Zod schema syntax. Computed property keys like [z.string().toString()] are not valid in Zod object schemas and will cause runtime errors. ",
+          "severity": "High",
+          "matched_candidate": "minimumTokenResponseSchema uses [z.string().toString()] as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field exists in the response, making the 'minimum token validation' ineffective.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: the invalid use of [z.string().toString()] as a computed property key in a Zod schema. The golden comment states it's 'Invalid Zod schema syntax' that will cause 'runtime errors'. The candidate comment explains the same problem in more detail - that it evaluates to a literal string like 'ZodString' at definition time rather than working as a dynamic key validator, making the validation ineffective. Both are pointing to the exact same bug with the computed property key syntax in the Zod schema."
+        },
+        {
+          "golden_comment": "parseRefreshTokenResponse returns a Zod safeParse result ({ success, data, error }), not the credential key object. Persisting that as key stores the wrapper instead of the token payload; we should store the parsed data or use schema parse.",
+          "severity": "High",
+          "matched_candidate": "parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously googleCredentialSchema.parse() returned just the data. Now the DB will store {success: true, data: {access_token, refresh_token, expiry_date}} instead of {access_token, refresh_token, expiry_date}, corrupting the credential for all subsequent reads.",
+          "confidence": 0.99,
+          "reasoning": "Both comments identify the exact same issue: parseRefreshTokenResponse returns a Zod safeParse result object (containing {success, data, error}) instead of just the parsed data. Both explain that this causes the credential key to be stored incorrectly - storing the wrapper object instead of the actual token payload. The candidate even provides more specific detail about the corruption impact."
+        },
+        {
+          "golden_comment": "When APP_CREDENTIAL_SHARING_ENABLED and CALCOM_CREDENTIAL_SYNC_ENDPOINT are set, the refreshFunction helper returns the fetch Response, but several callers (for example GoogleCalendarService.refreshAccessToken expecting res.data, and HubspotCalendarService.refreshAccessToken expecting a HubspotToken) assume it returns the integration-specific token object. That mismatch will cause runtime errors in the sync-enabled path unless the return type or those call sites are adjusted.",
+          "severity": "High",
+          "matched_candidate": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses res?.data (undefined on fetch Response \u2192 TypeError on token.access_token), Hubspot accesses .expiryDate/.expiresIn (undefined on Response), Zoho-bigin and Zohocrm access .data.error (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: when APP_CREDENTIAL_SHARING_ENABLED is true, the refreshOAuthTokens/refreshFunction returns a raw fetch Response object, but callers (Google Calendar, Hubspot, and others) expect integration-specific token objects with properties like res.data, .expiryDate, etc. This type mismatch will cause runtime errors. The candidate issue actually provides more specific examples (Zoho-bigin, Zohocrm) but covers the same fundamental problem."
+        },
+        {
+          "golden_comment": "When the sync endpoint path is used, res is a fetch Response and has no .data; res?.data will be undefined and token.access_token will throw at runtime. This relies on a consistent return shape from refreshOAuthTokens, which isn\u2019t guaranteed currently.",
+          "severity": "High",
+          "matched_candidate": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses res?.data (undefined on fetch Response \u2192 TypeError on token.access_token), Hubspot accesses .expiryDate/.expiresIn (undefined on Response), Zoho-bigin and Zohocrm access .data.error (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.",
+          "confidence": 0.95,
+          "reasoning": "The candidate issue identifies the same core problem as the golden comment: when the sync endpoint path is used (APP_CREDENTIAL_SHARING_ENABLED is true), refreshOAuthTokens returns a fetch Response instead of the expected object format. The candidate specifically mentions that Google Calendar accesses res?.data which will be undefined on a fetch Response, causing TypeError on token.access_token - this is exactly what the golden comment describes. The candidate also expands on the issue by noting similar problems in other integrations (Hubspot, Zoho), but the fundamental issue identified is the same: inconsistent return types from refreshOAuthTokens causing runtime errors."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Salesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using credentialKey.access_token \u2014 the OLD, pre-refresh token. The newly refreshed token from accessTokenParsed.data is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token."
+        },
+        {
+          "candidate": "The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The parseRefreshTokenResponse function incorrectly sets refresh_token to the hardcoded string 'refresh_token' when it's missing from the OAuth refresh token response. This invalidates the token, breaking subsequent token refreshes and causing authentication failures.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 4,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.8,
+      "recall": 0.8,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR11059__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR11059__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/7232": {
@@ -71645,6 +73561,49 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR7232__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR7232__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Asynchronous functions deleteScheduledEmailReminder and deleteScheduledSMSReminder are called without await inside forEach loops. This occurs during booking rescheduling/cancellation, and workflow/workflow step deletion/updates. Consequently, scheduled workflow reminders may not be reliably cancelled, potentially leaving them active.",
+          "severity": "Medium",
+          "matched_candidate": "In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget and the surrounding try/catch will never catch rejections from these promises",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: async functions deleteScheduledEmailReminder and deleteScheduledSMSReminder are called without await inside forEach loops. The candidate specifically mentions the reschedule path and explains that forEach doesn't await async callbacks, making them fire-and-forget. This matches the golden comment's concern about scheduled workflow reminders not being reliably cancelled. The candidate also adds the detail about try/catch not catching rejections, which is a consequence of the same underlying issue."
+        },
+        {
+          "golden_comment": "When immediateDelete is true, the deleteScheduledEmailReminder function cancels the SendGrid email but fails to delete the corresponding WorkflowReminder record from the database. This creates orphaned database entries and is inconsistent with the immediateDelete: false path, which marks the record as cancelled. The SendGrid DELETE API call is also omitted in this path.",
+          "severity": "High",
+          "matched_candidate": "When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record, causing the record to become permanently orphaned",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: when immediateDelete is true and referenceId is non-null, the function cancels the SendGrid email but returns without deleting or updating the workflowReminder database record, causing orphaned records. Both comments point to the same bug where the database record is not properly handled in this code path."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The CRON job first deletes ALL email reminders with a past `scheduledDate`, then queries for `cancelled: true` reminders to cancel via SendGrid. If a cancelled reminder's `scheduledDate` has already passed, the first `deleteMany` removes the DB record before the cancel loop processes it, so the email is never cancelled in SendGrid and gets delivered despite the booking being cancelled"
+        },
+        {
+          "candidate": "The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails, the exception aborts the loop and all remaining reminders are never cancelled or deleted, causing those emails to be sent despite their bookings being cancelled"
+        },
+        {
+          "candidate": "The sender name/sender ID input section was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}` but after the refactor is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR7232__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR7232__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/14740": {
@@ -73779,6 +75738,62 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR14740__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR14740__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Case sensitivity bypass in email blacklist",
+          "severity": "High",
+          "matched_candidate": "Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue directly identifies the same problem as the golden comment - a case sensitivity bypass in the email blacklist. The candidate provides specific technical details about how the blacklist entries are lowercased but incoming guest emails are compared as-is, allowing users to bypass the blacklist by using different case variations. This is exactly the case sensitivity bypass issue mentioned in the golden comment."
+        },
+        {
+          "golden_comment": "The logic for checking team admin/owner permissions is incorrect. This condition uses AND (&&) which requires both isTeamAdmin AND isTeamOwner to be true, but it should use OR (||) since a user needs to be either an admin OR an owner to have permission.",
+          "severity": "Critical",
+          "matched_candidate": "Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.",
+          "confidence": 1.0,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the authorization check incorrectly uses AND (&&) instead of OR (||) for checking team admin/owner permissions. Both comments point out that the condition requires BOTH isTeamAdmin AND isTeamOwner to be true, when it should only require one or the other. The candidate even provides additional context about why this is problematic (team admins who aren't owners will be denied access)."
+        },
+        {
+          "golden_comment": "This calls the email sender with the original guests, so existing attendees included in the input will be treated as new when sending notifications, leading to incorrect emails.",
+          "severity": "Medium",
+          "matched_candidate": "The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.",
+          "confidence": 0.95,
+          "reasoning": "The candidate issue identifies the same problem as the golden comment: the code passes the original `guests` array to the email sending function instead of the filtered unique guests, which means existing attendees will be incorrectly treated as new guests and receive wrong email notifications. Both comments point to the same bug where existing attendees included in the input will receive incorrect emails."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "When `booking.userId` is null, `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable if the user is an attendee of a booking whose organizer account was deleted."
+        },
+        {
+          "candidate": "In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. If the attendees array is ever empty, this would throw a runtime TypeError crashing the email send for the entire batch."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "uniqueGuests filters out existing attendees and blacklisted emails but does not deduplicate duplicates within the input; createMany can insert duplicate attendee rows if the client submits repeated emails.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Starting with an array containing an empty string may cause validation issues. Consider starting with an empty array [] and handling the empty state in the MultiEmail component instead.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 3,
+      "fp": 2,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.6,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR14740__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR14740__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/10600": {
@@ -75696,6 +77711,54 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR10600__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR10600__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Password bypass via backup code + dummy totpCode. An attacker with a valid backup code can send a request with a dummy non-empty totpCode (e.g., 'x') alongside the stolen backupCode. The non-empty totpCode causes the password check at line 121 to be skipped, then the backup code path at line 131 succeeds without ever verifying the password, allowing login with only email + backup code."
+        },
+        {
+          "candidate": "Backup code login is blocked for non-CAL identity provider users. The guard at line 113 throws ThirdPartyIdentityProviderEnabled when !credentials.totpCode is true. When a Google/SAML user with 2FA tries to use a backup code (no totpCode provided), this check blocks the request before the backup code path is reached, making backup codes completely non-functional for these users."
+        },
+        {
+          "candidate": "Unhandled exception in API route returns unstructured error. When CALENDSO_ENCRYPTION_KEY is missing in the backup code path, throw new Error(ErrorCode.InternalServerError) causes Next.js to return a generic 500 HTML error page instead of a structured JSON response, causing the client code to fail when calling response.json()."
+        },
+        {
+          "candidate": "Object URL created for backup codes download is never revoked on component unmount or dialog close. URL.createObjectURL(textBlob) creates a blob URL that persists until explicitly revoked, but resetState (called on dialog close/cancel) does not revoke it and there is no cleanup effect, causing each 2FA setup flow to leak one blob URL."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The exported function TwoFactor handles backup codes and is in BackupCode.tsx. Inconsistent naming.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Error message mentions 'backup code login' but this is a disable endpoint, not login",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Backup code validation is case-sensitive due to the use of indexOf(). This causes validation to fail if a user enters uppercase hex characters, as backup codes should be case-insensitive for a better user experience.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Because backupCodes are decrypted and mutated in memory before being written back, two concurrent login requests using the same backupCode could both pass this check and update, so a single backup code may effectively be accepted more than once if used concurrently, weakening the intended one-time-use semantics.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 4,
+      "tp": 0,
+      "fp": 4,
+      "fn": 4,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR10600__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10600__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/10967": {
@@ -77886,6 +79949,62 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR10967__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR10967__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Potential null reference if mainHostDestinationCalendar is undefined if evt.destinationCalendar is null or an empty array ",
+          "severity": "High",
+          "matched_candidate": "When `evt.destinationCalendar` is `null` or an empty array, `const [mainHostDestinationCalendar] = evt.destinationCalendar ?? []` sets `mainHostDestinationCalendar` to `undefined`. The next line accesses `mainHostDestinationCalendar.integration` without optional chaining, which throws `TypeError: Cannot read property 'integration' of undefined`. Multiple callers (e.g., webhook handlers, payment handlers) set `destinationCalendar` to `[]` when no calendar is configured.",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment. Both point out that when evt.destinationCalendar is null or an empty array, mainHostDestinationCalendar becomes undefined, and then accessing properties on it (like .integration) will cause a null/undefined reference error. The candidate provides more detail about the specific property access and the callers that trigger this condition, but it's fundamentally the same bug."
+        },
+        {
+          "golden_comment": "Logic inversion in organization creation: The slug property is now conditionally set when IS_TEAM_BILLING_ENABLED is true, instead of when it's false as originally intended. This change, combined with requestedSlug still being set when IS_TEAM_BILLING_ENABLED is true, results in both properties being set when billing is enabled, and neither when disabled",
+          "severity": "Medium",
+          "matched_candidate": "The slug/requestedSlug logic was inverted from the original. Old code: `...(!IS_TEAM_BILLING_ENABLED && { slug })` set `slug` when billing is DISABLED (org gets slug immediately, no payment needed). New code: `...(IS_TEAM_BILLING_ENABLED ? { slug } : {})` sets `slug` when billing is ENABLED. Combined with the `requestedSlug` line (which also only fires when billing is enabled), organizations without billing now get neither `slug` nor `requestedSlug`, while organizations with billing get both \u2014 the opposite of the original intent.",
+          "confidence": 0.99,
+          "reasoning": "The candidate issue identifies exactly the same problem as the golden comment: the logic inversion in the slug/requestedSlug conditional assignment. Both describe how the original code set slug when IS_TEAM_BILLING_ENABLED was false, but the new code sets slug when it's true. Both also note that this results in both properties being set when billing is enabled and neither when disabled, which is the opposite of the original intent."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In both `updateEvent` and `deleteEvent`, when `externalCalendarId` is falsy, the else-branch searches `event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId)` \u2014 but `externalCalendarId` is guaranteed falsy at that point, so the `.find()` will never match any element and always returns `undefined`. The old code correctly fell back to `event.destinationCalendar?.externalId`. This means calendar updates and deletes will use `undefined` as the calendar ID, likely falling back to the wrong calendar or failing."
+        },
+        {
+          "candidate": "In `updateAllCalendarEvents`, if `reference.credentialId` is truthy but the credential is not found in `this.calendarCredentials` AND the DB lookup fails (null result or no `app.slug`), `credential` remains `undefined`. It is then unconditionally passed to `updateEvent(credential, ...)` on line 542, which expects a `CredentialWithAppName`. This will crash when `updateEvent` calls `getCalendar(credential)` on the undefined value."
+        },
+        {
+          "candidate": "In `handleNewBooking`, when neither `eventType.destinationCalendar` nor `organizerUser.destinationCalendar` exist, `evt.destinationCalendar` is set to `null` (line 1067). The subsequent `evt.destinationCalendar?.push(...teamDestinationCalendars)` (line 1078) is then a no-op due to optional chaining. For collective team events where the organizer lacks a calendar but team members have destination calendars, those team member calendars are silently dropped \u2014 the event will have no destination calendars despite team members having them configured."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The optional chaining on mainHostDestinationCalendar?.integration is redundant since you already check mainHostDestinationCalendar in the ternary condition.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Logic error: when externalCalendarId is provided, you're searching for a calendar where externalId === externalCalendarId, but this will always fail since you're looking for a calendar that matches itself. Should likely find by credentialId or use different logic.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The Calendar interface now requires createEvent(event, credentialId), but some implementations (e.g., Lark/Office365) still declare createEvent(event) only\u2014this breaks the interface contract (also applies to other locations in the PR).",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 3,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR10967__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10967__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/8087": {
@@ -79329,6 +81448,47 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR8087__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR8087__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The code uses forEach with async callbacks, which causes asynchronous operations (e.g., calendar/video event deletions, payment refunds) to run concurrently without being awaited. This 'fire-and-forget' behavior leads to unhandled promise rejections, race conditions, and incomplete cleanup, as surrounding try-catch blocks cannot properly handle errors from these unawaited promises. Replace forEach with for...of loops or Promise.all() with map() to ensure proper sequential execution and error handling.",
+          "severity": "Critical",
+          "matched_candidate": "async-forEach fire-and-forget bug in packages/trpc/server/routers/viewer/bookings.tsx - the forEach on line 553 uses an async callback but forEach ignores the returned promises, the deleteEvent/deleteMeeting operations are unawaited, and any errors from the async operations are silently swallowed",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue identifies the exact same problem as the golden comment: using forEach with async callbacks causes promises to be ignored/unawaited, leading to 'fire-and-forget' behavior where async operations (like deleteEvent/deleteMeeting) run without proper error handling. Both comments point to the same file (packages/trpc/server/routers/viewer/bookings.tsx) and describe the same fundamental bug - async forEach not awaiting promises and silently swallowing errors."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "async-forEach fire-and-forget bug in packages/app-store/wipemycalother/lib/reschedule.ts - the async callback's promises are discarded, so calendar deleteEvent calls are unawaited, errors won't be caught and the function proceeds immediately to send reschedule emails"
+        },
+        {
+          "candidate": "async-forEach fire-and-forget in handleCancelBooking at packages/features/bookings/lib/handleCancelBooking.ts - the forEach on line 460 uses an async callback but forEach discards the promises, causing the apiDeletes array to be empty when Promise.all(apiDeletes) is called later, silently skipping calendar deletions for recurring events"
+        },
+        {
+          "candidate": "getCalendarCredentials is a synchronous function that calls getCalendar (now async) inside flatMap on line 28 in packages/core/CalendarManager.ts - since getCalendar returns Promise<Calendar | null>, the calendar field is actually a Promise not a Calendar, breaking the type contract and causing any caller that doesn't await item.calendar to get a truthy Promise object instead of null, bypassing null checks"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Consider adding try-catch around the await to handle import failures gracefully",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR8087__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8087__20260411/pull/1"
     }
   }
 }
\ No newline at end of file
diff --git a/offline/results/anthropic_claude-sonnet-4-5-20250929/candidates.json b/offline/results/anthropic_claude-sonnet-4-5-20250929/candidates.json
index 4fe8468..d88c5ff 100644
--- a/offline/results/anthropic_claude-sonnet-4-5-20250929/candidates.json
+++ b/offline/results/anthropic_claude-sonnet-4-5-20250929/candidates.json
@@ -977,6 +977,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian. This affects both the account and login message files, meaning Lithuanian users will see Italian text for the TOTP setup step.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value '\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a' uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f). The rest of the file uses Simplified Chinese, making this inconsistent and incorrect for zh_CN locale users.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. The regex `<[a-z]+[^>]*>` matches `<a href=\"...\">`, so if the English message contains only anchor tags (no other HTML), containsHtml returns true and POLICY_SOME_HTML is selected. This allows `<br>`, `<p>`, `<strong>`, `<b>` tags in translations even when the English source only has anchor tags, defeating the stated intent in the comment on line 90: 'Only if the English source string contains HTML we also allow HTML in the translation'.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/37634": {
@@ -1878,6 +1898,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Copy-paste bug in null check: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`, allowing null `rawTokenId` to pass through and potentially cause NPEs later when used in `encodeTokenId()`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Wrong substring indices in `isAccessTokenId` matcher: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType). Should be `substring(4, 6)`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Inverted boolean logic in `isAccessTokenId` matcher: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match. The condition should be negated to `if (!... .equals(...)) return false`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/38446": {
@@ -2659,6 +2699,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID. The stored credential has an ID (set in updateCredential at line 195), but createFromValues() does not set an ID on the model, and the stored ID is never propagated. Downstream, RecoveryAuthnCodesFormAuthenticator line 86-87 calls removeStoredCredentialById(recoveryCodeCredentialModel.getId()) \u2014 passing null here will fail or cause unexpected behavior when all codes are exhausted",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The test provider's isValid() for recovery codes accepts ANY code via anyMatch(), whereas the production RecoveryAuthnCodesCredentialProvider.isValid() only validates against the NEXT sequential code (getNextRecoveryAuthnCode), allowing out-of-order codes to be accepted",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "isValid() never consumes/removes the used code \u2014 the production provider calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation. This means the same code can be reused indefinitely and tests using this provider won't catch regressions in one-time-use behavior",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "For federated users, when all recovery codes are exhausted, the authenticator calls removeStoredCredentialById() which delegates to local storage (UserCredentialStore), not federated storage. The credential lives in the federated user storage provider, so this removal silently fails \u2014 the exhausted credential persists in federated storage, leading to users being stuck in a loop of regenerating codes without the old ones being cleaned up",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The updateCredential() method stores raw recovery codes JSON in the credentialData field, but in the standard RecoveryAuthnCodesCredentialModel, credentialData holds metadata and secretData holds hashed codes. The inconsistent use of credentialData for two different purposes (raw codes in storage vs metadata in standard model) is fragile and would silently break if the storage format were changed to match the standard model",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/36882": {
@@ -3220,7 +3292,27 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic (UpgradeLogicFactory, RecreateOnImageChangeUpgradeLogic, AlwaysRecreateUpgradeLogic) makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons \u2014 it never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode (it only calls `System.exit` when NOT in test mode). The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand (`UpdateCompatibilityMetadata.NAME`), but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/keycloak/keycloak/pull/36880": {
     "claude": [
@@ -4028,6 +4120,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getResourceTypeResource() can return null but the result is passed directly to findByResource(server, resource) without a null check, causing potential NullPointerException since PolicyStore.findByResource contract requires non-null resource parameter",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getClientsWithPermission uses findByType(server, \"Clients\", ...) which only finds the all-clients type resource, not individual per-client resources created via getOrCreateResource (which never calls resource.setType()), making per-client permissions invisible",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When all-clients permission exists, the type resource name \"Clients\" is added to the result set, causing AvailableRoleMappingResource.getRoleIdsWithPermissions to call realm.getClientById(\"Clients\") which returns null and causes NPE on .getRolesStream()",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "canMapRoles, canMapCompositeRoles, and canMapClientScopeRoles methods do not check the MANAGE_CLIENTS admin role, preventing users with MANAGE_CLIENTS realm admin role from mapping roles unless they have explicit fine-grained permissions",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Javadoc for requireView() incorrectly says 'Returns {@code true} if canView() returns true' but the method is void and throws ForbiddenException on failure",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Javadoc for requireViewClientScopes() incorrectly says it returns a boolean but the method is void and throws ForbiddenException on failure",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/37038": {
@@ -4926,6 +5056,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, allowing users with only VIEW permission to pass canManage() checks, creating a privilege escalation vulnerability",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getGroupIdsWithViewPermission() uses groupResource.getId() (authorization resource server's internal UUID) instead of groupResource.getName() (actual group model ID), causing hasPermission() to fail finding resources by name and always fall through to 'all-groups' check",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getGroupIdsWithViewPermission() adds resource UUIDs to granted set, but downstream JpaUserProvider.groupsWithPermissionsSubquery() compares them against groupId column containing group model IDs, causing query filter to never match",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Interface docstring for canManage() states 'Also if it has permission to VIEW or MANAGE groups', which either codifies privilege escalation as intentional or will become misleading if canManage() is fixed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ApiUtil.handleCreatedResponse(response) is called on responses wrapped in try-with-resources, causing double assertion of status 201 and double closing of response (at lines 73-77, 195-199, and 261-265)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/33832": {
@@ -5861,6 +6023,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures. This means bugs specific to handling full-size ES384/ES512 r/s values (e.g., off-by-one in `integerToBytes`) would not be caught.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name (e.g., `\"Ignored crypto providers: FooProvider, \"`). This produces a malformed log message.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/40940": {
@@ -6647,6 +6835,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getSubGroupsStream(String, Integer, Integer) at line 256 calls modelSupplier.get().getSubGroupsStream(...) without a null check, which will throw an NPE under the same concurrent-deletion scenario that this PR fixes in getSubGroupsCount()",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getSubGroupsStream overload at line 262 calls modelSupplier.get() without a null check, risking NPE in concurrent-deletion scenarios",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getSubGroupsStream overload at line 268 calls modelSupplier.get() without a null check, risking NPE in concurrent-deletion scenarios",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Reader thread is never joined before the assertion at line 157, allowing in-flight groups() calls to throw exceptions after the assertion has passed, causing the test to silently miss failures",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1": {
@@ -7367,6 +7581,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the method signature in parent class UsernamePasswordForm was changed to `isConditionalPasskeysEnabled(UserModel user)` requiring one parameter. The call should be `isConditionalPasskeysEnabled(context.getUser())`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Regression: passkeys no longer set up during initial login. In `authenticate()`, the new code calls `fillContextForm` only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This means WebAuthn challenge, rpId, and conditional UI attributes are never set on the login form for first-time visitors (when `context.getUser() == null`), breaking passkey conditional UI on the initial login page",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null`, but the new code calls it only when `user != null`. This means if a user enters an invalid username during initial login with passkeys enabled, the passkey UI elements will not be re-rendered on the error page",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Potential NullPointerException in `fillContextForm`: The refactored code at line 103 passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. Since `shouldShowWebAuthnAuthenticators` is now `protected` and designed for override, a future subclass could override it to return `true` without a null-user guard, causing NPE where `user` would be null",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/93824": {
@@ -8206,6 +8446,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Inconsistent metrics tag key: \"shards\" (plural) is used at line 199, while all other metrics in the same function use \"shard\" (singular), causing this metric to be silently missing from dashboards or alerts that filter on the shard tag",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When deadline is exceeded during join() at line 340, the break on line 341 exits the loop entirely, skipping terminate() for all remaining processes, leaving hung processes alive as zombies until parent exits",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Each sub-buffer's get_memory_info() at line 300 calls iter_cluster_memory_usage() which queries Redis INFO on all nodes, making N redundant Redis INFO roundtrips since all sub-buffers share the same Redis cluster, adding unnecessary latency to submit() calls on the hot path",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5": {
@@ -9578,6 +9838,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` in nodestore/base.py:205 appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list. This silently pairs wrong error IDs with wrong event data (title, message, timestamp), corrupting the error context sent to the AI summarizer",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds like 1713000000.0) are compared directly against replay segment event timestamps (`event.get('timestamp', 0)`). Standard rrweb replay events use Unix epoch milliseconds (like 1713000000000). This unit mismatch means all error events will always appear to have timestamps far earlier than replay events, causing them to be yielded before any replay breadcrumbs rather than interleaved in chronological order",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1": {
@@ -10707,6 +10981,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The only caller (organization_auditlogs.py) uses order_by='-datetime', meaning self.key is 'datetime' and getattr(item, self.key) returns a datetime. Compare with DateTimePaginator.get_item_key (line 235) which correctly converts via float(value.strftime('%s.%f')) before calling math.floor().",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), `queryset[start_offset:stop]` will raise AssertionError('Negative indexing is not supported.'). The comment 'The underlying Django ORM properly handles negative slicing automatically' is incorrect.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter (e.g., `datetime__lte=<int>`), producing incorrect results or a database type error. Compare with DateTimePaginator.value_from_cursor (line 238-241) which converts the numeric cursor value back to a timezone-aware datetime.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "organization_context.member can be None (declared as `RpcOrganizationMember | None = None` at model.py:346) when the user has no membership in the org. Accessing `.has_global_access` on None raises AttributeError. This executes on EVERY request to this endpoint (not just optimized ones) because the `enable_advanced` check runs unconditionally before the `use_optimized` guard.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BasePaginator (the base class used by ALL paginators across Sentry) now allows negative start_offset when cursor.is_prev is True: `start_offset = offset` without clamping. A malicious or malformed cursor string with a negative offset (e.g., '0:-3:1') will cause Django's AssertionError on queryset slicing.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/97529": {
@@ -11415,6 +11721,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Race condition from lock scope reduction: The old code held `cacheMu.Lock()` for the entire `BuildIndex` function, serializing all concurrent calls. The new code only locks the final cache write. This means two concurrent callers (e.g., a Search request and a handleEvent goroutine both hitting `getOrCreateIndex` for the same unindexed key) can simultaneously build an index for the same key. For file-based indexes, the second `bleve.New(dir, mapper)` call will fail because the directory already exists. For memory indexes, duplicate work is done and the first result is silently overwritten.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Data race on shared `err` variable in concurrent goroutines: The `err` variable from `stats, err := s.storage.GetResourceStats(...)` on line 181 is captured by reference in all `group.Go` closures. Multiple goroutines concurrently write to this same `err` via `_, _, err = s.build(...)` and then read it via `return err`. This is both a data race (detected by `-race`) and a correctness bug \u2014 a goroutine may return another goroutine's error (or nil) instead of its own.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Data race on `totalBatchesIndexed`: This variable is incremented from multiple goroutines launched via `errgroup` without any synchronization. With `initWorkers > 1`, concurrent increments are a data race.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/80168": {
@@ -12099,7 +12425,33 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "MetricAlertDetectorHandler extends StatefulDetectorHandler but doesn't implement required abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), causing TypeError when instantiated",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In TestEvaluateGroupKeyValue.test_dedupe, expected occurrence uses group_key='val1' but evaluate_group_key_value is called with group_key='group_key', creating mismatched fingerprints that aren't caught because IssueOccurrence.__eq__ only compares id field",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In test_state_results_multi_group, expected occurrence for group_2 is built with value=6 but actual data packet has group_2=10, test passes incorrectly because mock doesn't vary output based on value parameter",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In test_state_results, detector.detector_handler creates a new handler instance on each property access, so the handler used in process_detectors differs from the one used to build expected occurrence, creating fragile test that works only by coincidence",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/getsentry/sentry/pull/80528": {
     "gemini": [
@@ -12572,6 +12924,14 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In `get_monitor_environment_context`, line 168 returns `monitor_environment.monitor.config` (the original config) instead of the local `config` variable, causing the `schedule_type` override from line 162 to be silently discarded",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/77754": {
@@ -13135,6 +13495,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Classic Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition (module import) time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` (which doesn't pass `queued`) will share the same timestamp \u2014 the time the module was first imported. The fix is `queued: datetime = field(default_factory=timezone.now)`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `assignment_source` parameter was added to the abstract `sync_status_outbound` method, but this PR is about assignee sync cycles \u2014 status sync is unrelated.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` (line 161) does not accept `**kwargs`, so any future caller passing `assignment_source` to it will get a TypeError. The other concrete implementations (jira, jira_server) use `**kwargs` so they silently absorb it, but ExampleIntegration will crash.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array. The name is misleading and suggests the wrong data structure is being tested.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test method name `test_from_dict_inalid_data` has a typo \u2014 missing 'v' in 'invalid'. While the test still runs (Python doesn't validate spelling), this makes it harder to find via search and indicates insufficient review.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/95633": {
@@ -13886,6 +14278,62 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "close() joins the commit thread before shutting down the worker pool, causing processed items after final commit to have uncommitted offsets, leading to duplicate processing on restart",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_thread_queue_parallel_error_handling's docstring contradicts its assertion - docstring claims it tests that errors don't block commits, but assertion checks that no commits happened",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_thread_queue_parallel_error_handling is flaky because it relies on timing - passes only because commit loop waits 1 second but test waits 0.2s",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_thread_queue_parallel_error_handling assertion is incorrect - offsets will be committed for both failed and successful messages due to finally block in OrderedQueueWorker.run() line 149",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' but queues are created without maxsize parameter (line 185), making them unbounded with no backpressure mechanism",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SimpleQueueProcessingStrategy docstring claims 'Only commits offsets after successful processing' but finally block in OrderedQueueWorker.run() always calls complete_offset regardless of exceptions",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Unbounded queues in SimpleQueueProcessingStrategy can cause unbounded memory growth when workers are slow and producer is fast",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OffsetTracker._get_partition_lock accesses self.partition_locks without synchronization, relying on CPython GIL rather than language spec guarantees",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OffsetTracker.get_committable_offsets iterates list(self.all_offsets.keys()) without holding any lock while worker threads may be modifying all_offsets via defaultdict auto-creation",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2": {
@@ -15119,6 +15567,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. This paginator is used with order_by=\"-datetime\" on AuditLogEntry, so getattr(item, self.key) returns a datetime. Compare to DateTimePaginator.get_item_key which correctly converts to float first via `float(value.strftime(\"%s.%f\")) * self.multiplier` before applying math.floor/ceil.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing \u2014 it raises AssertionError(\"Negative indexing is not supported.\"). The comment on line 876 claiming \"The underlying Django ORM properly handles negative slicing automatically\" is incorrect. Any request to the audit logs endpoint with optimized_pagination=true and a crafted cursor with a negative offset will crash with a 500 error.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The old code had a safety check that skipped segments exceeding max_segment_spans (default 1001) during flush, with metrics and error logging. This check was completely removed. While the Lua script now caps spans at 1000 during insert, the flush-side guard protected against edge cases (e.g., race conditions between concurrent inserts, or spans inserted before the eviction logic was deployed). The max_segment_spans attribute is still initialized on the class but is now dead code.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3": {
@@ -16158,6 +16626,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Python's built-in hash() is randomized per-process (via PYTHONHASHSEED), causing different worker processes to compute different hash values for the same project IDs in shared Redis/Memcached cache, resulting in cache misses across processes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "invalidate_upsampling_cache function uses non-deterministic hash() in cache key construction, preventing cache invalidation from one process from clearing entries written by different processes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Cache key doesn't include dataset or request context, causing the same cache entry to be reused across different query types for the same org+projects, potentially returning inconsistent results",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Behavior changed from any() to all() for project allowlist checking, causing queries spanning both allowlisted and non-allowlisted projects to return raw sample counts instead of upsampled estimates",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Validation preventing 'timestamp' in raw_groupby was removed along with ParseError import, allowing users to group by 'timestamp' which was previously explicitly disallowed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/103633": {
@@ -16935,6 +17435,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. Line 981 comments \"Allow access to the dashboard to prove this is not checked\" but line 982 sets the permission cache value to `false` (`map[string]bool{\"dashboards:uid:dash1\": false}`). Since `checkPermission` checks `scopeMap[t.scope(req.Name)]` and `false` is falsy, the permission cache would ALSO deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance (both passed the same `cache` in `NewService`) but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix `ns.perm_UID_ACTION_NAME_PARENT` overlaps with the permission cache key prefix `ns.perm_UID_ACTION`. If a denial key accidentally matches a permission key (e.g., if a user UID contains underscore-delimited segments resembling `UID_ACTION_NAME_PARENT`), `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"` (line 137), suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates two semantically different situations (cache miss vs. cache hit with stale data) into the same metric label, making it difficult to distinguish genuine cache misses from outdated-cache fallbacks when monitoring.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/67876": {
@@ -17779,6 +18305,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OAuth `state` parameter uses deterministic MD5 hash (pipeline.signature) instead of cryptographic random nonce, allowing attackers to forge state parameter and bypass CSRF protection",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Accessing integration.metadata[\"sender\"][\"login\"] raises KeyError when sender key is missing from metadata (integrations created before code change or through non-webhook paths)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "First-time GitHub app installation bypasses sender/user validation when Integration.DoesNotExist exception occurs, allowing attacker to install app without being the initiating GitHub user",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "redirect_uri missing from token exchange request despite being included in authorization request, violating OAuth2 RFC 6749 Section 4.1.3 requirement",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "error() helper function accesses org.organization without verifying org type or structure, causing potential crash when active_organization is None or unexpected type",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/32918": {
@@ -18315,6 +18873,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Test cleanup registers removal for a literal \"alias\" instead of the actual IDP alias (e.g., \"idp-alias-\" + i), causing all 20 IDPs created in the loop to never be cleaned up and potentially leak state to other tests",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Same cleanup bug exists at line 425 for idp-alias-20",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getForLogin cached path returns results from a HashSet<IdentityProviderModel> which does not preserve ordering from the delegate's query, causing inconsistent behavior between cached and non-cached paths for future callers relying on delegate ordering",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In registerIDPLoginInvalidationOnUpdate, the original model is wrapped with createOrganizationAwareIdentityProviderModel while the updated model is raw, causing unnecessary cache invalidation when org-aware isEnabled() differs from the raw value",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/94942": {
@@ -19039,6 +19623,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code that misleadingly suggests the flag could re-enable the feature when it cannot",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `DB.QueryFramesInto` stub method always returns an error but the caller in `sql_command.go:Execute` still appends an empty frame to results after setting the error, creating a latent bug if the guard is ever removed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/90939": {
@@ -19498,6 +20096,14 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether entryPointAssetsCache was already populated by another goroutine, causing redundant I/O operations and thundering herd of file reads in non-dev mode with concurrent requests at startup",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/80329": {
@@ -20416,6 +21022,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "The cleanup service ticker was changed from 10 minutes to 1 minute, which appears to be an accidental debug leftover. The clean() method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running. This 10x increase in cleanup frequency will also increase database load significantly in production.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "All diagnostic log lines in the cleanup path use r.log.Error() instead of r.log.Debug() or r.log.Info(). These are clearly informational messages (e.g., 'Annotations to clean by time', 'cleaned annotations by time') that fire on every successful cleanup cycle. In production this will flood error-level logs and likely trigger monitoring/alerting systems, making real errors harder to find.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The SQLite parameter limit guard checks r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit but should check len(ids) > sqliteParameterLimit. The batch size is the maximum number of IDs that can be fetched, but the actual number returned may be less. If AnnotationCleanupJobBatchSize is e.g. 1000 but only 5 IDs are returned, the code will unnecessarily build an inline SQL string instead of using parameterized queries, which is the safer approach. The guard should reflect the actual parameter count, not the configured ceiling.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test helper createTestAnnotations hardcodes annotation IDs as int64(i + 1) via InsertMulti. Since tests share a single db.InitTestDB(t) and SQLite's auto-increment counter is not reset by DELETE FROM annotation, subsequent test cases that also insert IDs starting at 1 will collide with the auto-increment sequence on databases that enforce uniqueness on the primary key. If any test case inserts without explicit IDs, the auto-generated IDs may collide with IDs from a previous createTestAnnotations call that didn't fully clean up yet.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/90045": {
@@ -21684,6 +22316,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In Create's error path (line 45), recordLegacyDuration is called instead of recordStorageDuration, causing storage operation latency to be recorded against the legacy histogram",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In Update's error path (line 129), recordLegacyDuration is called instead of recordStorageDuration, causing storage operation latency to be recorded against the legacy histogram",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In DeleteCollection's goroutine (line 166), recordStorageDuration is called instead of recordLegacyDuration for the legacy write, causing legacy latency to be recorded in the storage histogram",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In Delete's success path (line 106), recordStorageDuration uses 'name' (object name) instead of 'options.Kind' as the third argument, creating high-cardinality metric labels",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In Delete method (line 97), klog.NewContext uses d.Log instead of the enriched log variable that includes name, kind, and method values",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/106778": {
@@ -22824,6 +23488,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Missing React `key` prop on GrafanaRuleListItem in FilterView.tsx line 157. The old GrafanaRuleLoader had key={key}, but the replacement GrafanaRuleListItem omits it, while other branches (DataSourceRuleLoader, UnknownRuleListItem) all have key={key}. This will cause React reconciliation issues when the list updates (e.g., items reordering, stale state, performance degradation).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Delete menu item will never render in the new list view. AlertRuleMenu at line 165 guards the delete menu item with `canDelete && rulerRule && (...)`. Since the new list view no longer fetches ruler data, rulerRule is always undefined when coming from GrafanaRuleListItem \u2192 RuleActionsButtons.V2 \u2192 AlertRuleMenu. Even though canDelete becomes true via the new grafanaDeleteAbility path, the rulerRule guard prevents the delete option from appearing.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Pause menu item will never render in the new list view. At line 123, `canPause && rulerRuleType.grafana.rule(rulerRule)` requires rulerRule to be a grafana ruler rule. Since rulerRule is undefined when coming from the new list view, rulerRuleType.grafana.rule(undefined) returns false, so the pause option never appears \u2014 even though the user has permission via the grafana prom abilities path.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Silence drawer won't open in the new list view. In RuleActionsButtons.V2.tsx, the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` at line 104. When rule (the RulerRuleDTO) is undefined (coming from GrafanaRuleListItem which only passes promRule), this check is false. So even though the 'Silence notifications' menu item appears (via canSilence from the grafana prom path), clicking it sets showSilenceDrawer to true but the drawer never renders.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The useGrafanaPromRuleAbility hook in RuleActionsButtons.V2.tsx is called with promRule typed as Rule (from unified-alerting.ts), but the hook expects GrafanaPromRuleDTO | SkipToken. While prometheusRuleType.grafana.rule() does a structural check, it returns boolean (not a type predicate narrowing to GrafanaPromRuleDTO), so TypeScript won't narrow the type. This is likely a compile error, or if it passes due to structural compatibility, the folderUid field used inside useIsGrafanaPromRuleEditable may be missing on non-Grafana Rule objects (where folderUid is optional), leading to incorrect permission checks.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/107534": {
@@ -23354,7 +24050,27 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "Double interpolation of queries in shard splitting path: runShardSplitQuery() calls applyTemplateVariables() at line 52, then splitQueriesByStreamShard() calls runSplitQuery() at line 163, which also calls applyTemplateVariables() at line 299. This causes addAdHocFilters() to unconditionally append label matchers twice via addLabelToQuery(), duplicating ad-hoc filters like env=\"prod\" in the query expression",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test mock for applyTemplateVariables mutates input query object directly (query.expr = query.expr.replace(...)) instead of returning a new object like the real implementation ({ ...target, expr, step, legendFormat }), hiding bugs where code relies on original targets remaining unmodified",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Filter ordering change in runShardSplitQuery causes behavioral difference: new code filters queries before interpolation, so template variables that resolve to empty strings (e.g., expr: '$optional_query' \u2192 '') pass the filter but result in empty expressions being sent to splitQueriesByStreamShard, whereas old code filtered after interpolation",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/grafana/grafana/pull/79265": {
     "gemini": [
@@ -24457,6 +25173,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "TOCTOU race in device limit enforcement. CreateOrUpdateDevice first counts devices, then conditionally calls updateDevice. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit. With high concurrency this could allow significantly more devices than intended.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The updateDevice method returns ErrDeviceLimitReached when rowsAffected == 0, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its updated_at falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with ErrDeviceLimitReached instead of being re-created or handled differently. The error is misleading for the second case.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The AnonymousDeviceLimit field in FrontendSettingsDTO is int64 with json:\"anonymousDeviceLimit\" \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as number | undefined. The Go DTO has no omitempty tag, so the field is never undefined in practice. Any frontend code checking config.anonymousDeviceLimit === undefined to detect 'no limit configured' will never match; it will see 0 instead.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9": {
@@ -25171,7 +25913,33 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "LRU translation cache key does not include SiteSetting.default_locale, causing stale translations to be served when admin changes the site's default locale (cached translations resolved through old fallback chain continue to be returned)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "No I18n.reload! triggered on default_locale change to clear the translation cache",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "translate method only loads config.locale but not fallback locales (default_locale, :en), causing fallback feature to silently fail in code paths outside controllers (Sidekiq jobs, rake tasks, console)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ensure_loaded! initializes @loaded_locales ||= [] outside LOAD_MUTEX, allowing race condition where concurrent threads create separate arrays and one gets discarded, potentially causing locales to appear unloaded or skip loading entirely",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/grafana/grafana/pull/76186": {
     "gemini": [
@@ -25842,6 +26610,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(), causing tests using TestLogger to silently lose all log messages when LoggerMiddleware.logRequest calls m.logger.FromContext(ctx).Info(...). The method should return the receiver (f) to preserve log capture, consistent with how the real implementation preserves the logger identity.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic. This is inconsistent with LoggerMiddleware which explicitly handles nil requests with guard clauses.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10": {
@@ -27389,6 +28171,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist (e.g., fresh installs or installs that never configured embedding), the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote (e.g., from a corrupted setting) will break the SQL or allow injection.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist (e.g., already deleted, invalid id). In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value (host is nil), `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it. The controller's `save_host` could pass nil if `params[:embeddable_host][:host]` is missing.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Fabricator file contents are swapped. `category_fabricator.rb` now contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators (`:category`, `:diff_category`, `:happy_category`, `:private_category`). The file names don't match their contents.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores (e.g., `'some_admin_model'` becomes `'some-admin_model'` instead of `'some-admin-model'`).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7": {
@@ -28403,6 +29223,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Light-theme lightness value changed from 30% to 70% (swapped with dark value) in app/assets/stylesheets/desktop/topic-post.scss line 291 for .embedded-posts link text, making it much lighter/more washed out",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Light-theme lightness value changed from 70% to 30% in app/assets/stylesheets/mobile/modal.scss line 102 for .custom-message-length, opposite of desktop version, making character count text much darker on mobile light themes and lighter on mobile dark themes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Light-theme lightness for h3 changed from 20% to 50% in app/assets/stylesheets/mobile/topic-post.scss line 182, making headings significantly more faded/grey instead of near-black",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Light-theme lightness for .name changed from 30% to 50% in app/assets/stylesheets/desktop/user.scss line 522, making user names noticeably lighter/more washed out, inconsistent with nearby .username a which preserves 30%",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8": {
@@ -29670,6 +30516,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Breaking change in GroupsController#members endpoint response format - changed from flat JSON array to object with members array and meta fields, breaking existing API consumers",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "GroupsController#members endpoint now always paginates with default limit of 50, whereas old code returned all members for non-automatic groups, causing silent truncation for existing callers",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Off-by-one error in totalPages calculation using Math.floor(user_count / limit) + 1 instead of Math.ceil(user_count / limit), creating extra empty page when user_count is divisible by limit",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test for 'cannot remove members from automatic groups' uses xhr :put but route is defined as DELETE, not validating actual HTTP method used by clients",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In admin/groups_controller update action, group.visible = params[:visible] == 'true' executes unconditionally, causing partial updates without visible parameter to reset visibility to false",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Group model asJSON() sends visible: !!this.get('visible') which evaluates to false for new groups where visible is undefined/null, causing new groups to always be created with visible: false regardless of checkbox state",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3": {
@@ -30470,6 +31354,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation, which can run multiple times per user action, inflating blocked-email statistics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "record.save inside should_block? triggers BlockedEmail's own validations including a uniqueness DB query, making validation unnecessarily expensive",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "JavaScript code adds email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures, permanently marking valid emails that belong to other users as rejected",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "rejectedEmails is initialized as Em.A([]) directly in the class definition, making it a shared prototype property across all instances, causing cross-instance state pollution",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BlockedEmail.should_block? calls record.save but doesn't check the return value, silently ignoring save failures and operating on dirty in-memory state with incremented match_count that wasn't persisted",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5": {
@@ -31230,7 +32146,45 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "The `order` mixin passes raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values (default is 1) while the standard `order` property is 0-based (default is 0). The correct value for `-webkit-box-ordinal-group` should be `$val + 1`, otherwise `@include order(1)` would get `-webkit-box-ordinal-group: 1` which equals the default, breaking intended ordering in old WebKit browsers",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `.small-action`, child `.topic-avatar` still has `float: left` despite parent now being `display: flex`. While floats are ignored on flex items, this indicates incomplete migration and should be removed for clarity",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that created spacing from the avatar. With `1.5%` horizontal padding, the gap between avatar and description text will shrink significantly on narrow viewports and may overlap or appear too tight",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`-ms-align-items` (line 121) is not a valid CSS property and will be silently ignored by IE10/11. The correct IE prefix for `align-items` is only `-ms-flex-align`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `-ms-flex-align` property uses different value keywords than standard `align-items`: it expects `start`/`end` instead of `flex-start`/`flex-end`. While current usages (`center`, `baseline`) happen to match, the mixin does not translate values, so future callers passing `flex-start` or `flex-end` would silently break in IE",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `.badge-wrapper.bullet` styling is inconsistent: `margin-top: 5px` was removed in the header's `.extra-info-wrapper` context to accommodate flexbox alignment, but the same selector in `#topic-title .title-wrapper` retains `margin-top: 5px`, creating unintentional inconsistency",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6": {
     "gemini": [
@@ -32004,6 +32958,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized \u2014 even when `website` is nil/blank, defeating the intent of the guard.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`website_name` is not included in the `untrusted_attributes` list (lines 105-111), but `website` is. Since `website_name` bypasses this check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `Discourse.stubs(:current_hostname)` call happens inside the `it` block, after `json` is already defined as a lazy let. If `json` were ever evaluated before the stub (e.g., by a shared example or `before` block), the stub would not be in effect. The stub should be in a `before` block before `json` is accessed.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4": {
@@ -33859,6 +34833,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression (e.g., `end if condition`). This will raise a SyntaxError when the `best.html.erb` template is rendered.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. The `<%= %>` ERB tag HTML-escapes the output, but HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`. SimpleRSS returns nil for missing fields.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a *substring* of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain (e.g., `http://evil.com` if discourseUrl is `http://evil.com.attacker.com/...`) to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`contents << \"\\n<hr>...\"` on line 13 uses the `<<` operator which mutates the string in place. This modifies the caller's original string. In `import_remote`, `doc.content` gets mutated. If `import` is called twice with the same string (e.g., in tests via `let(:contents)`), the imported-from footer is appended multiple times. More critically, when `import` is called for an update path (content_sha1 check on line 34), the SHA1 was computed *after* mutation, so re-importing the same content will always differ from the stored SHA1 (since the stored one includes one footer but the new computation includes two).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1": {
@@ -35102,6 +36108,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The 413 error handler hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2": {
@@ -36052,6 +37084,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "NullPointerError crash: `TopicUser.find_by` returns `nil` when no TopicUser record exists (e.g., user received a notification but never visited the topic). The very next line calls `tu.notification_level` on the nil result, which raises `NoMethodError: undefined method 'notification_level' for nil:NilClass`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `unsubscribe_link` i18n string now requires an `%{unsubscribe_url}` interpolation variable, but existing email paths that use `add_unsubscribe_link: true` without passing `unsubscribe_url` (e.g., the `mailing_list_notify` path which calls `send_notification_email` without an `unsubscribe_url` key) will raise an `I18n::MissingInterpolationArgument` error or render a broken `[click here]()` link with an empty URL.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS vulnerability: the `title` value is injected directly into HTML via string concatenation (`buffer.push(\"<h4 class='title'>\" + title + \"</h4>\")`). If `title` contains user-controlled content (e.g., a topic title with `<script>` tags), it will be rendered as raw HTML. The old code had the same issue but always rendered; the new code adds a nil guard but still doesn't escape the value.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The unsubscribe controller uses `stopNotificiationsText` (note the typo: 'Notificiations' instead of 'Notifications') as the computed property name. The template references `{{{stopNotificiationsText}}}` with the same typo so it works, but the triple-stache `{{{ }}}` renders raw HTML. The computed property interpolates `model.fancyTitle` (which can contain user-supplied HTML entities) into the i18n string. If `fancyTitle` contains malicious HTML, it would be rendered unescaped.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `html_part` method only substitutes `%{respond_instructions}` and `%{unsubscribe_link}` placeholders when `add_unsubscribe_link` is true. But the notification.html.erb template always contains `%{respond_instructions}` (both inside the `context_posts.present?` block and at the bottom). When `add_unsubscribe_link` is false, these literal `%{respond_instructions}` and `%{unsubscribe_link}` strings will appear in the rendered email.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/22532": {
@@ -37202,6 +38266,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive. The comment says 'Update SelectedCalendar.updatedAt' but nothing is actually updated, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `deleteCache` handler only checks `userId: user.id` on the credential, but does not account for delegation credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `deleteCache` handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The Prisma schema uses `@default(now()) @updatedAt` together. For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` is semantically incorrect. The `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. `?? null` would be more correct.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `CredentialActionsDropdown` component initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them. When `canDisconnect` is false and `hasCache` is false, the component returns null but still sets up subscription handlers unnecessarily.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/8330": {
@@ -38166,6 +39268,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range (returns true \u2192 function returns false), it prevents checking whether the slot is valid for a different user's override. The combination of mutation (`dateOverrideExist = true`) and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/14943": {
@@ -39485,6 +40613,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Public methods getAuthorizationConditions() and getFilterConditions() return Prisma.Sql fragments containing OR operators without outer parentheses, causing SQL operator precedence bugs when composed with AND operators. This creates a privilege escalation vulnerability where org team bookings could bypass date filters.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Constructor parameter type changed from discriminated union InsightsBookingServiceOptions to flat type InsightsBookingServicePublicOptions, removing compile-time type safety for required teamId parameter when scope='team'. Missing teamId now silently fails at runtime instead of causing compile error.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Falsy guard 'if (this.filters.eventTypeId)' skips filter when eventTypeId is 0, and 'if (this.filters.memberUserId)' skips when memberUserId is 0, causing unfiltered results instead of filtering by ID=0.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getBaseConditions method has unreachable dead code in 'else if (filterConditions)' and 'else return NOTHING_CONDITION' branches because getAuthorizationConditions() always returns truthy Prisma.Sql object. Also prevents short-circuiting when auth fails.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Entire 'Caching' test suite removed, leaving caching logic in service (lines 84-87 and 91-94) untested. No tests to catch regressions if caching is accidentally broken.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/11059": {
@@ -41462,6 +42622,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously googleCredentialSchema.parse() returned just the data. Now the DB will store {success: true, data: {access_token, refresh_token, expiry_date}} instead of {access_token, refresh_token, expiry_date}, corrupting the credential for all subsequent reads.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses res?.data (undefined on fetch Response \u2192 TypeError on token.access_token), Hubspot accesses .expiryDate/.expiresIn (undefined on Response), Zoho-bigin and Zohocrm access .data.error (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "minimumTokenResponseSchema uses [z.string().toString()] as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field (like expires_in or expiry_date) exists in the response, making the 'minimum token validation' ineffective.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Salesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using credentialKey.access_token \u2014 the OLD, pre-refresh token. The newly refreshed token from accessTokenParsed.data is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/7232": {
@@ -42718,6 +43910,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record. The record becomes permanently orphaned\u2014it's never cleaned up by the CRON's cancel loop (since `cancelled` is not set to true) and won't be found by the unscheduled query either (since `scheduled` is true)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The CRON job first deletes ALL email reminders with a past `scheduledDate` (line 34-41), then queries for `cancelled: true` reminders to cancel via SendGrid (line 44-51). If a cancelled reminder's `scheduledDate` has already passed, the first `deleteMany` removes the DB record before the cancel loop processes it. The email is never cancelled in SendGrid and gets delivered despite the booking being cancelled",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails (e.g., network error, invalid batch_id), the exception aborts the loop and all remaining reminders are never cancelled or deleted. Those emails will be sent despite their bookings being cancelled",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget. The surrounding try/catch will never catch rejections from these promises\u2014they'll become unhandled promise rejections instead of being logged as intended",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The sender name/sender ID input section (lines 463-492) was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}`. After the refactor, this section is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/14740": {
@@ -44000,6 +45224,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. In `sendAddGuestsEmails`, `newGuests.includes(attendee.email)` determines which email template to use: new guests get `AttendeeScheduledEmail` (booking confirmation), existing attendees get `AttendeeAddGuestsEmail` (notification). If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When `booking.userId` is null (it's `Int?` in the Prisma schema), `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable: if the user is an attendee (not the organizer) of a booking whose organizer account was deleted, the authorization check passes but this query crashes.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. While in practice attendees should exist, if the attendees array is ever empty (e.g., all attendees removed in a race condition), this would throw a runtime TypeError crashing the email send for the entire batch (since all emails are sent via `Promise.all`).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/10600": {
@@ -45241,6 +46497,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Password bypass via backup code + dummy totpCode. The password check at line 121 is guarded by `!credentials.totpCode`. An attacker who has obtained a valid backup code can send a request with a dummy non-empty `totpCode` (e.g., \"x\") alongside the stolen `backupCode`. The non-empty `totpCode` causes the password check to be skipped (line 121 evaluates to false). Then at line 131, `credentials.backupCode` is truthy so the backup code path is entered \u2014 which succeeds without ever verifying the password. The TOTP path (line 157) is never reached because it's in the `else if`. This allows login with only email + backup code, no password required.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Backup code login is blocked for non-CAL identity provider users. The guard at line 113 throws `ThirdPartyIdentityProviderEnabled` when `!credentials.totpCode` is true. When a Google/SAML user with 2FA tries to use a backup code (which means no totpCode is provided), this check blocks the request before the backup code path at line 131 is ever reached. While the UI currently doesn't show the 'Lost access' option for external TOTP flows (ExternalTotpFooter has no such button), this means backup codes are completely non-functional for these users \u2014 defeating the purpose of backup codes as a recovery mechanism.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Unhandled exception in API route returns unstructured error. When `CALENDSO_ENCRYPTION_KEY` is missing in the backup code path, `throw new Error(ErrorCode.InternalServerError)` causes Next.js to return a generic 500 HTML error page instead of a structured JSON response. The client code (`DisableTwoFactorModal`) calls `response.json()` and checks `body.error`, which would fail on an HTML response, causing a silent client-side error.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Object URL created for backup codes download is never revoked on component unmount or dialog close. `URL.createObjectURL(textBlob)` at line 99 creates a blob URL that persists until explicitly revoked. While `handleSetup` revokes the previous URL before creating a new one, the `resetState` function (called on dialog close/cancel) does not revoke it, and there is no cleanup effect. Each 2FA setup flow leaks one blob URL.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/10967": {
@@ -46830,6 +48112,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In updateEvent and deleteEvent, when externalCalendarId is falsy, the else-branch searches event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId) but externalCalendarId is guaranteed falsy at that point, so .find() will never match and always returns undefined instead of falling back to event.destinationCalendar?.externalId",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When evt.destinationCalendar is null or empty array, const [mainHostDestinationCalendar] = evt.destinationCalendar ?? [] sets mainHostDestinationCalendar to undefined, then accessing mainHostDestinationCalendar.integration without optional chaining throws TypeError",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In updateAllCalendarEvents, if reference.credentialId is truthy but credential is not found in this.calendarCredentials AND DB lookup fails, credential remains undefined and is passed to updateEvent which will crash when calling getCalendar(credential)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In handleNewBooking, when neither eventType.destinationCalendar nor organizerUser.destinationCalendar exist, evt.destinationCalendar is set to null, making evt.destinationCalendar?.push(...teamDestinationCalendars) a no-op that silently drops team member calendars in collective events",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The slug/requestedSlug logic is inverted: old code set slug when IS_TEAM_BILLING_ENABLED is false, new code sets slug when IS_TEAM_BILLING_ENABLED is true, causing organizations without billing to get neither slug nor requestedSlug while organizations with billing get both",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/8087": {
@@ -48045,6 +49359,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "async callback inside forEach is fire-and-forget in packages/app-store/vital/lib/reschedule.ts line 125. The forEach does not await the async callback, so: (1) the try/catch block on line 124 will never catch errors from the async operations, (2) the deleteEvent/deleteMeeting calls are not awaited before the function proceeds to send reschedule emails on line 143, meaning calendar/video deletions may silently fail or race with subsequent operations",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "async callback inside forEach is fire-and-forget in packages/app-store/wipemycalother/lib/reschedule.ts line 125. The async callback's promises are discarded, so calendar deleteEvent calls are unawaited. Errors won't be caught and the function proceeds immediately to send reschedule emails",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "async callback inside forEach is fire-and-forget in packages/trpc/server/routers/viewer/bookings.tsx line 553. The forEach uses an async callback but forEach ignores the returned promises. The deleteEvent/deleteMeeting operations are unawaited, and the code immediately proceeds to send reschedule emails on line 570. Any errors from the async operations are silently swallowed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "async callback inside forEach is fire-and-forget in packages/features/bookings/lib/handleCancelBooking.ts line 460. The forEach uses an async callback, but forEach discards the promises. The `apiDeletes` array is populated inside the async callback, but since the callback is not awaited, `apiDeletes` will be empty when `Promise.all(apiDeletes)` is called later, causing calendar deletions for recurring events to be silently skipped",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getCalendarCredentials in packages/core/CalendarManager.ts line 28 is a synchronous function that calls getCalendar (now async) inside flatMap. Since getCalendar now returns Promise<Calendar | null>, the `calendar` field in each returned object is a Promise, not a Calendar. The return type claims `calendar` is `Calendar | null` when it's actually `Promise<Calendar | null>`. Any caller that doesn't await `item.calendar` will get a truthy Promise object instead of null, bypassing null checks",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   }
 }
\ No newline at end of file
diff --git a/offline/results/anthropic_claude-sonnet-4-5-20250929/evaluations.json b/offline/results/anthropic_claude-sonnet-4-5-20250929/evaluations.json
index 509dba9..b9e04d3 100644
--- a/offline/results/anthropic_claude-sonnet-4-5-20250929/evaluations.json
+++ b/offline/results/anthropic_claude-sonnet-4-5-20250929/evaluations.json
@@ -1811,6 +1811,52 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37429__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37429__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The translation is in Italian instead of Lithuanian. This should be translated to Lithuanian to match the file's locale (messages_lt.properties).",
+          "severity": "Medium",
+          "matched_candidate": "Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian. This affects both the account and login message files, meaning Lithuanian users will see Italian text for the TOTP setup step.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: Italian text appearing in Lithuanian translation files. The golden comment notes the translation is in Italian instead of Lithuanian for messages_lt.properties, while the candidate provides more detail by specifying the Italian text itself and noting it affects multiple files. Both are flagging the same fundamental problem - wrong language in locale files."
+        },
+        {
+          "golden_comment": "The totpStep1 value uses Traditional Chinese terms in the Simplified Chinese file (zh_CN), which is likely incorrect for this locale. Please verify the locale\u2011appropriate translation.",
+          "severity": "Medium",
+          "matched_candidate": "Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value '\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a' uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f). The rest of the file uses Simplified Chinese, making this inconsistent and incorrect for zh_CN locale users.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: the zh_CN (Simplified Chinese) file contains Traditional Chinese characters instead of Simplified Chinese. The golden comment mentions 'totpStep1 value uses Traditional Chinese terms in the Simplified Chinese file' and the candidate provides specific examples of Traditional vs Simplified characters (\u624b\u6a5f vs \u624b\u673a, \u5b89\u88dd vs \u5b89\u88c5, \u61c9\u7528\u7a0b\u5f0f vs \u5e94\u7528\u7a0b\u5e8f) for the same translation key. Both flag this as a locale-inappropriate translation issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. The regex `<[a-z]+[^>]*>` matches `<a href=\"...\">`, so if the English message contains only anchor tags (no other HTML), containsHtml returns true and POLICY_SOME_HTML is selected. This allows `<br>`, `<p>`, `<strong>`, `<b>` tags in translations even when the English source only has anchor tags, defeating the stated intent in the comment on line 90: 'Only if the English source string contains HTML we also allow HTML in the translation'."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The anchor sanitization logic has a potential issue where it consumes English matcher groups without proper validation. If the translated text has more anchor tags than the English text, this could lead to incorrect validation results.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The method name 'santizeAnchors' should be 'sanitizeAnchors' (missing 'i').",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37429__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37429__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/37634": {
@@ -3590,6 +3636,52 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37634__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37634__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Wrong parameter in null check (grantType vs. rawTokenId)",
+          "severity": "Critical",
+          "matched_candidate": "Copy-paste bug in null check: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`, allowing null `rawTokenId` to pass through and potentially cause NPEs later when used in `encodeTokenId()`",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same bug: the null check is validating the wrong parameter (grantType instead of rawTokenId). The candidate provides more detail about the consequences (potential NPEs in encodeTokenId()), but the core issue - wrong parameter in the null check - is identical."
+        },
+        {
+          "golden_comment": "In isAccessTokenId, the substring for the grant shortcut and the equality check look inverted: the grant shortcut occupies indices 4\u20135 (substring(4,6)), and a match should return true (combined with UUID check), not false.",
+          "severity": "High",
+          "matched_candidate": "Wrong substring indices in `isAccessTokenId` matcher: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType). Should be `substring(4, 6)`",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same bug: the substring indices in isAccessTokenId are wrong. The golden comment states the grant shortcut is at indices 4-5 (substring(4,6)), while the candidate specifies the same issue - that substring(3,5) extracts indices 3-4 when it should be substring(4,6) to get indices 4-5 where the grant type shortcut is located. Both are pointing to the exact same incorrect substring operation."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Inverted boolean logic in `isAccessTokenId` matcher: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match. The condition should be negated to `if (!... .equals(...)) return false`"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Javadoc mentions \"usually like 3-letters shortcut\" but some implementations use 2-letter shortcuts (\"ac\", \"cc\", \"rt\", \"te\", \"pc\", \"ci\", \"ro\"). Consider updating documentation to reflect actual usage pattern.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": " Catching generic RuntimeException is too broad. The implementation throws IllegalArgumentException specifically - catch that instead for more precise testing.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37634__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37634__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/38446": {
@@ -5020,6 +5112,50 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR38446__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR38446__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "After creating the RecoveryAuthnCodesCredentialModel, consider setting its id from the stored credential (e.g., myUser.recoveryCodes.getId()); otherwise getId() will be null and downstream removal by id (e.g., removeStoredCredentialById in the authenticator flow) may not work.",
+          "severity": "Low",
+          "matched_candidate": "getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID. The stored credential has an ID (set in updateCredential at line 195), but createFromValues() does not set an ID on the model, and the stored ID is never propagated. Downstream, RecoveryAuthnCodesFormAuthenticator line 86-87 calls removeStoredCredentialById(recoveryCodeCredentialModel.getId()) \u2014 passing null here will fail or cause unexpected behavior when all codes are exhausted",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same core issue: after creating RecoveryAuthnCodesCredentialModel, the ID is not set from the stored credential, resulting in getId() returning null. This causes problems downstream when removeStoredCredentialById is called with a null ID. The candidate provides more detail about the flow (mentioning createFromValues, line numbers, and the specific failure scenario), but both are describing the identical bug - the missing ID propagation from stored credential to the model."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The test provider's isValid() for recovery codes accepts ANY code via anyMatch(), whereas the production RecoveryAuthnCodesCredentialProvider.isValid() only validates against the NEXT sequential code (getNextRecoveryAuthnCode), allowing out-of-order codes to be accepted"
+        },
+        {
+          "candidate": "isValid() never consumes/removes the used code \u2014 the production provider calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation. This means the same code can be reused indefinitely and tests using this provider won't catch regressions in one-time-use behavior"
+        },
+        {
+          "candidate": "For federated users, when all recovery codes are exhausted, the authenticator calls removeStoredCredentialById() which delegates to local storage (UserCredentialStore), not federated storage. The credential lives in the federated user storage provider, so this removal silently fails \u2014 the exhausted credential persists in federated storage, leading to users being stuck in a loop of regenerating codes without the old ones being cleaned up"
+        },
+        {
+          "candidate": "The updateCredential() method stores raw recovery codes JSON in the credentialData field, but in the standard RecoveryAuthnCodesCredentialModel, credentialData holds metadata and secretData holds hashed codes. The inconsistent use of credentialData for two different purposes (raw codes in storage vs metadata in standard model) is fragile and would silently break if the storage format were changed to match the standard model"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Unsafe raw List deserialization without type safety. Calling Optional.get() directly on the Optional returned by RecoveryAuthnCodesUtils.getCredential(user) without checking isPresent() can lead to a NoSuchElementException if the Optional is empty.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR38446__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR38446__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/36882": {
@@ -6180,6 +6316,39 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR36882__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR36882__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic (UpgradeLogicFactory, RecreateOnImageChangeUpgradeLogic, AlwaysRecreateUpgradeLogic) makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons \u2014 it never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments."
+        },
+        {
+          "candidate": "The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode (it only calls `System.exit` when NOT in test mode). The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure."
+        },
+        {
+          "candidate": "`testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand (`UpdateCompatibilityMetadata.NAME`), but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Incorrect method call for exit codes. The picocli.exit() method calls System.exit() directly, which is problematic:",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR36882__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36882__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/36880": {
@@ -7777,6 +7946,57 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR36880__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR36880__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In getClientsWithPermission(String scope), iterating resourceStore.findByType(server, AdminPermissionsSchema.CLIENTS_RESOURCE_TYPE) and returning resource.getName() will only ever consider the type-level 'Clients' resource (per-client resources have no type) and return its name, while AvailableRoleMappingResource#getRoleIdsWithPermissions expects actual client IDs to pass to realm.getClientById, which can lead to incorrect behavior or a null client and subsequent failures.",
+          "severity": "High",
+          "matched_candidate": "getClientsWithPermission uses findByType(server, \"Clients\", ...) which only finds the all-clients type resource, not individual per-client resources created via getOrCreateResource (which never calls resource.setType()), making per-client permissions invisible",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: getClientsWithPermission uses findByType to search for client resources, but this only finds the type-level 'Clients' resource, not the individual per-client resources. The golden comment explains that per-client resources have no type set, and the candidate explicitly states that getOrCreateResource never calls setType(), making per-client permissions invisible. Both describe the same bug with the same root cause - the type-based search misses per-client resources because they lack a type."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "getResourceTypeResource() can return null but the result is passed directly to findByResource(server, resource) without a null check, causing potential NullPointerException since PolicyStore.findByResource contract requires non-null resource parameter"
+        },
+        {
+          "candidate": "When all-clients permission exists, the type resource name \"Clients\" is added to the result set, causing AvailableRoleMappingResource.getRoleIdsWithPermissions to call realm.getClientById(\"Clients\") which returns null and causes NPE on .getRolesStream()"
+        },
+        {
+          "candidate": "canMapRoles, canMapCompositeRoles, and canMapClientScopeRoles methods do not check the MANAGE_CLIENTS admin role, preventing users with MANAGE_CLIENTS realm admin role from mapping roles unless they have explicit fine-grained permissions"
+        },
+        {
+          "candidate": "Javadoc for requireView() incorrectly says 'Returns {@code true} if canView() returns true' but the method is void and throws ForbiddenException on failure"
+        },
+        {
+          "candidate": "Javadoc for requireViewClientScopes() incorrectly says it returns a boolean but the method is void and throws ForbiddenException on failure"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Inconsistent feature flag bug causing orphaned permissions. The AdminPermissions event listener, responsible for cleaning up permissions upon role, client, or group removal, is incorrectly guarded by the ADMIN_FINE_GRAINED_AUTHZ (V1) feature flag. This is inconsistent with other methods in the class that use ADMIN_FINE_GRAINED_AUTHZ_V2. Consequently, if ADMIN_FINE_GRAINED_AUTHZ_V2 is enabled but V1 is not, the permission cleanup logic will not execute, leading to orphaned permission data. Cleanup should occur regardless of which fine-grained authorization version is enabled.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "In hasPermission(ClientModel client, String scope), the resource lookup uses findByName(server, client.getId(), server.getId()), but AdminPermissionsSchema.getOrCreateResource creates per-client resources with the owner set to resourceServer.getClientId(), so this lookup will never find those resources and will always fall back to the 'all-clients' resource, effectively ignoring client-specific permissions.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 5,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR36880__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36880__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/37038": {
@@ -9226,6 +9446,43 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37038__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37038__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Incorrect permission check in canManage() method",
+          "severity": "High",
+          "matched_candidate": "GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, allowing users with only VIEW permission to pass canManage() checks, creating a privilege escalation vulnerability",
+          "confidence": 0.95,
+          "reasoning": "Both the golden comment and candidate issue identify the same underlying problem: the canManage() method has an incorrect permission check. The golden comment states there's an 'incorrect permission check' while the candidate provides more detail, explaining that the method checks for both VIEW and MANAGE scopes when it should only check MANAGE, leading to a privilege escalation vulnerability. The candidate is describing the specific nature of the incorrect permission check mentioned in the golden comment."
+        },
+        {
+          "golden_comment": "In getGroupIdsWithViewPermission, hasPermission is called with groupResource.getId() and the same groupResource.getId() is added to granted, but hasPermission resolves resources by name (treating the argument as a group id) and the GroupPermissionEvaluator contract says this method returns group IDs that are later used as UserModel.GROUPS and in getUsersCount group filters. This mismatch means per-group VIEW_MEMBERS/MANAGE_MEMBERS permissions may not yield the expected group IDs for filtering and counts, and evaluation may effectively only look at the type-level 'all-groups' resource; consider revisiting whether this should operate on the underlying group ids (resource names) instead so it aligns with the JPA queries and the interface contract.",
+          "severity": "High",
+          "matched_candidate": "getGroupIdsWithViewPermission() uses groupResource.getId() (authorization resource server's internal UUID) instead of groupResource.getName() (actual group model ID), causing hasPermission() to fail finding resources by name and always fall through to 'all-groups' check",
+          "confidence": 0.95,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same underlying bug: getGroupIdsWithViewPermission() incorrectly uses groupResource.getId() (the authorization resource server's internal UUID) when it should use groupResource.getName() (the actual group model ID). Both explain that this causes hasPermission() to fail to find resources properly since it resolves by name, resulting in the method falling back to only checking the type-level 'all-groups' resource instead of per-group permissions. The candidate is more concise but captures the same core issue about the ID/name mismatch and its consequence on permission evaluation."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "ApiUtil.handleCreatedResponse(response) is called on responses wrapped in try-with-resources, causing double assertion of status 201 and double closing of response (at lines 73-77, 195-199, and 261-265)"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 1,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37038__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37038__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/33832": {
@@ -10709,6 +10966,47 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR33832__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR33832__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Dead code exists where ASN1Encoder instances are created and written to, but their results are immediately discarded. The actual encoding is performed by new ASN1Encoder instances created in the subsequent return statement, rendering the earlier operations useless.",
+          "severity": "Low",
+          "matched_candidate": "Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.",
+          "confidence": 1.0,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same problem: ASN1Encoder instances are created and written to on lines 114-115, but their results are immediately discarded because the actual encoding happens in subsequent lines (117-121 per candidate, or in a return statement per golden). Both recognize this as dead/useless code. The candidate provides more specific line numbers and adds context about it being wasteful allocations and possibly leftover debugging code, but the core issue identified is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking."
+        },
+        {
+          "candidate": "The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures. This means bugs specific to handling full-size ES384/ES512 r/s values (e.g., off-by-one in `integerToBytes`) would not be caught."
+        },
+        {
+          "candidate": "In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name (e.g., `\"Ignored crypto providers: FooProvider, \"`). This produces a malformed log message."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Returns wrong provider (default keystore instead of BouncyCastle)",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR33832__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR33832__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/40940": {
@@ -12091,6 +12389,47 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR40940__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR40940__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The reader thread isn\u2019t waited for; flipping deletedAll to true and asserting immediately can race and miss exceptions added just after the flag change, making this test flaky.",
+          "severity": "Medium",
+          "matched_candidate": "Reader thread is never joined before the assertion at line 157, allowing in-flight groups() calls to throw exceptions after the assertion has passed, causing the test to silently miss failures",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same race condition: the reader thread is not properly synchronized (not waited for/joined) before assertions are made, which allows exceptions thrown by the thread to be missed after the test has already passed. The golden comment mentions 'flipping deletedAll to true and asserting immediately' while the candidate mentions 'never joined before the assertion at line 157', but both describe the same underlying synchronization bug that makes the test flaky."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "getSubGroupsStream(String, Integer, Integer) at line 256 calls modelSupplier.get().getSubGroupsStream(...) without a null check, which will throw an NPE under the same concurrent-deletion scenario that this PR fixes in getSubGroupsCount()"
+        },
+        {
+          "candidate": "getSubGroupsStream overload at line 262 calls modelSupplier.get() without a null check, risking NPE in concurrent-deletion scenarios"
+        },
+        {
+          "candidate": "getSubGroupsStream overload at line 268 calls modelSupplier.get() without a null check, risking NPE in concurrent-deletion scenarios"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Returning null from getSubGroupsCount() violates the GroupModel contract (Javadoc says it never returns null) and may lead to NPEs in callers that expect a non-null count.",
+          "severity": "Critical"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR40940__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR40940__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1": {
@@ -13477,6 +13816,46 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak-greptile__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak-greptile__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "ConditionalPasskeysEnabled() called without UserModel parameter",
+          "severity": "Medium",
+          "matched_candidate": "Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the method signature in parent class UsernamePasswordForm was changed to `isConditionalPasskeysEnabled(UserModel user)` requiring one parameter. The call should be `isConditionalPasskeysEnabled(context.getUser())`",
+          "confidence": 1.0,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same problem: the method `ConditionalPasskeysEnabled()` (or `isConditionalPasskeysEnabled()`) is being called without the required `UserModel` parameter. The candidate provides more detail about the context (UsernameForm.java, parent class, and suggested fix), but the core issue is identical - a method that now requires a UserModel parameter is being called without one."
+        },
+        {
+          "golden_comment": "With isConditionalPasskeysEnabled(UserModel user) requiring user != null, authenticate(...) will not call webauthnAuth.fillContextForm(context) on the initial login page where context.getUser() is still null, so conditional passkey UI will not be set up for first-time passkey login. Consider whether this should also be enabled when no user has been selected yet so ID-less passkey authentication on the initial login form continues to work.",
+          "severity": "Medium",
+          "matched_candidate": "Regression: passkeys no longer set up during initial login. In `authenticate()`, the new code calls `fillContextForm` only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This means WebAuthn challenge, rpId, and conditional UI attributes are never set on the login form for first-time visitors (when `context.getUser() == null`), breaking passkey conditional UI on the initial login page",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same core issue: the authenticate() method's new conditional check `isConditionalPasskeysEnabled(context.getUser())` requires a non-null user, which prevents `fillContextForm` from being called on the initial login page where `context.getUser()` is null. This breaks conditional passkey UI for first-time/ID-less passkey authentication. The candidate provides more technical detail about what gets broken (WebAuthn challenge, rpId, conditional UI attributes), while the golden comment is more concise, but they both describe the same regression and its impact on initial login passkey functionality."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null`, but the new code calls it only when `user != null`. This means if a user enters an invalid username during initial login with passkeys enabled, the passkey UI elements will not be re-rendered on the error page"
+        },
+        {
+          "candidate": "Potential NullPointerException in `fillContextForm`: The refactored code at line 103 passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. Since `shouldShowWebAuthnAuthenticators` is now `protected` and designed for override, a future subclass could override it to return `true` without a null-user guard, causing NPE where `user` would be null"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak-greptile__tusk__PR1__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak-greptile__tusk__PR1__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/93824": {
@@ -15412,6 +15791,56 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR93824__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR93824__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Inconsistent metric tagging with 'shard' and 'shards'",
+          "severity": "Medium",
+          "matched_candidate": "Inconsistent metrics tag key: \"shards\" (plural) is used at line 199, while all other metrics in the same function use \"shard\" (singular), causing this metric to be silently missing from dashboards or alerts that filter on the shard tag",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same underlying issue: inconsistent use of 'shard' vs 'shards' as a metric tag key. The golden comment states the inconsistency exists between these two tag names, while the candidate provides more specific detail about where the inconsistency occurs (line 199 uses 'shards' plural while other metrics use 'shard' singular) and explains the impact (metrics missing from dashboards/alerts). These are describing the exact same bug with different levels of detail."
+        },
+        {
+          "golden_comment": "Breaking out of the loop when the deadline has elapsed can skip terminating remaining flusher processes, potentially leaving them running after shutdown; consider ensuring termination is attempted even if the deadline is exceeded.",
+          "severity": "Medium",
+          "matched_candidate": "When deadline is exceeded during join() at line 340, the break on line 341 exits the loop entirely, skipping terminate() for all remaining processes, leaving hung processes alive as zombies until parent exits",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same issue: breaking out of the loop when the deadline is exceeded (line 341) causes the code to skip the terminate() call for remaining processes. The golden comment notes this could leave processes running after shutdown, while the candidate is more specific about zombie processes. Both are describing the same underlying bug - incomplete cleanup of flusher processes when the deadline elapses."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Each sub-buffer's get_memory_info() at line 300 calls iter_cluster_memory_usage() which queries Redis INFO on all nodes, making N redundant Redis INFO roundtrips since all sub-buffers share the same Redis cluster, adding unnecessary latency to submit() calls on the hot path"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Fixed sleep in tests can be flaky; wait on condition instead",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Because flusher processes are created via multiprocessing.get_context('spawn').Process, they are instances of multiprocessing.context.SpawnProcess, which on POSIX is not a subclass of multiprocessing.Process, so this isinstance check will always be false and hung processes won't be killed here.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "Sleep in test_consumer.py won\u2019t actually wait because time.sleep was monkeypatched above; consider restoring sleep or using a different sync to ensure the flusher has time to process.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 1,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR93824__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR93824__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5": {
@@ -17295,6 +17724,45 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR5__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR5__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Using zip(error_ids, events.values()) assumes the get_multi result preserves the input order; dict value order is not guaranteed to match error_ids, so event data can be paired with the wrong ID (missing nodes also shift alignment).",
+          "severity": "Low",
+          "matched_candidate": "fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` in nodestore/base.py:205 appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list. This silently pairs wrong error IDs with wrong event data (title, message, timestamp), corrupting the error context sent to the AI summarizer",
+          "confidence": 0.95,
+          "reasoning": "Both issues identify the same core problem: using zip(error_ids, events.values()) incorrectly assumes that the dictionary values from get_multi() will be in the same order as the input error_ids list. The golden comment notes that dict value order is not guaranteed and missing nodes shift alignment. The candidate issue provides more implementation detail about how the misalignment occurs (via cache_items update in nodestore/base.py), but fundamentally describes the same bug - error IDs being paired with wrong event data due to order assumptions."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds like 1713000000.0) are compared directly against replay segment event timestamps (`event.get('timestamp', 0)`). Standard rrweb replay events use Unix epoch milliseconds (like 1713000000000). This unit mismatch means all error events will always appear to have timestamps far earlier than replay events, causing them to be yielded before any replay breadcrumbs rather than interleaved in chronological order"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Breaking changes in error response format",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Detector validator uses wrong key when updating type",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR5__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR5__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1": {
@@ -19163,6 +19631,55 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Django querysets do not support negative slicing",
+          "severity": "High",
+          "matched_candidate": "Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), `queryset[start_offset:stop]` will raise AssertionError('Negative indexing is not supported.'). The comment 'The underlying Django ORM properly handles negative slicing automatically' is incorrect.",
+          "confidence": 0.95,
+          "reasoning": "Both the golden comment and candidate issue identify the same underlying problem: Django querysets do not support negative slicing/indexing. The candidate provides more technical detail about how this manifests (via AssertionError when cursor.offset < 0), and explicitly calls out an incorrect comment in the code, but the core issue is identical - negative slicing is not supported by Django querysets."
+        },
+        {
+          "golden_comment": "When requests are authenticated with API keys or org auth tokens (which have user_id=None), organization_context.member is None. Line 71 attempts to access organization_context.member.has_global_access without checking if member is None, causing an AttributeError crash when optimized_pagination=true is used, even though the request passed all permission checks with valid org:write scope.",
+          "severity": "High",
+          "matched_candidate": "organization_context.member can be None (declared as `RpcOrganizationMember | None = None` at model.py:346) when the user has no membership in the org. Accessing `.has_global_access` on None raises AttributeError. This executes on EVERY request to this endpoint (not just optimized ones) because the `enable_advanced` check runs unconditionally before the `use_optimized` guard.",
+          "confidence": 0.85,
+          "reasoning": "Both issues identify the same core bug: organization_context.member can be None, and line 71 accesses .has_global_access without a None check, causing an AttributeError. The golden comment specifies this happens with API keys/org auth tokens (user_id=None), while the candidate mentions users with no org membership. Both describe the same technical failure (AttributeError on None.has_global_access). The candidate has a minor inaccuracy about when it executes (claims EVERY request vs just optimized_pagination=true), but the fundamental bug being identified is identical."
+        },
+        {
+          "golden_comment": "get_item_key assumes a numeric key, but the paginator is used with order_by=-datetime in the audit logs endpoint; calling math.floor/ceil on a datetime will raise a TypeError.",
+          "severity": "High",
+          "matched_candidate": "OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The only caller (organization_auditlogs.py) uses order_by='-datetime', meaning self.key is 'datetime' and getattr(item, self.key) returns a datetime. Compare with DateTimePaginator.get_item_key (line 235) which correctly converts via float(value.strftime('%s.%f')) before calling math.floor().",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: OptimizedCursorPaginator.get_item_key incorrectly calls math.floor/ceil on a datetime object (when used with order_by=-datetime in audit logs), which raises a TypeError. The candidate provides more detail by referencing the correct implementation in DateTimePaginator, but the core bug is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter (e.g., `datetime__lte=<int>`), producing incorrect results or a database type error. Compare with DateTimePaginator.value_from_cursor (line 238-241) which converts the numeric cursor value back to a timezone-aware datetime."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Importing non-existent OptimizedCursorPaginator",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 4,
+      "tp": 3,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.75,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR1__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR1__20260410/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/97529": {
@@ -20540,6 +21057,44 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR97529__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR97529__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "A race condition in BuildIndex allows multiple goroutines to concurrently build the same expensive index for the same key. This is caused by moving the b.cacheMu lock from protecting the entire function to only protecting the final cache assignment. ",
+          "severity": "High",
+          "matched_candidate": "Race condition from lock scope reduction: The old code held `cacheMu.Lock()` for the entire `BuildIndex` function, serializing all concurrent calls. The new code only locks the final cache write. This means two concurrent callers (e.g., a Search request and a handleEvent goroutine both hitting `getOrCreateIndex` for the same unindexed key) can simultaneously build an index for the same key. For file-based indexes, the second `bleve.New(dir, mapper)` call will fail because the directory already exists. For memory indexes, duplicate work is done and the first result is silently overwritten.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same core issue: reducing the lock scope in BuildIndex from protecting the entire function to only protecting the cache write creates a race condition where multiple goroutines can concurrently build the same index for the same key. The candidate provides more implementation details (file vs memory indexes, specific failure modes), but the fundamental bug is identical - the lock reduction allows duplicate expensive index building work."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Data race on shared `err` variable in concurrent goroutines: The `err` variable from `stats, err := s.storage.GetResourceStats(...)` on line 181 is captured by reference in all `group.Go` closures. Multiple goroutines concurrently write to this same `err` via `_, _, err = s.build(...)` and then read it via `return err`. This is both a data race (detected by `-race`) and a correctness bug \u2014 a goroutine may return another goroutine's error (or nil) instead of its own."
+        },
+        {
+          "candidate": "Data race on `totalBatchesIndexed`: This variable is incremented from multiple goroutines launched via `errgroup` without any synchronization. With `initWorkers > 1`, concurrent increments are a data race."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Calling s.search.TotalDocs() here may race with concurrent index creation: TotalDocs iterates b.cache without synchronization, and the event watcher goroutine started just above could trigger BuildIndex writes concurrently, potentially causing a concurrent map read/write panic.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR97529__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR97529__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/80168": {
@@ -21905,6 +22460,47 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR80168__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR80168__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "MetricAlertDetectorHandler inherits from StatefulDetectorHandler but only contains pass, failing to implement its required abstract methods: counter_names (property), get_dedupe_value(), get_group_key_values(), and build_occurrence_and_event_data(). This will cause a TypeError at runtime when the class is instantiated.",
+          "severity": "High",
+          "matched_candidate": "MetricAlertDetectorHandler extends StatefulDetectorHandler but doesn't implement required abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), causing TypeError when instantiated",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: MetricAlertDetectorHandler inherits from StatefulDetectorHandler but fails to implement the required abstract methods (counter_names, get_dedupe_value, get_group_key_values, and build_occurrence_and_event_data), which will cause a TypeError at runtime when instantiated. The wording is slightly different but they describe the identical problem with the same class, same parent class, same missing methods, and same runtime consequence."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In TestEvaluateGroupKeyValue.test_dedupe, expected occurrence uses group_key='val1' but evaluate_group_key_value is called with group_key='group_key', creating mismatched fingerprints that aren't caught because IssueOccurrence.__eq__ only compares id field"
+        },
+        {
+          "candidate": "In test_state_results_multi_group, expected occurrence for group_2 is built with value=6 but actual data packet has group_2=10, test passes incorrectly because mock doesn't vary output based on value parameter"
+        },
+        {
+          "candidate": "In test_state_results, detector.detector_handler creates a new handler instance on each property access, so the handler used in process_detectors differs from the one used to build expected occurrence, creating fragile test that works only by coincidence"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Docstring says this returns a list of DetectorEvaluationResult, but the method now returns a dict keyed by DetectorGroupKey. Consider updating the docstring to match the new return type.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR80168__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80168__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/80528": {
@@ -23179,6 +23775,37 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR80528__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR80528__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The function modifies the config variable to include display values but then returns the original monitor.config instead of the modified version.",
+          "severity": "High",
+          "matched_candidate": "In `get_monitor_environment_context`, line 168 returns `monitor_environment.monitor.config` (the original config) instead of the local `config` variable, causing the `schedule_type` override from line 162 to be silently discarded",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same bug: the function modifies a local 'config' variable but returns the original unmodified 'monitor.config' instead. The golden comment describes it generally as 'modified version' vs 'original', while the candidate is more specific about line 168 and the schedule_type override on line 162, but they're pointing to the identical underlying issue."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "The code fetches MonitorCheckIn objects by ID when the required data already exists in previous_checkins. This creates an unnecessary database query.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 1,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 0,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR80528__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80528__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/77754": {
@@ -24806,6 +25433,58 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR77754__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR77754__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Shared mutable default in dataclass timestamp",
+          "severity": "Medium",
+          "matched_candidate": "Classic Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition (module import) time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` (which doesn't pass `queued`) will share the same timestamp \u2014 the time the module was first imported. The fix is `queued: datetime = field(default_factory=timezone.now)`.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: a mutable default argument problem in a dataclass where `timezone.now()` is evaluated at class definition time rather than instance creation time, causing all instances to share the same timestamp value. The candidate provides more technical detail about when the evaluation occurs and suggests the correct fix using `field(default_factory=timezone.now)`, while the golden comment is more concise. They are describing the exact same bug."
+        },
+        {
+          "golden_comment": "The method name has a typo: test_from_dict_inalid_data should be test_from_dict_invalid_data.",
+          "severity": "Low",
+          "matched_candidate": "Test method name `test_from_dict_inalid_data` has a typo \u2014 missing 'v' in 'invalid'. While the test still runs (Python doesn't validate spelling), this makes it harder to find via search and indicates insufficient review.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same typo in the test method name: 'inalid' should be 'invalid' (missing 'v'). The candidate provides additional context about the impact, but the core issue - the misspelling of 'invalid' as 'inalid' in the method name - is identical."
+        },
+        {
+          "golden_comment": "Method name says 'empty_array' but tests empty dict - consider renaming to 'test_from_dict_empty_dict' for clarity.",
+          "severity": "Low",
+          "matched_candidate": "Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array. The name is misleading and suggests the wrong data structure is being tested.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: the test method is named 'test_from_dict_empty_array' but it actually tests an empty dictionary ({}), not an empty array. The golden comment suggests renaming to 'test_from_dict_empty_dict' for clarity, while the candidate describes the name as misleading. Both are pointing to the same naming inconsistency problem."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `assignment_source` parameter was added to the abstract `sync_status_outbound` method, but this PR is about assignee sync cycles \u2014 status sync is unrelated."
+        },
+        {
+          "candidate": "`ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` (line 161) does not accept `**kwargs`, so any future caller passing `assignment_source` to it will get a TypeError. The other concrete implementations (jira, jira_server) use `**kwargs` so they silently absorb it, but ExampleIntegration will crash."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "to_dict() returns a datetime for queued; if this dict is passed in task kwargs (e.g., via apply_async), JSON serialization may fail depending on the serializer, which can cause enqueue errors.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 4,
+      "tp": 3,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.75,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR77754__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR77754__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/95633": {
@@ -26367,6 +27046,66 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR95633__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR95633__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The test test_thread_queue_parallel_error_handling has a docstring that doesn't match the test implementation.",
+          "severity": "Low",
+          "matched_candidate": "test_thread_queue_parallel_error_handling's docstring contradicts its assertion - docstring claims it tests that errors don't block commits, but assertion checks that no commits happened",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: a mismatch between the test's docstring and its actual implementation/assertion. The golden comment states the docstring doesn't match the implementation, while the candidate provides more specific detail about the nature of the contradiction (docstring says errors don't block commits, but assertion checks no commits happened). The candidate is describing the same problem with more specificity."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "close() joins the commit thread before shutting down the worker pool, causing processed items after final commit to have uncommitted offsets, leading to duplicate processing on restart"
+        },
+        {
+          "candidate": "test_thread_queue_parallel_error_handling is flaky because it relies on timing - passes only because commit loop waits 1 second but test waits 0.2s"
+        },
+        {
+          "candidate": "test_thread_queue_parallel_error_handling assertion is incorrect - offsets will be committed for both failed and successful messages due to finally block in OrderedQueueWorker.run() line 149"
+        },
+        {
+          "candidate": "SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' but queues are created without maxsize parameter (line 185), making them unbounded with no backpressure mechanism"
+        },
+        {
+          "candidate": "SimpleQueueProcessingStrategy docstring claims 'Only commits offsets after successful processing' but finally block in OrderedQueueWorker.run() always calls complete_offset regardless of exceptions"
+        },
+        {
+          "candidate": "Unbounded queues in SimpleQueueProcessingStrategy can cause unbounded memory growth when workers are slow and producer is fast"
+        },
+        {
+          "candidate": "OffsetTracker._get_partition_lock accesses self.partition_locks without synchronization, relying on CPython GIL rather than language spec guarantees"
+        },
+        {
+          "candidate": "OffsetTracker.get_committable_offsets iterates list(self.all_offsets.keys()) without holding any lock while worker threads may be modifying all_offsets via defaultdict auto-creation"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The queue.shutdown() method with 'immediate=False' parameter may not exist in the standard Python queue module. This could cause AttributeError at runtime. Verify the correct API or implement a custom shutdown mechanism.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The magic number 50 for max_wait is used repeatedly throughout the tests. Consider extracting this as a named constant to improve maintainability.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 9,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 8,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.1111111111111111,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR95633__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR95633__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2": {
@@ -28206,6 +28945,50 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR2__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR2__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "OptimizedCursorPaginator negative-offset branch slices QuerySet with a negative start index",
+          "severity": "Critical",
+          "matched_candidate": "The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing \u2014 it raises AssertionError(\"Negative indexing is not supported.\"). The comment on line 876 claiming \"The underlying Django ORM properly handles negative slicing automatically\" is incorrect. Any request to the audit logs endpoint with optimized_pagination=true and a crafted cursor with a negative offset will crash with a 500 error.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same underlying issue: the OptimizedCursorPaginator uses negative indexing on a Django QuerySet (queryset[negative_offset:stop]), which is not supported by Django ORM. The golden comment describes it as 'slices QuerySet with a negative start index' while the candidate provides more detail about the crash and incorrect comment, but they're pointing to the exact same bug."
+        },
+        {
+          "golden_comment": "BasePaginator negative-offset branch slices QuerySet with a negative start index",
+          "severity": "High",
+          "matched_candidate": "The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing \u2014 it raises AssertionError(\"Negative indexing is not supported.\"). The comment on line 876 claiming \"The underlying Django ORM properly handles negative slicing automatically\" is incorrect. Any request to the audit logs endpoint with optimized_pagination=true and a crafted cursor with a negative offset will crash with a 500 error.",
+          "confidence": 1.0,
+          "reasoning": "Both the golden comment and candidate issue identify the same underlying bug: BasePaginator attempts to slice a QuerySet with a negative start index (negative offset), which is not supported by Django's ORM. The golden comment succinctly states 'slices QuerySet with a negative start index' while the candidate provides more detail about the consequences (AssertionError, 500 error) and context (optimized_pagination, crafted cursor), but they are describing the exact same code defect."
+        },
+        {
+          "golden_comment": "OptimizedCursorPaginator.get_item_key uses floor/ceil on a datetime key (order_by='-datetime'), causing TypeError.",
+          "severity": "High",
+          "matched_candidate": "OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. This paginator is used with order_by=\"-datetime\" on AuditLogEntry, so getattr(item, self.key) returns a datetime. Compare to DateTimePaginator.get_item_key which correctly converts to float first via `float(value.strftime(\"%s.%f\")) * self.multiplier` before applying math.floor/ceil.",
+          "confidence": 1.0,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same bug: OptimizedCursorPaginator.get_item_key incorrectly uses math.floor()/math.ceil() on a datetime object (when order_by='-datetime'), which causes a TypeError. The candidate provides more detail about the context (AuditLogEntry) and the correct approach (DateTimePaginator's conversion to float first), but the core issue is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The old code had a safety check that skipped segments exceeding max_segment_spans (default 1001) during flush, with metrics and error logging. This check was completely removed. While the Lua script now caps spans at 1000 during insert, the flush-side guard protected against edge cases (e.g., race conditions between concurrent inserts, or spans inserted before the eviction logic was deployed). The max_segment_spans attribute is still initialized on the class but is now dead code."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 1,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR2__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR2__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3": {
@@ -29901,6 +30684,51 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR3__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR3__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Using Python\u2019s built-in hash() to build cache keys is non-deterministic across processes (hash randomization), so keys won\u2019t match across workers and invalidate_upsampling_cache may fail to delete them. Use a deterministic serialization of project_ids for the cache key.",
+          "severity": "Low",
+          "matched_candidate": "Python's built-in hash() is randomized per-process (via PYTHONHASHSEED), causing different worker processes to compute different hash values for the same project IDs in shared Redis/Memcached cache, resulting in cache misses across processes",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same core issue: Python's hash() function is non-deterministic across processes due to hash randomization (PYTHONHASHSEED), which causes cache key mismatches when using hash() to build cache keys in a multi-process environment with shared cache (Redis/Memcached). The golden comment mentions that invalidate_upsampling_cache may fail to delete keys, while the candidate mentions cache misses - these are both consequences of the same root cause. Both recommend using deterministic serialization instead."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Cache key doesn't include dataset or request context, causing the same cache entry to be reused across different query types for the same org+projects, potentially returning inconsistent results"
+        },
+        {
+          "candidate": "Behavior changed from any() to all() for project allowlist checking, causing queries spanning both allowlisted and non-allowlisted projects to return raw sample counts instead of upsampled estimates"
+        },
+        {
+          "candidate": "Validation preventing 'timestamp' in raw_groupby was removed along with ParseError import, allowing users to group by 'timestamp' which was previously explicitly disallowed"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "sample_rate = 0.0 is falsy and skipped",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The upsampling eligibility check passes the outer dataset instead of the actual dataset used by scoped_dataset. In paths where the query ultimately runs against discover (e.g., dashboard split) while the original dataset is metrics, upsampling may be skipped even when all projects are allowlisted.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 3,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR3__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR3__20260410/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/103633": {
@@ -31330,6 +32158,47 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR103633__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR103633__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The test comment says the cached permissions 'allow access', but the map stores false for dashboards:uid:dash1, so checkPermission will still treat this scope as not allowed.",
+          "severity": "Low",
+          "matched_candidate": "Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. Line 981 comments \"Allow access to the dashboard to prove this is not checked\" but line 982 sets the permission cache value to `false` (`map[string]bool{\"dashboards:uid:dash1\": false}`). Since `checkPermission` checks `scopeMap[t.scope(req.Name)]` and `false` is falsy, the permission cache would ALSO deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: the test comment on line 981 claims to 'allow access' to prove the permission cache is not checked, but line 982 actually sets the permission to false in the cache map. This means the permission cache would also deny access, making the test ineffective at proving what it claims to test. The golden comment is more concise while the candidate provides more detail, but they both point to the same fundamental problem with the test's logic."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource."
+        },
+        {
+          "candidate": "The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance (both passed the same `cache` in `NewService`) but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix `ns.perm_UID_ACTION_NAME_PARENT` overlaps with the permission cache key prefix `ns.perm_UID_ACTION`. If a denial key accidentally matches a permission key (e.g., if a user UID contains underscore-delimited segments resembling `UID_ACTION_NAME_PARENT`), `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs."
+        },
+        {
+          "candidate": "After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"` (line 137), suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates two semantically different situations (cache miss vs. cache hit with stale data) into the same metric label, making it difficult to distinguish genuine cache misses from outdated-cache fallbacks when monitoring."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The Check operation exhibits asymmetric cache trust logic: cached permission grants are trusted and returned immediately, but cached denials from the same permission cache are ignored, leading to a fresh database lookup. This allows stale cached grants to provide access to revoked resources, posing a security risk. ",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR103633__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR103633__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/67876": {
@@ -32958,6 +33827,53 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR67876__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR67876__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Null reference if github_authenticated_user state is missing",
+          "severity": "Medium",
+          "matched_candidate": "error() helper function accesses org.organization without verifying org type or structure, causing potential crash when active_organization is None or unexpected type",
+          "confidence": 0.85,
+          "reasoning": "Both issues identify the same underlying problem: accessing attributes on an object that could be None or missing without proper validation. The golden comment refers to 'github_authenticated_user state is missing' causing a null reference, while the candidate describes accessing 'org.organization' when 'active_organization is None'. These are describing the same type of null/None reference error in what appears to be the same codebase context - the error() helper function attempting to access properties on potentially None objects without verification."
+        },
+        {
+          "golden_comment": "OAuth state uses pipeline.signature (static) instead of a per-request random value",
+          "severity": "Medium",
+          "matched_candidate": "OAuth `state` parameter uses deterministic MD5 hash (pipeline.signature) instead of cryptographic random nonce, allowing attackers to forge state parameter and bypass CSRF protection",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same core security issue: the OAuth state parameter is using a static/deterministic value (pipeline.signature) instead of a random per-request value. The candidate provides more detail about the security implications (CSRF protection bypass via forged state parameters), but both are pointing to the exact same underlying vulnerability in the code."
+        },
+        {
+          "golden_comment": "The code attempts to access integration.metadata[sender][login] without checking for the existence of the sender key. This causes a KeyError for integrations where the sender metadata was not set during creation",
+          "severity": "High",
+          "matched_candidate": "Accessing integration.metadata[\"sender\"][\"login\"] raises KeyError when sender key is missing from metadata (integrations created before code change or through non-webhook paths)",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same underlying issue: accessing integration.metadata['sender']['login'] without first checking if the 'sender' key exists in the metadata dictionary, which causes a KeyError. The candidate provides additional context about when this occurs (integrations created before code change or through non-webhook paths), but the core problem is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "First-time GitHub app installation bypasses sender/user validation when Integration.DoesNotExist exception occurs, allowing attacker to install app without being the initiating GitHub user"
+        },
+        {
+          "candidate": "redirect_uri missing from token exchange request despite being included in authorization request, violating OAuth2 RFC 6749 Section 4.1.3 requirement"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR67876__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR67876__20260410/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/32918": {
@@ -34259,6 +35175,44 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR32918__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR32918__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Cleanup reference uses incorrect alias - should be 'idp-alias-' + i instead of 'alias'.",
+          "severity": "Medium",
+          "matched_candidate": "Test cleanup registers removal for a literal \"alias\" instead of the actual IDP alias (e.g., \"idp-alias-\" + i), causing all 20 IDPs created in the loop to never be cleaned up and potentially leak state to other tests",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same bug: the cleanup code uses the literal string 'alias' instead of the dynamically generated alias 'idp-alias-' + i. The candidate provides more context about the consequences (20 IDPs not cleaned up, potential test state leakage), but the core issue is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "getForLogin cached path returns results from a HashSet<IdentityProviderModel> which does not preserve ordering from the delegate's query, causing inconsistent behavior between cached and non-cached paths for future callers relying on delegate ordering"
+        },
+        {
+          "candidate": "In registerIDPLoginInvalidationOnUpdate, the original model is wrapped with createOrganizationAwareIdentityProviderModel while the updated model is raw, causing unnecessary cache invalidation when org-aware isEnabled() differs from the raw value"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Recursive caching call using session instead of delegate",
+          "severity": "Critical"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR32918__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR32918__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/94942": {
@@ -35613,6 +36567,41 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR94942__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR94942__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The enableSqlExpressions function has flawed logic that always returns false, effectively disabling SQL expressions unconditionally:",
+          "severity": "Critical",
+          "matched_candidate": "The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code that misleadingly suggests the flag could re-enable the feature when it cannot",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same bug: the enableSqlExpressions function has logic that always returns false, making it impossible to enable SQL expressions. The golden comment states it 'always returns false, effectively disabling SQL expressions unconditionally' while the candidate states it 'always returns false regardless of the feature flag value'. Both recognize that the feature flag check is ineffective/dead code. This is clearly the same underlying issue with semantically equivalent descriptions."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `DB.QueryFramesInto` stub method always returns an error but the caller in `sql_command.go:Execute` still appends an empty frame to results after setting the error, creating a latent bug if the guard is ever removed"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Several methods such as NewInMemoryDB().RunCommands and db.QueryFramesInto return 'not implemented'.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR94942__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR94942__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/90939": {
@@ -36874,6 +37863,37 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR90939__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR90939__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The GetWebAssets function implements an incomplete double-checked locking pattern for caching web assets. The function first checks if the cache is populated using a read lock (RLock), and if the cache is empty, it acquires a write lock to populate it. However, it fails to re-check whether the cache was populated by another goroutine while waiting to acquire the write lock.",
+          "severity": "Medium",
+          "matched_candidate": "Double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether entryPointAssetsCache was already populated by another goroutine, causing redundant I/O operations and thundering herd of file reads in non-dev mode with concurrent requests at startup",
+          "confidence": 1.0,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same bug: an incomplete double-checked locking pattern in the GetWebAssets function. Both describe that after releasing the read lock and acquiring the write lock, the code fails to re-check if another goroutine populated the cache in the meantime. The candidate provides additional detail about the consequences (redundant I/O, thundering herd) and specific line numbers, but the core issue is identical."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "In addition to the missing double-check, the function has a critical flaw in its error handling: it unconditionally assigns the fetch result to the cache (line 69: entryPointAssetsCache = result) regardless of whether the fetch succeeded or failed. When an error occurs during asset fetching, result is nil, and this nil value overwrites any previously valid cache entry.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 1,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 0,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR90939__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90939__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/80329": {
@@ -38199,6 +39219,42 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR80329__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR80329__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The code uses Error log level for what appears to be debugging information. This will pollute error logs in production. Consider using Debug or Info level instead.",
+          "severity": "Low",
+          "matched_candidate": "All diagnostic log lines in the cleanup path use r.log.Error() instead of r.log.Debug() or r.log.Info(). These are clearly informational messages (e.g., 'Annotations to clean by time', 'cleaned annotations by time') that fire on every successful cleanup cycle. In production this will flood error-level logs and likely trigger monitoring/alerting systems, making real errors harder to find.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same core issue: using Error log level for informational/debugging messages instead of Debug or Info level, which will pollute production error logs. The candidate provides more specific details about the cleanup path and mentions monitoring/alerting implications, but fundamentally addresses the identical logging level misuse problem that the golden comment describes."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The cleanup service ticker was changed from 10 minutes to 1 minute, which appears to be an accidental debug leftover. The clean() method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running. This 10x increase in cleanup frequency will also increase database load significantly in production."
+        },
+        {
+          "candidate": "The SQLite parameter limit guard checks r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit but should check len(ids) > sqliteParameterLimit. The batch size is the maximum number of IDs that can be fetched, but the actual number returned may be less. If AnnotationCleanupJobBatchSize is e.g. 1000 but only 5 IDs are returned, the code will unnecessarily build an inline SQL string instead of using parameterized queries, which is the safer approach. The guard should reflect the actual parameter count, not the configured ceiling."
+        },
+        {
+          "candidate": "Test helper createTestAnnotations hardcodes annotation IDs as int64(i + 1) via InsertMulti. Since tests share a single db.InitTestDB(t) and SQLite's auto-increment counter is not reset by DELETE FROM annotation, subsequent test cases that also insert IDs starting at 1 will collide with the auto-increment sequence on databases that enforce uniqueness on the primary key. If any test case inserts without explicit IDs, the auto-generated IDs may collide with IDs from a previous createTestAnnotations call that didn't fully clean up yet."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 1,
+      "tp": 1,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR80329__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR80329__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/90045": {
@@ -39990,6 +41046,53 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR90045__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR90045__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The context is being created with d.Log instead of the log variable that was initialized with additional context values (name, kind, method). This means those values won't be propagated to the logging context.",
+          "severity": "Medium",
+          "matched_candidate": "In Delete method (line 97), klog.NewContext uses d.Log instead of the enriched log variable that includes name, kind, and method values",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: in the Delete method at line 97, klog.NewContext is using d.Log instead of the enriched log variable that contains the additional context values (name, kind, method). The candidate issue is more specific about the line number but describes the identical problem with the same root cause and impact."
+        },
+        {
+          "golden_comment": "Bug: calling recordLegacyDuration when storage operation fails should be recordStorageDuration.",
+          "severity": "High",
+          "matched_candidate": "In Create's error path (line 45), recordLegacyDuration is called instead of recordStorageDuration, causing storage operation latency to be recorded against the legacy histogram",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same bug: on line 45 in the error path of the Create function, recordLegacyDuration is incorrectly called when recordStorageDuration should be used instead. The candidate provides more context about the consequence (storage operation latency being recorded to the wrong histogram), but the core issue is identical."
+        },
+        {
+          "golden_comment": "Inconsistency: using name instead of options.Kind for metrics recording differs from other methods.",
+          "severity": "Medium",
+          "matched_candidate": "In Delete's success path (line 106), recordStorageDuration uses 'name' (object name) instead of 'options.Kind' as the third argument, creating high-cardinality metric labels",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: in the Delete method's success path, the recordStorageDuration call uses 'name' instead of 'options.Kind' as an argument. The golden comment frames this as an inconsistency with other methods, while the candidate provides additional context about high-cardinality metrics. However, they both point to the exact same code problem - using the wrong variable (name vs options.Kind) in the same location."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In Update's error path (line 129), recordLegacyDuration is called instead of recordStorageDuration, causing storage operation latency to be recorded against the legacy histogram"
+        },
+        {
+          "candidate": "In DeleteCollection's goroutine (line 166), recordStorageDuration is called instead of recordLegacyDuration for the legacy write, causing legacy latency to be recorded in the storage histogram"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR90045__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90045__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/106778": {
@@ -41584,6 +42687,49 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR106778__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR106778__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The rendered GrafanaRuleListItem is missing the required key prop for React list items. This can cause rendering issues when the list order changes.",
+          "severity": "Medium",
+          "matched_candidate": "Missing React `key` prop on GrafanaRuleListItem in FilterView.tsx line 157. The old GrafanaRuleLoader had key={key}, but the replacement GrafanaRuleListItem omits it, while other branches (DataSourceRuleLoader, UnknownRuleListItem) all have key={key}. This will cause React reconciliation issues when the list updates (e.g., items reordering, stale state, performance degradation).",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: the missing `key` prop on GrafanaRuleListItem in a React list. The golden comment mentions it can cause rendering issues when list order changes, while the candidate provides more technical detail about React reconciliation issues, performance degradation, and stale state. The candidate also provides specific line numbers and context about other branches having the key prop. Despite different levels of detail, they are identifying the same underlying bug."
+        },
+        {
+          "golden_comment": "RuleActionsButtons is invoked with only promRule, but SilenceGrafanaRuleDrawer inside RuleActionsButtons still depends on a Grafana Ruler rule being present, so for Grafana rules coming from list views the 'Silence notifications' menu entry (now driven by Grafana Prom abilities) will toggle showSilenceDrawer without ever rendering the drawer. This means clicking 'Silence notifications' for these rules has no visible effect, even when abilities indicate silencing is allowed.",
+          "severity": "High",
+          "matched_candidate": "Silence drawer won't open in the new list view. In RuleActionsButtons.V2.tsx, the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` at line 104. When rule (the RulerRuleDTO) is undefined (coming from GrafanaRuleListItem which only passes promRule), this check is false. So even though the 'Silence notifications' menu item appears (via canSilence from the grafana prom path), clicking it sets showSilenceDrawer to true but the drawer never renders.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: in the new list view, RuleActionsButtons receives only promRule (not the full Grafana ruler rule), causing the silence drawer to not render even though the 'Silence notifications' menu item appears and can be clicked. The candidate correctly identifies the guard condition at line 104 checking for rulerRuleType.grafana.alertingRule(rule) as the cause, and explains that showSilenceDrawer gets set to true but the drawer never renders because the rule (RulerRuleDTO) is undefined. This matches the golden comment's description of the same problem with slightly different wording."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Delete menu item will never render in the new list view. AlertRuleMenu at line 165 guards the delete menu item with `canDelete && rulerRule && (...)`. Since the new list view no longer fetches ruler data, rulerRule is always undefined when coming from GrafanaRuleListItem \u2192 RuleActionsButtons.V2 \u2192 AlertRuleMenu. Even though canDelete becomes true via the new grafanaDeleteAbility path, the rulerRule guard prevents the delete option from appearing."
+        },
+        {
+          "candidate": "Pause menu item will never render in the new list view. At line 123, `canPause && rulerRuleType.grafana.rule(rulerRule)` requires rulerRule to be a grafana ruler rule. Since rulerRule is undefined when coming from the new list view, rulerRuleType.grafana.rule(undefined) returns false, so the pause option never appears \u2014 even though the user has permission via the grafana prom abilities path."
+        },
+        {
+          "candidate": "The useGrafanaPromRuleAbility hook in RuleActionsButtons.V2.tsx is called with promRule typed as Rule (from unified-alerting.ts), but the hook expects GrafanaPromRuleDTO | SkipToken. While prometheusRuleType.grafana.rule() does a structural check, it returns boolean (not a type predicate narrowing to GrafanaPromRuleDTO), so TypeScript won't narrow the type. This is likely a compile error, or if it passes due to structural compatibility, the folderUid field used inside useIsGrafanaPromRuleEditable may be missing on non-Grafana Rule objects (where folderUid is optional), leading to incorrect permission checks."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR106778__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR106778__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/107534": {
@@ -42719,6 +43865,39 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR107534__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR107534__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Double interpolation of queries in shard splitting path: runShardSplitQuery() calls applyTemplateVariables() at line 52, then splitQueriesByStreamShard() calls runSplitQuery() at line 163, which also calls applyTemplateVariables() at line 299. This causes addAdHocFilters() to unconditionally append label matchers twice via addLabelToQuery(), duplicating ad-hoc filters like env=\"prod\" in the query expression"
+        },
+        {
+          "candidate": "Test mock for applyTemplateVariables mutates input query object directly (query.expr = query.expr.replace(...)) instead of returning a new object like the real implementation ({ ...target, expr, step, legendFormat }), hiding bugs where code relies on original targets remaining unmodified"
+        },
+        {
+          "candidate": "Filter ordering change in runShardSplitQuery causes behavioral difference: new code filters queries before interpolation, so template variables that resolve to empty strings (e.g., expr: '$optional_query' \u2192 '') pass the filter but result in empty expressions being sent to splitQueriesByStreamShard, whereas old code filtered after interpolation"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The applyTemplateVariables method is called with request.filters as the third parameter, but this parameter is not used in the corresponding test setup.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR107534__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR107534__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/79265": {
@@ -44745,6 +45924,59 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR79265__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR79265__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Race condition: Multiple concurrent requests could pass the device count check simultaneously and create devices beyond the limit. Consider using a database transaction or lock.",
+          "severity": "High",
+          "matched_candidate": "TOCTOU race in device limit enforcement. CreateOrUpdateDevice first counts devices, then conditionally calls updateDevice. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit. With high concurrency this could allow significantly more devices than intended.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same race condition vulnerability in device limit enforcement. The golden comment describes how concurrent requests can pass the device count check simultaneously and exceed the limit. The candidate issue provides more technical detail (TOCTOU - Time Of Check Time Of Use) but describes the exact same problem: the gap between counting devices and creating/updating them allows concurrent requests to bypass the limit check. Both recognize this as a concurrency issue that could result in exceeding the device limit."
+        },
+        {
+          "golden_comment": "Returning ErrDeviceLimitReached when no rows were updated is misleading; the device might not exist.",
+          "severity": "Low",
+          "matched_candidate": "The updateDevice method returns ErrDeviceLimitReached when rowsAffected == 0, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its updated_at falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with ErrDeviceLimitReached instead of being re-created or handled differently. The error is misleading for the second case.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: returning ErrDeviceLimitReached when rowsAffected == 0 is misleading because it conflates multiple scenarios. The golden comment points out that the device might not exist, while the candidate issue provides more detail by explaining two specific cases (device doesn't exist vs. device exists but is outside the time range). Both are highlighting that this single error code is being used inappropriately for different failure scenarios, making it misleading. The candidate issue is more detailed but identifies the same fundamental problem."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout."
+        },
+        {
+          "candidate": "The AnonymousDeviceLimit field in FrontendSettingsDTO is int64 with json:\"anonymousDeviceLimit\" \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as number | undefined. The Go DTO has no omitempty tag, so the field is never undefined in practice. Any frontend code checking config.anonymousDeviceLimit === undefined to detect 'no limit configured' will never match; it will see 0 instead."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Anonymous authentication now fails entirely if anonDeviceService.TagDevice returns ErrDeviceLimitReached. Previously, device tagging was asynchronous and non-blocking. This change prevents anonymous users from authenticating when the device limit is reached.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "This call won\u2019t compile: dbSession.Exec(args...) is given a []interface{} where the first element is the query, but Exec\u2019s signature requires a first parameter of type string (not an interface{} splat).",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Time window calculation inconsistency: Using device.UpdatedAt.UTC().Add(-anonymousDeviceExpiration) as the lower bound but device.UpdatedAt as the current time may not match the intended logic. Consider using time.Now().UTC() consistently.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 2,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR79265__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR79265__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9": {
@@ -46135,6 +47367,47 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR9__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR9__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Thread-safety issue with lazy @loaded_locales",
+          "severity": "Low",
+          "matched_candidate": "ensure_loaded! initializes @loaded_locales ||= [] outside LOAD_MUTEX, allowing race condition where concurrent threads create separate arrays and one gets discarded, potentially causing locales to appear unloaded or skip loading entirely",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same thread-safety issue: the lazy initialization of @loaded_locales using the ||= operator occurs outside the LOAD_MUTEX, creating a race condition where concurrent threads can create separate arrays. The candidate provides more technical detail about the consequences (discarded arrays, locales appearing unloaded), but the core issue is identical - improper synchronization of the @loaded_locales initialization."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "LRU translation cache key does not include SiteSetting.default_locale, causing stale translations to be served when admin changes the site's default locale (cached translations resolved through old fallback chain continue to be returned)"
+        },
+        {
+          "candidate": "No I18n.reload! triggered on default_locale change to clear the translation cache"
+        },
+        {
+          "candidate": "translate method only loads config.locale but not fallback locales (default_locale, :en), causing fallback feature to silently fail in code paths outside controllers (Sidekiq jobs, rake tasks, console)"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Consider normalizing the input locale (e.g., to a symbol) when checking/loading here to avoid double-loading if the same locale is passed as a String vs Symbol (also applies to other locations in the PR).",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR9__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR9__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/76186": {
@@ -47506,6 +48779,41 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR76186__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR76186__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
+          "severity": "High",
+          "matched_candidate": "ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic. This is inconsistent with LoggerMiddleware which explicitly handles nil requests with guard clauses.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: the ContextualLoggerMiddleware methods access req.PluginContext without checking if req is nil first, which causes a nil pointer dereference panic. The golden comment mentions this happens via the instrumentContext function and notes it's a regression from previous middleware that handled nil requests gracefully. The candidate makes the same core observation and explicitly compares it to LoggerMiddleware which does handle nil requests properly. These are semantically identical - they identify the same bug in the same methods with the same root cause."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(), causing tests using TestLogger to silently lose all log messages when LoggerMiddleware.logRequest calls m.logger.FromContext(ctx).Info(...). The method should return the receiver (f) to preserve log capture, consistent with how the real implementation preserves the logger identity."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR76186__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR76186__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10": {
@@ -49622,6 +50930,61 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR10__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR10__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "NoMethodError before_validation in EmbeddableHost",
+          "severity": "Critical",
+          "matched_candidate": "`before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value (host is nil), `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it. The controller's `save_host` could pass nil if `params[:embeddable_host][:host]` is missing.",
+          "confidence": 0.95,
+          "reasoning": "Both the golden comment and candidate issue identify the same underlying problem: a NoMethodError occurring in the before_validation callback of EmbeddableHost when the host attribute is nil. The candidate provides more detailed explanation of the root cause (calling sub! on nil) and the scenario that triggers it, but fundamentally describes the same bug that the golden comment references."
+        },
+        {
+          "golden_comment": "The update and destroy methods in Admin::EmbeddableHostsController do not validate the existence of the EmbeddableHost record retrieved by ID. If EmbeddableHost.where(id: params[:id]).first returns nil (i.e., the host does not exist), attempting to call methods on the nil object (e.g., save_host or destroy) will result in a NoMethodError.",
+          "severity": "Medium",
+          "matched_candidate": "Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist (e.g., already deleted, invalid id). In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same underlying issue: the `update` and `destroy` methods in Admin::EmbeddableHostsController fail to validate whether `EmbeddableHost.where(id: params[:id]).first` returns nil before attempting to call methods on it. Both mention that this will result in a NoMethodError when the record doesn't exist. The candidate provides slightly more specific examples (mentioning `host.destroy` and `host.host = ...`) but describes the exact same vulnerability and failure mode as the golden comment."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist (e.g., fresh installs or installs that never configured embedding), the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached."
+        },
+        {
+          "candidate": "SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote (e.g., from a corrupted setting) will break the SQL or allow injection."
+        },
+        {
+          "candidate": "Fabricator file contents are swapped. `category_fabricator.rb` now contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators (`:category`, `:diff_category`, `:happy_category`, `:private_category`). The file names don't match their contents."
+        },
+        {
+          "candidate": "JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores (e.g., `'some_admin_model'` becomes `'some-admin_model'` instead of `'some-admin-model'`)."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "record_for_host compares lower(host) = ? but does not normalize the parameter\u2019s case, so mixed\u2011case referer hosts may fail to match even though comparison intends to be case\u2011insensitive.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Because this migration inserts embeddable_hosts rows with raw SQL, any existing embeddable_hosts values that include http:// or /https:// or path segments won\u2019t go through the EmbeddableHost model\u2019s normalization, so the new host lookup (which compares only the bare host) may fail for migrated data. Consider ensuring that migrated hosts are normalized to the same format as newly created EmbeddableHost records so existing embedding configurations keep working.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 4,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR10__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR10__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7": {
@@ -51343,6 +52706,51 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR7__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR7__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "This change for desktop/user.css changes $primary from 30% to 50% for the light theme; most other changes preserve the original $primary value and move the complement to $secondary for dark. Consider reviewing this (also applies to a similar .name change in the mobile variant).",
+          "severity": "Low",
+          "matched_candidate": "Light-theme lightness for .name changed from 30% to 50% in app/assets/stylesheets/desktop/user.scss line 522, making user names noticeably lighter/more washed out, inconsistent with nearby .username a which preserves 30%",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: the $primary lightness value for .name in the light theme was changed from 30% to 50% in desktop/user.css/scss, making it lighter/more washed out. The golden comment additionally mentions this pattern applies to mobile variant and notes that other changes preserved the original $primary value, but the core issue being flagged is identical - the problematic 30% to 50% change for .name in the light theme."
+        },
+        {
+          "golden_comment": "In topic-post.css the original code used $lightness: 70% but the replacement uses $lightness: 30% for the light theme. This makes the text significantly darker than intended.",
+          "severity": "Low",
+          "matched_candidate": "Light-theme lightness value changed from 30% to 70% (swapped with dark value) in app/assets/stylesheets/desktop/topic-post.scss line 291 for .embedded-posts link text, making it much lighter/more washed out",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: a lightness value swap between light and dark themes in the same file (topic-post.css/topic-post.scss). The golden comment states the light theme incorrectly uses 30% instead of 70%, while the candidate states it changed from 30% to 70% (swapped with dark value). Both are describing the same bug - the lightness values for light and dark themes were swapped, resulting in incorrect text darkness for the light theme. The candidate's phrasing about 'changed from 30% to 70%' describes the fix direction, while the golden describes the problem state, but they're identifying the same swap issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Light-theme lightness value changed from 70% to 30% in app/assets/stylesheets/mobile/modal.scss line 102 for .custom-message-length, opposite of desktop version, making character count text much darker on mobile light themes and lighter on mobile dark themes"
+        },
+        {
+          "candidate": "Light-theme lightness for h3 changed from 20% to 50% in app/assets/stylesheets/mobile/topic-post.scss line 182, making headings significantly more faded/grey instead of near-black"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "In .topic-meta-data h5 a, the original code had color: scale-color($primary, $lightness: 30%) but was changed to dark-light-choose(scale-color($primary, $lightness: 70%), scale-color($secondary, $lightness: 30%)). The lightness for the light theme changed from 30% to 70%, which is a dramatic inversion",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR7__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR7__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8": {
@@ -53159,6 +54567,57 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR8__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR8__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In the next action, capping the next offset at user_count can produce an empty page (e.g., total equal to limit results in offset == total, showing 2/2 with no members). This can cause confusing UX on the last page.",
+          "severity": "Medium",
+          "matched_candidate": "Off-by-one error in totalPages calculation using Math.floor(user_count / limit) + 1 instead of Math.ceil(user_count / limit), creating extra empty page when user_count is divisible by limit",
+          "confidence": 0.95,
+          "reasoning": "Both issues identify the same underlying problem: incorrect pagination logic that creates an empty last page when user_count is divisible by limit. The golden comment describes the symptom (empty page, confusing UX) caused by capping offset at user_count. The candidate identifies the root cause (off-by-one error in totalPages calculation using Math.floor + 1 instead of Math.ceil) that leads to this exact same empty page issue. These are two perspectives on the same bug."
+        },
+        {
+          "golden_comment": "HTTP method mismatch in .remove_member - test uses PUT but remove_member action expects DELETE",
+          "severity": "Medium",
+          "matched_candidate": "Test for 'cannot remove members from automatic groups' uses xhr :put but route is defined as DELETE, not validating actual HTTP method used by clients",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same underlying issue: there is an HTTP method mismatch in the test for remove_member functionality. The golden comment states the test uses PUT but remove_member expects DELETE. The candidate issue describes the identical problem - the test uses 'xhr :put' but the route is defined as DELETE. They are pointing to the exact same bug with slightly different wording."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Breaking change in GroupsController#members endpoint response format - changed from flat JSON array to object with members array and meta fields, breaking existing API consumers"
+        },
+        {
+          "candidate": "GroupsController#members endpoint now always paginates with default limit of 50, whereas old code returned all members for non-automatic groups, causing silent truncation for existing callers"
+        },
+        {
+          "candidate": "In admin/groups_controller update action, group.visible = params[:visible] == 'true' executes unconditionally, causing partial updates without visible parameter to reset visibility to false"
+        },
+        {
+          "candidate": "Group model asJSON() sends visible: !!this.get('visible') which evaluates to false for new groups where visible is undefined/null, causing new groups to always be created with visible: false regardless of checkbox state"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": " The findMembers() call is now asynchronous and unhandled. The controller may not have member data immediately available, creating a race condition.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR8__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR8__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3": {
@@ -54569,6 +56028,50 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR3__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR3__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "BlockedEmail.should_block_email? method has side effects during a read operation - it updates statistics even when just checking if an email should be blocked. This could cause race conditions in concurrent environments and makes the method name misleading.",
+          "severity": "Medium",
+          "matched_candidate": "BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation, which can run multiple times per user action, inflating blocked-email statistics",
+          "confidence": 0.95,
+          "reasoning": "Both issues identify the same core problem: the BlockedEmail method performs database writes (updating statistics like match_count and last_match_at) as a side effect during what should be a read-only check operation. The golden comment emphasizes the misleading method name and race condition risks, while the candidate focuses on the validation context and statistics inflation, but both are describing the same underlying issue of inappropriate side effects in a checking/validation method."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "record.save inside should_block? triggers BlockedEmail's own validations including a uniqueness DB query, making validation unnecessarily expensive"
+        },
+        {
+          "candidate": "JavaScript code adds email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures, permanently marking valid emails that belong to other users as rejected"
+        },
+        {
+          "candidate": "rejectedEmails is initialized as Em.A([]) directly in the class definition, making it a shared prototype property across all instances, causing cross-instance state pollution"
+        },
+        {
+          "candidate": "BlockedEmail.should_block? calls record.save but doesn't check the return value, silently ignoring save failures and operating on dirty in-memory state with incremented match_count that wasn't persisted"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Regex pattern @(#{domains}) only matches domain suffixes, not full domains. evil.example.com would match whitelist entry example.com.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR3__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR3__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5": {
@@ -55981,6 +57484,53 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR5__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR5__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "-ms-align-items never existed in any version of IE/Edge; the correct legacy property is -ms-flex-align.",
+          "severity": "Low",
+          "matched_candidate": "`-ms-align-items` (line 121) is not a valid CSS property and will be silently ignored by IE10/11. The correct IE prefix for `align-items` is only `-ms-flex-align`",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: the use of `-ms-align-items` which is not a valid CSS property. Both correctly state that the proper IE/Edge legacy property should be `-ms-flex-align` instead. The candidate provides additional context about which browsers will ignore it (IE10/11) and the line number, but the core issue being identified is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `order` mixin passes raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values (default is 1) while the standard `order` property is 0-based (default is 0). The correct value for `-webkit-box-ordinal-group` should be `$val + 1`, otherwise `@include order(1)` would get `-webkit-box-ordinal-group: 1` which equals the default, breaking intended ordering in old WebKit browsers"
+        },
+        {
+          "candidate": "In `.small-action`, child `.topic-avatar` still has `float: left` despite parent now being `display: flex`. While floats are ignored on flex items, this indicates incomplete migration and should be removed for clarity"
+        },
+        {
+          "candidate": "The padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that created spacing from the avatar. With `1.5%` horizontal padding, the gap between avatar and description text will shrink significantly on narrow viewports and may overlap or appear too tight"
+        },
+        {
+          "candidate": "The `-ms-flex-align` property uses different value keywords than standard `align-items`: it expects `start`/`end` instead of `flex-start`/`flex-end`. While current usages (`center`, `baseline`) happen to match, the mixin does not translate values, so future callers passing `flex-start` or `flex-end` would silently break in IE"
+        },
+        {
+          "candidate": "The `.badge-wrapper.bullet` styling is inconsistent: `margin-top: 5px` was removed in the header's `.extra-info-wrapper` context to accommodate flexbox alignment, but the same selector in `#topic-title .title-wrapper` retains `margin-top: 5px`, creating unintentional inconsistency"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Mixing float: left with flexbox causes layout issues. Further this PR removes the float-based right alignment for .d-header .panel, which may cause the login panel in the non-Ember/noscript header (where .panel is nested inside .row and not a flex item) to stack under the title instead of remaining right-aligned.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR5__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR5__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6": {
@@ -57239,6 +58789,39 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR6__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR6__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The include_website_name method is missing the required ? suffix. Rails serializers expect include_ methods to end with ? for conditional attribute inclusion, a convention followed by other methods in this serializer. Without it, the website_name attribute may not be conditionally included as intended. Additionally, the '.' << website_host string concatenation should be replaced with '.' + website_host or '.#{website_host}' to avoid mutating string literals, which can lead to issues.",
+          "severity": "Medium",
+          "matched_candidate": "Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized \u2014 even when `website` is nil/blank, defeating the intent of the guard.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same primary issue: the `include_website_name` method is missing the required `?` suffix that Rails/ActiveModel::Serializer expects for conditional attribute inclusion methods. Both explain that without the `?`, the method won't function as intended for conditional inclusion. The golden comment additionally mentions a string concatenation issue with `<<`, but the candidate focuses solely on the missing `?` suffix. Since the core bug (missing `?` suffix) is identical in both, this is a match."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "`website_name` is not included in the `untrusted_attributes` list (lines 105-111), but `website` is. Since `website_name` bypasses this check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden."
+        },
+        {
+          "candidate": "The `Discourse.stubs(:current_hostname)` call happens inside the `it` block, after `json` is already defined as a lazy let. If `json` were ever evaluated before the stub (e.g., by a shared example or `before` block), the stub would not be in effect. The stub should be in a `before` block before `json` is accessed."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 1,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR6__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR6__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4": {
@@ -59782,6 +61365,66 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR4__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR4__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The current origin validation using indexOf is insufficient and can be bypassed. An attacker could use a malicious domain like evil-discourseUrl.com to pass this check.",
+          "severity": "Medium",
+          "matched_candidate": "The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a *substring* of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain (e.g., `http://evil.com` if discourseUrl is `http://evil.com.attacker.com/...`) to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same security vulnerability: the indexOf-based origin validation is insufficient and can be bypassed. The golden comment gives an example of 'evil-discourseUrl.com' passing the check, while the candidate provides a similar example with 'evil.com' as a prefix of 'evil.com.attacker.com'. Both recognize that substring matching allows malicious domains to bypass validation. The candidate additionally suggests a fix (checking if discourseUrl starts with e.origin + '/' or comparing parsed origins), but the core issue identified is identical."
+        },
+        {
+          "golden_comment": "The ERB block closes with end if, which is invalid Ruby/ERB and will raise at render; it should just be end to close the if block.",
+          "severity": "Medium",
+          "matched_candidate": "ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression (e.g., `end if condition`). This will raise a SyntaxError when the `best.html.erb` template is rendered.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same syntax error: the invalid use of 'end if' on line 6 of the ERB template. The golden comment states it should be just 'end' to close the if block, and the candidate provides the same correction (should be '<%- end %>') with additional technical detail about why it's invalid. They both recognize this will cause a runtime error when rendering."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. The `<%= %>` ERB tag HTML-escapes the output, but HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped."
+        },
+        {
+          "candidate": "In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`. SimpleRSS returns nil for missing fields."
+        },
+        {
+          "candidate": "`contents << \"\\n<hr>...\"` on line 13 uses the `<<` operator which mutates the string in place. This modifies the caller's original string. In `import_remote`, `doc.content` gets mutated. If `import` is called twice with the same string (e.g., in tests via `let(:contents)`), the imported-from footer is appended multiple times. More critically, when `import` is called for an update path (content_sha1 check on line 34), the SHA1 was computed *after* mutation, so re-importing the same content will always differ from the stored SHA1 (since the stored one includes one footer but the new computation includes two)."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "SSRF vulnerability using open(url) without validation",
+          "severity": "Critical"
+        },
+        {
+          "golden_comment": "postMessage targetOrigin should be the origin (scheme+host+port), not the full referrer URL; using the full URL will cause the message to be dropped and prevent resizing.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "The code sets X-Frame-Options: ALLOWALL which completely disables clickjacking protection. The referer validation can be bypassed (referer headers are easily spoofed), and the fallback to empty string for nil referer masks validation failures.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "The TopicEmbed.import method is susceptible to a NoMethodError if the contents parameter is nil when attempting to append a string, and an XSS vulnerability due to unescaped url interpolation in the generated HTML.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 6,
+      "tp": 2,
+      "fp": 3,
+      "fn": 4,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR4__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR4__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1": {
@@ -61503,6 +63146,51 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The downsize method is defined twice. The second definition, which expects a single dimensions string parameter, overrides the first, which expected separate max_width and max_height parameters. This makes the original method unreachable and breaks existing code that calls it with separate width and height arguments.",
+          "severity": "Medium",
+          "matched_candidate": "Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: the downsize method is defined twice, with the second definition overriding the first, making the original method unreachable. The candidate provides additional detail about the specific line numbers (145 and 149), the exact method signatures, and a concrete example of where this breaks existing code (resize_emoji.rb:14), but the core issue is identical - duplicate method definitions where the second replaces the first, breaking code that calls the original signature."
+        },
+        {
+          "golden_comment": "Hardcoding maxSizeKB = 10 * 1024 ignores Discourse.SiteSettings['max_' + type + '_size_kb'], so the client-side limit can diverge from server-side and per-type settings (also applies to the 413 handler below).",
+          "severity": "Low",
+          "matched_candidate": "The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the same core issue: the hardcoded `maxSizeKB = 10 * 1024` value replaces the dynamic lookup of `Discourse.SiteSettings['max_' + type + '_size_kb']`, causing client-side validation to ignore admin-configurable per-type size limits and potentially diverge from server-side settings. The candidate provides more detail about the consequences (e.g., non-image attachments, server rejection), but the underlying bug is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The 413 error handler hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message."
+        },
+        {
+          "candidate": "The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Passing 80% as the dimensions can fail for animated GIFs when allow_animated_thumbnails is true, since the animated path uses gifsicle --resize-fit which expects WxH geometry, not a percentage; downsizing would then silently fail.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR1__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR1__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2": {
@@ -63007,6 +64695,49 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR2__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR2__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "logic: Potential nil pointer exception - if no TopicUser record exists, tu will be nil and calling methods on it will crash",
+          "severity": "High",
+          "matched_candidate": "NullPointerError crash: `TopicUser.find_by` returns `nil` when no TopicUser record exists (e.g., user received a notification but never visited the topic). The very next line calls `tu.notification_level` on the nil result, which raises `NoMethodError: undefined method 'notification_level' for nil:NilClass`.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same issue: TopicUser.find_by can return nil, and the subsequent line attempts to call a method (notification_level) on that nil value, which will cause a crash (nil pointer/NoMethodError). The candidate provides more implementation detail about the specific error type and scenario, but the core bug is identical."
+        },
+        {
+          "golden_comment": "Typo in property name: 'stopNotificiationsText' should be 'stopNotificationsText' (missing 'n' in 'Notifications')",
+          "severity": "Low",
+          "matched_candidate": "The unsubscribe controller uses `stopNotificiationsText` (note the typo: 'Notificiations' instead of 'Notifications') as the computed property name. The template references `{{{stopNotificiationsText}}}` with the same typo so it works, but the triple-stache `{{{ }}}` renders raw HTML. The computed property interpolates `model.fancyTitle` (which can contain user-supplied HTML entities) into the i18n string. If `fancyTitle` contains malicious HTML, it would be rendered unescaped.",
+          "confidence": 0.95,
+          "reasoning": "The golden comment identifies a typo in the property name 'stopNotificiationsText' (missing 'n' in 'Notifications'). The candidate issue also identifies this exact same typo ('Notificiations' instead of 'Notifications'), though it goes on to discuss an additional security concern about unescaped HTML rendering. Since the candidate clearly identifies the same typo that the golden comment is looking for, this is a match - the candidate just provides additional context and identifies a second issue as well."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `unsubscribe_link` i18n string now requires an `%{unsubscribe_url}` interpolation variable, but existing email paths that use `add_unsubscribe_link: true` without passing `unsubscribe_url` (e.g., the `mailing_list_notify` path which calls `send_notification_email` without an `unsubscribe_url` key) will raise an `I18n::MissingInterpolationArgument` error or render a broken `[click here]()` link with an empty URL."
+        },
+        {
+          "candidate": "XSS vulnerability: the `title` value is injected directly into HTML via string concatenation (`buffer.push(\"<h4 class='title'>\" + title + \"</h4>\")`). If `title` contains user-controlled content (e.g., a topic title with `<script>` tags), it will be rendered as raw HTML. The old code had the same issue but always rendered; the new code adds a nil guard but still doesn't escape the value."
+        },
+        {
+          "candidate": "The `html_part` method only substitutes `%{respond_instructions}` and `%{unsubscribe_link}` placeholders when `add_unsubscribe_link` is true. But the notification.html.erb template always contains `%{respond_instructions}` (both inside the `context_posts.present?` block and at the bottom). When `add_unsubscribe_link` is false, these literal `%{respond_instructions}` and `%{unsubscribe_link}` strings will appear in the rendered email."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR2__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR2__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/22532": {
@@ -64607,6 +66338,53 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR22532__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR22532__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The updateManyByCredentialId call uses an empty data object, which prevents Prisma's @updatedAt decorator from updating the updatedAt timestamp. This results in inaccurate cache status tracking, as the timestamp isn't updated when the cache is refreshed. To fix this, explicitly set the updatedAt field.",
+          "severity": "Medium",
+          "matched_candidate": "Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive. The comment says 'Update SelectedCalendar.updatedAt' but nothing is actually updated, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the exact same issue: calling updateManyByCredentialId with an empty data object fails to update the updatedAt timestamp because Prisma's @updatedAt decorator doesn't work with updateMany (or in this case, doesn't trigger when the data object is empty). The golden comment states the empty object 'prevents Prisma's @updatedAt decorator from updating' while the candidate explains 'Prisma's updateMany does NOT auto-update @updatedAt fields' and notes the timestamp 'will never change via this code path'. Both recognize this results in inaccurate timestamp tracking and that the field needs to be explicitly set. The core issue, impact, and root cause are identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `deleteCache` handler only checks `userId: user.id` on the credential, but does not account for delegation credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache."
+        },
+        {
+          "candidate": "The `deleteCache` handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials."
+        },
+        {
+          "candidate": "The Prisma schema uses `@default(now()) @updatedAt` together. For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries."
+        },
+        {
+          "candidate": "Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` is semantically incorrect. The `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. `?? null` would be more correct."
+        },
+        {
+          "candidate": "The `CredentialActionsDropdown` component initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them. When `canDisconnect` is false and `hasCache` is false, the component returns null but still sets up subscription handlers unnecessarily."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "logic: macOS-specific sed syntax with empty string after -i flag will fail on Linux systems",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR22532__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22532__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/8330": {
@@ -66111,6 +67889,46 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR8330__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR8330__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Incorrect end time calculation using slotStartTime instead of slotEndTime",
+          "severity": "Medium",
+          "matched_candidate": "Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same bug: the 'end' variable is incorrectly calculated using 'slotStartTime' instead of 'slotEndTime'. The golden comment states this concisely, while the candidate provides more detail about the consequences (allowing bookings outside working hours), but they are describing the same underlying code defect."
+        },
+        {
+          "golden_comment": "Using === for dayjs object comparison will always return false as it compares object references, not values. Use .isSame() method instead: dayjs(date.start).add(utcOffset, 'minutes').isSame(dayjs(date.end).add(utcOffset, minutes))",
+          "severity": "Medium",
+          "matched_candidate": "Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: using === to compare dayjs objects compares references rather than values, which will always return false. The golden comment suggests using .isSame() method, while the candidate provides more context about the business logic impact (unavailable-day overrides being treated incorrectly). Despite different levels of detail and explanation, they both point to the exact same bug - incorrect comparison of dayjs objects using reference equality instead of value equality."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC."
+        },
+        {
+          "candidate": "The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range (returns true \u2192 function returns false), it prevents checking whether the slot is valid for a different user's override. The combination of mutation (`dateOverrideExist = true`) and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR8330__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8330__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/14943": {
@@ -67463,6 +69281,33 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR14943__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR14943__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "Using retryCount: reminder.retryCount + 1 reads a possibly stale value and can lose increments under concurrency; consider an atomic increment via Prisma (increment: 1) to avoid race conditions (also applies to the similar update in the catch block).",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The deletion logic in scheduleSMSReminders.ts incorrectly deletes non-SMS workflow reminders (e.g., Email, WhatsApp) that have retryCount > 1. This occurs because the retryCount condition within the OR clause for deletion lacks a method: WorkflowMethods.SMS filter, causing it to apply to all reminder types instead of only SMS reminders, which is the intended scope of this function.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 0,
+      "total_golden": 2,
+      "tp": 0,
+      "fp": 0,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR14943__20260411",
+      "pr_url": null
     }
   },
   "https://github.com/calcom/cal.com/pull/22345": {
@@ -68815,6 +70660,50 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR22345__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR22345__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In getBaseConditions(), the else if (filterConditions) and final else branches are unreachable. This is because getAuthorizationConditions() always returns a non-null Prisma.Sql object, making authConditions always truthy, which means only the first two if/else if conditions are ever evaluated.",
+          "severity": "Low",
+          "matched_candidate": "getBaseConditions method has unreachable dead code in 'else if (filterConditions)' and 'else return NOTHING_CONDITION' branches because getAuthorizationConditions() always returns truthy Prisma.Sql object. Also prevents short-circuiting when auth fails.",
+          "confidence": 0.95,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same problem: the else if (filterConditions) and final else branches in getBaseConditions() are unreachable because getAuthorizationConditions() always returns a non-null/truthy Prisma.Sql object, causing only the first two conditions to be evaluated. The candidate adds an additional observation about preventing short-circuiting when auth fails, but the core issue about unreachable code is identical."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Public methods getAuthorizationConditions() and getFilterConditions() return Prisma.Sql fragments containing OR operators without outer parentheses, causing SQL operator precedence bugs when composed with AND operators. This creates a privilege escalation vulnerability where org team bookings could bypass date filters."
+        },
+        {
+          "candidate": "Constructor parameter type changed from discriminated union InsightsBookingServiceOptions to flat type InsightsBookingServicePublicOptions, removing compile-time type safety for required teamId parameter when scope='team'. Missing teamId now silently fails at runtime instead of causing compile error."
+        },
+        {
+          "candidate": "Falsy guard 'if (this.filters.eventTypeId)' skips filter when eventTypeId is 0, and 'if (this.filters.memberUserId)' skips when memberUserId is 0, causing unfiltered results instead of filtering by ID=0."
+        },
+        {
+          "candidate": "Entire 'Caching' test suite removed, leaving caching logic in service (lines 84-87 and 91-94) untested. No tests to catch regressions if caching is accidentally broken."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Fetching userIdsFromOrg only when teamsFromOrg.length > 0 can exclude org-level members for orgs without child teams; consider deriving from teamIds (which includes orgId) or removing the guard so org-only orgs still include member user bookings.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR22345__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22345__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/11059": {
@@ -71296,6 +73185,65 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR11059__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR11059__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Invalid Zod schema syntax. Computed property keys like [z.string().toString()] are not valid in Zod object schemas and will cause runtime errors. ",
+          "severity": "High",
+          "matched_candidate": "minimumTokenResponseSchema uses [z.string().toString()] as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field (like expires_in or expiry_date) exists in the response, making the 'minimum token validation' ineffective.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: using computed property keys like [z.string().toString()] in a Zod schema is invalid/incorrect. The golden comment states this causes runtime errors, while the candidate provides more detail about why it's wrong (evaluates to a literal string like 'ZodString' at definition time rather than creating a dynamic validator). Both are describing the same fundamental bug - misuse of computed property syntax in Zod schemas."
+        },
+        {
+          "golden_comment": "parseRefreshTokenResponse returns a Zod safeParse result ({ success, data, error }), not the credential key object. Persisting that as key stores the wrapper instead of the token payload; we should store the parsed data or use schema parse.",
+          "severity": "High",
+          "matched_candidate": "parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously googleCredentialSchema.parse() returned just the data. Now the DB will store {success: true, data: {access_token, refresh_token, expiry_date}} instead of {access_token, refresh_token, expiry_date}, corrupting the credential for all subsequent reads.",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the exact same issue: parseRefreshTokenResponse returns a Zod safeParse wrapper object ({success, data, error}) instead of the actual credential data, and this wrapper is being stored directly in the database instead of extracting the parsed data. The golden comment says 'stores the wrapper instead of the token payload' and suggests storing 'the parsed data'. The candidate issue provides more detail about the structure but describes the identical problem - the DB will store the wrapper object instead of just the credential data, corrupting subsequent reads."
+        },
+        {
+          "golden_comment": "When APP_CREDENTIAL_SHARING_ENABLED and CALCOM_CREDENTIAL_SYNC_ENDPOINT are set, the refreshFunction helper returns the fetch Response, but several callers (for example GoogleCalendarService.refreshAccessToken expecting res.data, and HubspotCalendarService.refreshAccessToken expecting a HubspotToken) assume it returns the integration-specific token object. That mismatch will cause runtime errors in the sync-enabled path unless the return type or those call sites are adjusted.",
+          "severity": "High",
+          "matched_candidate": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses res?.data (undefined on fetch Response \u2192 TypeError on token.access_token), Hubspot accesses .expiryDate/.expiresIn (undefined on Response), Zoho-bigin and Zohocrm access .data.error (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: when APP_CREDENTIAL_SHARING_ENABLED is true, the refreshFunction/refreshOAuthTokens returns a fetch Response object, but callers (GoogleCalendarService, HubspotCalendarService, and the candidate adds Zoho-bigin/Zohocrm) expect integration-specific token objects with properties like .data, .expiryDate, .expiresIn. This type mismatch causes runtime errors. The candidate provides more specific examples of the crash scenarios, but the underlying bug is identical."
+        },
+        {
+          "golden_comment": "When the sync endpoint path is used, res is a fetch Response and has no .data; res?.data will be undefined and token.access_token will throw at runtime. This relies on a consistent return shape from refreshOAuthTokens, which isn\u2019t guaranteed currently.",
+          "severity": "High",
+          "matched_candidate": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses res?.data (undefined on fetch Response \u2192 TypeError on token.access_token), Hubspot accesses .expiryDate/.expiresIn (undefined on Response), Zoho-bigin and Zohocrm access .data.error (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: when the sync endpoint path is used (which corresponds to APP_CREDENTIAL_SHARING_ENABLED being true), refreshOAuthTokens returns a fetch Response object instead of the expected shape. This causes res?.data to be undefined, leading to runtime errors when accessing token.access_token. The candidate provides more comprehensive detail about the impact across multiple integrations (Google Calendar, Hubspot, Zoho), but the fundamental bug is identical - inconsistent return types from refreshOAuthTokens causing crashes when callers expect a .data property."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Salesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using credentialKey.access_token \u2014 the OLD, pre-refresh token. The newly refreshed token from accessTokenParsed.data is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token."
+        },
+        {
+          "candidate": "The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The parseRefreshTokenResponse function incorrectly sets refresh_token to the hardcoded string 'refresh_token' when it's missing from the OAuth refresh token response. This invalidates the token, breaking subsequent token refreshes and causing authentication failures.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 4,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.8,
+      "recall": 0.8,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR11059__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR11059__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/7232": {
@@ -72900,6 +74848,49 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR7232__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR7232__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Asynchronous functions deleteScheduledEmailReminder and deleteScheduledSMSReminder are called without await inside forEach loops. This occurs during booking rescheduling/cancellation, and workflow/workflow step deletion/updates. Consequently, scheduled workflow reminders may not be reliably cancelled, potentially leaving them active.",
+          "severity": "Medium",
+          "matched_candidate": "In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget. The surrounding try/catch will never catch rejections from these promises\u2014they'll become unhandled promise rejections instead of being logged as intended",
+          "confidence": 0.95,
+          "reasoning": "Both issues identify the same core problem: async functions `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` are called without await inside forEach loops. The golden comment focuses on the consequence that scheduled reminders may not be reliably cancelled. The candidate issue provides additional technical detail about the fire-and-forget nature and unhandled promise rejections, but fundamentally describes the same bug - missing await on async deletion functions in forEach loops during booking operations."
+        },
+        {
+          "golden_comment": "When immediateDelete is true, the deleteScheduledEmailReminder function cancels the SendGrid email but fails to delete the corresponding WorkflowReminder record from the database. This creates orphaned database entries and is inconsistent with the immediateDelete: false path, which marks the record as cancelled. The SendGrid DELETE API call is also omitted in this path.",
+          "severity": "High",
+          "matched_candidate": "When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record. The record becomes permanently orphaned\u2014it's never cleaned up by the CRON's cancel loop (since `cancelled` is not set to true) and won't be found by the unscheduled query either (since `scheduled` is true)",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same core issue: when immediateDelete is true, the function cancels the SendGrid email but fails to delete or properly mark the WorkflowReminder database record, creating orphaned entries. The candidate provides additional detail about why the record becomes orphaned (won't be caught by CRON cancel loop or unscheduled query), but this is elaborating on the same fundamental bug that the golden comment describes."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The CRON job first deletes ALL email reminders with a past `scheduledDate` (line 34-41), then queries for `cancelled: true` reminders to cancel via SendGrid (line 44-51). If a cancelled reminder's `scheduledDate` has already passed, the first `deleteMany` removes the DB record before the cancel loop processes it. The email is never cancelled in SendGrid and gets delivered despite the booking being cancelled"
+        },
+        {
+          "candidate": "The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails (e.g., network error, invalid batch_id), the exception aborts the loop and all remaining reminders are never cancelled or deleted. Those emails will be sent despite their bookings being cancelled"
+        },
+        {
+          "candidate": "The sender name/sender ID input section (lines 463-492) was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}`. After the refactor, this section is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR7232__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR7232__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/14740": {
@@ -75045,6 +77036,62 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR14740__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR14740__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Case sensitivity bypass in email blacklist",
+          "severity": "High",
+          "matched_candidate": "Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.",
+          "confidence": 1.0,
+          "reasoning": "Both the golden comment and candidate issue identify the exact same vulnerability: the email blacklist can be bypassed using different letter casing. The golden comment states 'Case sensitivity bypass in email blacklist' while the candidate provides detailed explanation that blacklisted emails are lowercased but incoming guest emails are compared without case normalization, allowing variants like 'Blocked@Example.COM' to bypass 'blocked@example.com'. This is the same underlying security issue with different levels of detail."
+        },
+        {
+          "golden_comment": "The logic for checking team admin/owner permissions is incorrect. This condition uses AND (&&) which requires both isTeamAdmin AND isTeamOwner to be true, but it should use OR (||) since a user needs to be either an admin OR an owner to have permission.",
+          "severity": "Critical",
+          "matched_candidate": "Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.",
+          "confidence": 1.0,
+          "reasoning": "Both comments identify the exact same bug: the use of AND (&&) instead of OR (||) in the authorization check for team admin/owner permissions. The golden comment states that the condition requires both isTeamAdmin AND isTeamOwner to be true when it should require either one. The candidate issue makes the same point with additional detail, explaining that the variable name suggests OR logic but the implementation uses AND, and that this makes the admin check ineffective since only owners will pass. Both are describing the same logical error in the permission check."
+        },
+        {
+          "golden_comment": "This calls the email sender with the original guests, so existing attendees included in the input will be treated as new when sending notifications, leading to incorrect emails.",
+          "severity": "Medium",
+          "matched_candidate": "The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. In `sendAddGuestsEmails`, `newGuests.includes(attendee.email)` determines which email template to use: new guests get `AttendeeScheduledEmail` (booking confirmation), existing attendees get `AttendeeAddGuestsEmail` (notification). If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: the code passes the original/unfiltered guests array to the email sender function, which causes existing attendees to be incorrectly treated as new guests when determining which email template to send. The golden comment states this leads to 'incorrect emails' while the candidate provides more specific detail about the exact consequence (existing attendees receiving booking confirmation instead of 'guests added' notification), but they are describing the same bug."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "When `booking.userId` is null (it's `Int?` in the Prisma schema), `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable: if the user is an attendee (not the organizer) of a booking whose organizer account was deleted, the authorization check passes but this query crashes."
+        },
+        {
+          "candidate": "In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. While in practice attendees should exist, if the attendees array is ever empty (e.g., all attendees removed in a race condition), this would throw a runtime TypeError crashing the email send for the entire batch (since all emails are sent via `Promise.all`)."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "uniqueGuests filters out existing attendees and blacklisted emails but does not deduplicate duplicates within the input; createMany can insert duplicate attendee rows if the client submits repeated emails.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Starting with an array containing an empty string may cause validation issues. Consider starting with an empty array [] and handling the empty state in the MultiEmail component instead.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 3,
+      "fp": 2,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.6,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR14740__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR14740__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/10600": {
@@ -77015,6 +79062,54 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR10600__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR10600__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Password bypass via backup code + dummy totpCode. The password check at line 121 is guarded by `!credentials.totpCode`. An attacker who has obtained a valid backup code can send a request with a dummy non-empty `totpCode` (e.g., \"x\") alongside the stolen `backupCode`. The non-empty `totpCode` causes the password check to be skipped (line 121 evaluates to false). Then at line 131, `credentials.backupCode` is truthy so the backup code path is entered \u2014 which succeeds without ever verifying the password. The TOTP path (line 157) is never reached because it's in the `else if`. This allows login with only email + backup code, no password required."
+        },
+        {
+          "candidate": "Backup code login is blocked for non-CAL identity provider users. The guard at line 113 throws `ThirdPartyIdentityProviderEnabled` when `!credentials.totpCode` is true. When a Google/SAML user with 2FA tries to use a backup code (which means no totpCode is provided), this check blocks the request before the backup code path at line 131 is ever reached. While the UI currently doesn't show the 'Lost access' option for external TOTP flows (ExternalTotpFooter has no such button), this means backup codes are completely non-functional for these users \u2014 defeating the purpose of backup codes as a recovery mechanism."
+        },
+        {
+          "candidate": "Unhandled exception in API route returns unstructured error. When `CALENDSO_ENCRYPTION_KEY` is missing in the backup code path, `throw new Error(ErrorCode.InternalServerError)` causes Next.js to return a generic 500 HTML error page instead of a structured JSON response. The client code (`DisableTwoFactorModal`) calls `response.json()` and checks `body.error`, which would fail on an HTML response, causing a silent client-side error."
+        },
+        {
+          "candidate": "Object URL created for backup codes download is never revoked on component unmount or dialog close. `URL.createObjectURL(textBlob)` at line 99 creates a blob URL that persists until explicitly revoked. While `handleSetup` revokes the previous URL before creating a new one, the `resetState` function (called on dialog close/cancel) does not revoke it, and there is no cleanup effect. Each 2FA setup flow leaks one blob URL."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The exported function TwoFactor handles backup codes and is in BackupCode.tsx. Inconsistent naming.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Error message mentions 'backup code login' but this is a disable endpoint, not login",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Backup code validation is case-sensitive due to the use of indexOf(). This causes validation to fail if a user enters uppercase hex characters, as backup codes should be case-insensitive for a better user experience.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Because backupCodes are decrypted and mutated in memory before being written back, two concurrent login requests using the same backupCode could both pass this check and update, so a single backup code may effectively be accepted more than once if used concurrently, weakening the intended one-time-use semantics.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 4,
+      "tp": 0,
+      "fp": 4,
+      "fn": 4,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR10600__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10600__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/10967": {
@@ -79260,6 +81355,62 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR10967__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR10967__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Potential null reference if mainHostDestinationCalendar is undefined if evt.destinationCalendar is null or an empty array ",
+          "severity": "High",
+          "matched_candidate": "When evt.destinationCalendar is null or empty array, const [mainHostDestinationCalendar] = evt.destinationCalendar ?? [] sets mainHostDestinationCalendar to undefined, then accessing mainHostDestinationCalendar.integration without optional chaining throws TypeError",
+          "confidence": 0.95,
+          "reasoning": "Both comments identify the same underlying issue: when evt.destinationCalendar is null or an empty array, mainHostDestinationCalendar becomes undefined, and accessing its properties (like .integration) without proper null checking will cause a runtime error. The candidate provides more technical detail about the TypeError, while the golden comment is more concise, but they describe the exact same bug."
+        },
+        {
+          "golden_comment": "Logic inversion in organization creation: The slug property is now conditionally set when IS_TEAM_BILLING_ENABLED is true, instead of when it's false as originally intended. This change, combined with requestedSlug still being set when IS_TEAM_BILLING_ENABLED is true, results in both properties being set when billing is enabled, and neither when disabled",
+          "severity": "Medium",
+          "matched_candidate": "The slug/requestedSlug logic is inverted: old code set slug when IS_TEAM_BILLING_ENABLED is false, new code sets slug when IS_TEAM_BILLING_ENABLED is true, causing organizations without billing to get neither slug nor requestedSlug while organizations with billing get both",
+          "confidence": 0.98,
+          "reasoning": "Both comments identify the same logic inversion bug: the condition for setting the slug property has been flipped from IS_TEAM_BILLING_ENABLED being false to true. Both also note the same consequence - that when billing is enabled, both slug and requestedSlug are set, while when billing is disabled, neither is set. The candidate uses slightly different wording ('inverted logic' vs 'logic inversion') but describes the identical problem and its effects."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In updateEvent and deleteEvent, when externalCalendarId is falsy, the else-branch searches event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId) but externalCalendarId is guaranteed falsy at that point, so .find() will never match and always returns undefined instead of falling back to event.destinationCalendar?.externalId"
+        },
+        {
+          "candidate": "In updateAllCalendarEvents, if reference.credentialId is truthy but credential is not found in this.calendarCredentials AND DB lookup fails, credential remains undefined and is passed to updateEvent which will crash when calling getCalendar(credential)"
+        },
+        {
+          "candidate": "In handleNewBooking, when neither eventType.destinationCalendar nor organizerUser.destinationCalendar exist, evt.destinationCalendar is set to null, making evt.destinationCalendar?.push(...teamDestinationCalendars) a no-op that silently drops team member calendars in collective events"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The optional chaining on mainHostDestinationCalendar?.integration is redundant since you already check mainHostDestinationCalendar in the ternary condition.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Logic error: when externalCalendarId is provided, you're searching for a calendar where externalId === externalCalendarId, but this will always fail since you're looking for a calendar that matches itself. Should likely find by credentialId or use different logic.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The Calendar interface now requires createEvent(event, credentialId), but some implementations (e.g., Lark/Office365) still declare createEvent(event) only\u2014this breaks the interface contract (also applies to other locations in the PR).",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 3,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR10967__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10967__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/8087": {
@@ -80857,6 +83008,50 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR8087__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR8087__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The code uses forEach with async callbacks, which causes asynchronous operations (e.g., calendar/video event deletions, payment refunds) to run concurrently without being awaited. This 'fire-and-forget' behavior leads to unhandled promise rejections, race conditions, and incomplete cleanup, as surrounding try-catch blocks cannot properly handle errors from these unawaited promises. Replace forEach with for...of loops or Promise.all() with map() to ensure proper sequential execution and error handling.",
+          "severity": "Critical",
+          "matched_candidate": "async callback inside forEach is fire-and-forget in packages/app-store/vital/lib/reschedule.ts line 125. The forEach does not await the async callback, so: (1) the try/catch block on line 124 will never catch errors from the async operations, (2) the deleteEvent/deleteMeeting calls are not awaited before the function proceeds to send reschedule emails on line 143, meaning calendar/video deletions may silently fail or race with subsequent operations",
+          "confidence": 0.95,
+          "reasoning": "Both the golden comment and candidate issue identify the same core problem: using forEach with async callbacks creates a fire-and-forget pattern where asynchronous operations are not properly awaited. Both mention: (1) try-catch blocks cannot catch errors from unawaited promises, (2) operations like calendar/video event deletions run without proper awaiting, and (3) this leads to race conditions and incomplete cleanup. The candidate is more specific to one file location while the golden is more general, but they describe the identical underlying bug."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "async callback inside forEach is fire-and-forget in packages/app-store/wipemycalother/lib/reschedule.ts line 125. The async callback's promises are discarded, so calendar deleteEvent calls are unawaited. Errors won't be caught and the function proceeds immediately to send reschedule emails"
+        },
+        {
+          "candidate": "async callback inside forEach is fire-and-forget in packages/trpc/server/routers/viewer/bookings.tsx line 553. The forEach uses an async callback but forEach ignores the returned promises. The deleteEvent/deleteMeeting operations are unawaited, and the code immediately proceeds to send reschedule emails on line 570. Any errors from the async operations are silently swallowed"
+        },
+        {
+          "candidate": "async callback inside forEach is fire-and-forget in packages/features/bookings/lib/handleCancelBooking.ts line 460. The forEach uses an async callback, but forEach discards the promises. The `apiDeletes` array is populated inside the async callback, but since the callback is not awaited, `apiDeletes` will be empty when `Promise.all(apiDeletes)` is called later, causing calendar deletions for recurring events to be silently skipped"
+        },
+        {
+          "candidate": "getCalendarCredentials in packages/core/CalendarManager.ts line 28 is a synchronous function that calls getCalendar (now async) inside flatMap. Since getCalendar now returns Promise<Calendar | null>, the `calendar` field in each returned object is a Promise, not a Calendar. The return type claims `calendar` is `Calendar | null` when it's actually `Promise<Calendar | null>`. Any caller that doesn't await `item.calendar` will get a truthy Promise object instead of null, bypassing null checks"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Consider adding try-catch around the await to handle import failures gracefully",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR8087__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8087__20260411/pull/1"
     }
   }
 }
\ No newline at end of file
diff --git a/offline/results/benchmark_data.json b/offline/results/benchmark_data.json
index 8f19cf0..de32429 100644
--- a/offline/results/benchmark_data.json
+++ b/offline/results/benchmark_data.json
@@ -1173,6 +1173,37 @@
             "created_at": "2026-03-18T20:42:38Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR37429__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37429__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "themes/src/main/resources-community/theme/base/account/messages/messages_lt.properties",
+            "line": 101,
+            "body": "P2: Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian. This affects both the account and login message files, meaning Lithuanian users will see Italian text for the TOTP setup step.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>themes/src/main/resources-community/theme/base/account/messages/messages_lt.properties</file_path>\n<line_numbers>101</line_numbers>\n\n<comment>Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian. This affects both the account and login message files, meaning Lithuanian users will see Italian text for the TOTP setup step.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>themes/src/main/resources-community/theme/base/account/messages/messages_lt.properties</file_path>\n<line_numbers>101</line_numbers>\n\n<comment>Lithuanian (lt) translation replaced with Italian text. The value 'Installa una delle seguenti applicazioni sul tuo cellulare:' is Italian, not Lithuanian. This affects both the account and login message files, meaning Lithuanian users will see Italian text for the TOTP setup step.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:21:56Z"
+          },
+          {
+            "path": "themes/src/main/resources-community/theme/base/account/messages/messages_zh_CN.properties",
+            "line": 112,
+            "body": "P2: Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value '\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a' uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f). The rest of the file uses Simplified Chinese, making this inconsistent and incorrect for zh_CN locale users.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>themes/src/main/resources-community/theme/base/account/messages/messages_zh_CN.properties</file_path>\n<line_numbers>112</line_numbers>\n\n<comment>Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value '\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a' uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f). The rest of the file uses Simplified Chinese, making this inconsistent and incorrect for zh_CN locale users.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>themes/src/main/resources-community/theme/base/account/messages/messages_zh_CN.properties</file_path>\n<line_numbers>112</line_numbers>\n\n<comment>Chinese Simplified (zh_CN) account translation replaced with Traditional Chinese characters. The new value '\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a' uses Traditional Chinese (\u624b\u6a5f, \u5b89\u88dd, \u61c9\u7528\u7a0b\u5f0f) instead of Simplified Chinese (\u624b\u673a, \u5b89\u88c5, \u5e94\u7528\u7a0b\u5e8f). The rest of the file uses Simplified Chinese, making this inconsistent and incorrect for zh_CN locale users.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:21:56Z"
+          },
+          {
+            "path": "misc/theme-verifier/src/main/java/org/keycloak/themeverifier/VerifyMessageProperties.java",
+            "line": 91,
+            "body": "P3: containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. The regex `<[a-z]+[^>]*>` matches `<a href=\"...\">`, so if the English message contains only anchor tags (no other HTML), containsHtml returns true and POLICY_SOME_HTML is selected. This allows `<br>`, `<p>`, `<strong>`, `<b>` tags in translations even when the English source only has anchor tags, defeating the stated intent in the comment on line 90: 'Only if the English source string contains HTML we also allow HTML in the translation'.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>misc/theme-verifier/src/main/java/org/keycloak/themeverifier/VerifyMessageProperties.java</file_path>\n<line_numbers>91</line_numbers>\n\n<comment>containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. The regex `<[a-z]+[^>]*>` matches `<a href=\"...\">`, so if the English message contains only anchor tags (no other HTML), containsHtml returns true and POLICY_SOME_HTML is selected. This allows `<br>`, `<p>`, `<strong>`, `<b>` tags in translations even when the English source only has anchor tags, defeating the stated intent in the comment on line 90: 'Only if the English source string contains HTML we also allow HTML in the translation'.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>misc/theme-verifier/src/main/java/org/keycloak/themeverifier/VerifyMessageProperties.java</file_path>\n<line_numbers>91</line_numbers>\n\n<comment>containsHtml() checks englishValue BEFORE anchor tags are stripped from it, but AFTER santizeAnchors() already stripped anchors from the translation value. The regex `<[a-z]+[^>]*>` matches `<a href=\"...\">`, so if the English message contains only anchor tags (no other HTML), containsHtml returns true and POLICY_SOME_HTML is selected. This allows `<br>`, `<p>`, `<strong>`, `<b>` tags in translations even when the English source only has anchor tags, defeating the stated intent in the comment on line 90: 'Only if the English source string contains HTML we also allow HTML in the translation'.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:21:56Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:21:56Z"
+          }
+        ]
       }
     ]
   },
@@ -2170,6 +2201,31 @@
             "created_at": "2026-03-18T20:26:50Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR37634__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37634__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "services/src/main/java/org/keycloak/protocol/oidc/encode/AccessTokenContext.java",
+            "line": 73,
+            "body": "P2: Copy-paste bug: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`. This means `rawTokenId` is never null-checked, and a null `rawTokenId` will be silently accepted, potentially causing NPEs later when the token ID is used (e.g., in `encodeTokenId()` which concatenates it).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/protocol/oidc/encode/AccessTokenContext.java</file_path>\n<line_numbers>73</line_numbers>\n\n<comment>Copy-paste bug: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`. This means `rawTokenId` is never null-checked, and a null `rawTokenId` will be silently accepted, potentially causing NPEs later when the token ID is used (e.g., in `encodeTokenId()` which concatenates it).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/protocol/oidc/encode/AccessTokenContext.java</file_path>\n<line_numbers>73</line_numbers>\n\n<comment>Copy-paste bug: `Objects.requireNonNull(grantType, \"Null rawTokenId not allowed\")` checks `grantType` instead of `rawTokenId`. This means `rawTokenId` is never null-checked, and a null `rawTokenId` will be silently accepted, potentially causing NPEs later when the token ID is used (e.g., in `encodeTokenId()` which concatenates it).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:14:35Z"
+          },
+          {
+            "path": "testsuite/integration-arquillian/tests/base/src/test/java/org/keycloak/testsuite/AssertEvents.java",
+            "line": 483,
+            "body": "P2: Two bugs in `isAccessTokenId` matcher: (1) Wrong substring indices: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType). Should be `substring(4, 6)`. (2) Inverted boolean logic: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match. The condition should be negated: `if (!... .equals(...)) return false`. Combined, these two bugs cause the matcher to accept tokens with wrong grant types and reject tokens with correct ones. However, due to the off-by-one in the substring, the matcher will accidentally pass for some inputs. The comment says \"starts at character 4th char\" but the code uses index 3.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>testsuite/integration-arquillian/tests/base/src/test/java/org/keycloak/testsuite/AssertEvents.java</file_path>\n<line_numbers>483</line_numbers>\n\n<comment>Two bugs in `isAccessTokenId` matcher: (1) Wrong substring indices: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType). Should be `substring(4, 6)`. (2) Inverted boolean logic: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match. The condition should be negated: `if (!... .equals(...)) return false`. Combined, these two bugs cause the matcher to accept tokens with wrong grant types and reject tokens with correct ones. However, due to the off-by-one in the substring, the matcher will accidentally pass for some inputs. The comment says \"starts at character 4th char\" but the code uses index 3.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>testsuite/integration-arquillian/tests/base/src/test/java/org/keycloak/testsuite/AssertEvents.java</file_path>\n<line_numbers>483</line_numbers>\n\n<comment>Two bugs in `isAccessTokenId` matcher: (1) Wrong substring indices: `items[0].substring(3, 5)` extracts characters at index 3-4, but the grant type shortcut is at index 4-5 in the 6-char prefix (format: 2-char sessionType + 2-char tokenType + 2-char grantType). Should be `substring(4, 6)`. (2) Inverted boolean logic: `if (items[0].substring(3, 5).equals(expectedGrantShortcut)) return false` rejects tokens when the grant shortcut MATCHES and accepts tokens when it DOESN'T match. The condition should be negated: `if (!... .equals(...)) return false`. Combined, these two bugs cause the matcher to accept tokens with wrong grant types and reject tokens with correct ones. However, due to the off-by-one in the substring, the matcher will accidentally pass for some inputs. The comment says \"starts at character 4th char\" but the code uses index 3.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:14:35Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 2 issues",
+            "created_at": "2026-04-12T02:14:35Z"
+          }
+        ]
       }
     ]
   },
@@ -3188,6 +3244,43 @@
             "created_at": "2026-03-18T20:26:44Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR38446__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR38446__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java",
+            "line": 237,
+            "body": "P3: getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID. The stored credential has an ID (set in updateCredential at line 195), but createFromValues() does not set an ID on the model, and the stored ID is never propagated. Downstream, RecoveryAuthnCodesFormAuthenticator line 86-87 calls removeStoredCredentialById(recoveryCodeCredentialModel.getId()) \u2014 passing null here will fail or cause unexpected behavior when all codes are exhausted. Fix: after createFromValues(), set model.setId(myUser.recoveryCodes.getId()).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java</file_path>\n<line_numbers>237</line_numbers>\n\n<comment>getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID. The stored credential has an ID (set in updateCredential at line 195), but createFromValues() does not set an ID on the model, and the stored ID is never propagated. Downstream, RecoveryAuthnCodesFormAuthenticator line 86-87 calls removeStoredCredentialById(recoveryCodeCredentialModel.getId()) \u2014 passing null here will fail or cause unexpected behavior when all codes are exhausted. Fix: after createFromValues(), set model.setId(myUser.recoveryCodes.getId()).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java</file_path>\n<line_numbers>237</line_numbers>\n\n<comment>getCredentials() returns a RecoveryAuthnCodesCredentialModel with a null ID. The stored credential has an ID (set in updateCredential at line 195), but createFromValues() does not set an ID on the model, and the stored ID is never propagated. Downstream, RecoveryAuthnCodesFormAuthenticator line 86-87 calls removeStoredCredentialById(recoveryCodeCredentialModel.getId()) \u2014 passing null here will fail or cause unexpected behavior when all codes are exhausted. Fix: after createFromValues(), set model.setId(myUser.recoveryCodes.getId()).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:18Z"
+          },
+          {
+            "path": "testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java",
+            "line": 340,
+            "body": "P3: The test provider's isValid() for recovery codes accepts ANY code via anyMatch(), whereas the production RecoveryAuthnCodesCredentialProvider.isValid() only validates against the NEXT sequential code (getNextRecoveryAuthnCode). Additionally, isValid() never consumes/removes the used code \u2014 the production provider calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation. This means: (1) the same code can be reused indefinitely, and (2) out-of-order codes are accepted. Tests using this provider won't catch regressions in sequential ordering or one-time-use behavior.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java</file_path>\n<line_numbers>340</line_numbers>\n\n<comment>The test provider's isValid() for recovery codes accepts ANY code via anyMatch(), whereas the production RecoveryAuthnCodesCredentialProvider.isValid() only validates against the NEXT sequential code (getNextRecoveryAuthnCode). Additionally, isValid() never consumes/removes the used code \u2014 the production provider calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation. This means: (1) the same code can be reused indefinitely, and (2) out-of-order codes are accepted. Tests using this provider won't catch regressions in sequential ordering or one-time-use behavior.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java</file_path>\n<line_numbers>340</line_numbers>\n\n<comment>The test provider's isValid() for recovery codes accepts ANY code via anyMatch(), whereas the production RecoveryAuthnCodesCredentialProvider.isValid() only validates against the NEXT sequential code (getNextRecoveryAuthnCode). Additionally, isValid() never consumes/removes the used code \u2014 the production provider calls removeRecoveryAuthnCode() and updateStoredCredential() after successful validation. This means: (1) the same code can be reused indefinitely, and (2) out-of-order codes are accepted. Tests using this provider won't catch regressions in sequential ordering or one-time-use behavior.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:18Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/authentication/authenticators/browser/RecoveryAuthnCodesFormAuthenticator.java",
+            "line": 83,
+            "body": "P3: _Note: Issue is at line 86, shown here on nearest changed line._\n\nFor federated users, when all recovery codes are exhausted, the authenticator calls removeStoredCredentialById() which delegates to local storage (UserCredentialStore), not federated storage. The credential lives in the federated user storage provider, so this removal silently fails \u2014 the exhausted credential persists in federated storage. While the required action to generate new codes IS correctly added (line 90-91), the stale credential remains and getCredential() will continue to find it on next login, leading to allCodesUsed() being true and the user being stuck in a loop of regenerating codes without the old ones being cleaned up.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/RecoveryAuthnCodesFormAuthenticator.java</file_path>\n<line_numbers>83</line_numbers>\n\n<comment>_Note: Issue is at line 86, shown here on nearest changed line._\n\nFor federated users, when all recovery codes are exhausted, the authenticator calls removeStoredCredentialById() which delegates to local storage (UserCredentialStore), not federated storage. The credential lives in the federated user storage provider, so this removal silently fails \u2014 the exhausted credential persists in federated storage. While the required action to generate new codes IS correctly added (line 90-91), the stale credential remains and getCredential() will continue to find it on next login, leading to allCodesUsed() being true and the user being stuck in a loop of regenerating codes without the old ones being cleaned up.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/RecoveryAuthnCodesFormAuthenticator.java</file_path>\n<line_numbers>83</line_numbers>\n\n<comment>_Note: Issue is at line 86, shown here on nearest changed line._\n\nFor federated users, when all recovery codes are exhausted, the authenticator calls removeStoredCredentialById() which delegates to local storage (UserCredentialStore), not federated storage. The credential lives in the federated user storage provider, so this removal silently fails \u2014 the exhausted credential persists in federated storage. While the required action to generate new codes IS correctly added (line 90-91), the stale credential remains and getCredential() will continue to find it on next login, leading to allCodesUsed() being true and the user being stuck in a loop of regenerating codes without the old ones being cleaned up.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:18Z"
+          },
+          {
+            "path": "testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java",
+            "line": 197,
+            "body": "P4: The updateCredential() method stores the raw recovery codes JSON in the credentialData field (line 197: setCredentialData(input.getChallengeResponse())). However, in the standard RecoveryAuthnCodesCredentialModel, credentialData holds metadata (algorithm, remaining count, total count) and secretData holds the hashed codes. The getCredentials() method at line 238 reads credentialData as a List of raw codes and passes them to createFromValues() which re-hashes them \u2014 this works because createFromValues expects raw codes. But isValid() at line 334 also reads credentialData as a raw List and directly compares against user input. If the user storage format were ever changed to match the standard model format (credentialData = metadata, secretData = hashed codes), isValid() would silently break. The inconsistent use of credentialData for two different purposes is fragile.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java</file_path>\n<line_numbers>197</line_numbers>\n\n<comment>The updateCredential() method stores the raw recovery codes JSON in the credentialData field (line 197: setCredentialData(input.getChallengeResponse())). However, in the standard RecoveryAuthnCodesCredentialModel, credentialData holds metadata (algorithm, remaining count, total count) and secretData holds the hashed codes. The getCredentials() method at line 238 reads credentialData as a List of raw codes and passes them to createFromValues() which re-hashes them \u2014 this works because createFromValues expects raw codes. But isValid() at line 334 also reads credentialData as a raw List and directly compares against user input. If the user storage format were ever changed to match the standard model format (credentialData = metadata, secretData = hashed codes), isValid() would silently break. The inconsistent use of credentialData for two different purposes is fragile.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>testsuite/integration-arquillian/servers/auth-server/services/testsuite-providers/src/main/java/org/keycloak/testsuite/federation/BackwardsCompatibilityUserStorage.java</file_path>\n<line_numbers>197</line_numbers>\n\n<comment>The updateCredential() method stores the raw recovery codes JSON in the credentialData field (line 197: setCredentialData(input.getChallengeResponse())). However, in the standard RecoveryAuthnCodesCredentialModel, credentialData holds metadata (algorithm, remaining count, total count) and secretData holds the hashed codes. The getCredentials() method at line 238 reads credentialData as a List of raw codes and passes them to createFromValues() which re-hashes them \u2014 this works because createFromValues expects raw codes. But isValid() at line 334 also reads credentialData as a raw List and directly compares against user input. If the user storage format were ever changed to match the standard model format (credentialData = metadata, secretData = hashed codes), isValid() would silently break. The inconsistent use of credentialData for two different purposes is fragile.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:18Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:18:18Z"
+          }
+        ]
       }
     ]
   },
@@ -4028,6 +4121,37 @@
             "created_at": "2026-03-18T20:25:38Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR36882__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36882__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "docs/guides/operator/advanced-configuration.adoc",
+            "line": 448,
+            "body": "P3: Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic (UpgradeLogicFactory, RecreateOnImageChangeUpgradeLogic, AlwaysRecreateUpgradeLogic) makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons \u2014 it never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>docs/guides/operator/advanced-configuration.adoc</file_path>\n<line_numbers>448</line_numbers>\n\n<comment>Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic (UpgradeLogicFactory, RecreateOnImageChangeUpgradeLogic, AlwaysRecreateUpgradeLogic) makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons \u2014 it never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>docs/guides/operator/advanced-configuration.adoc</file_path>\n<line_numbers>448</line_numbers>\n\n<comment>Documentation claims the Operator will fail without the `rolling-updates` feature enabled, but the operator never checks this feature flag. The operator's upgrade logic (UpgradeLogicFactory, RecreateOnImageChangeUpgradeLogic, AlwaysRecreateUpgradeLogic) makes rolling vs recreate decisions purely based on Kubernetes StatefulSet image/args/env comparisons \u2014 it never references `Profile.Feature.ROLLING_UPDATES` or invokes `kc.sh update-compatibility`. This misleading statement could cause users to unnecessarily enable a preview feature for operator-managed deployments.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:28:36Z"
+          },
+          {
+            "path": "quarkus/tests/integration/src/test/java/org/keycloak/it/cli/dist/UpdateCommandDistTest.java",
+            "line": 48,
+            "body": "P3: The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode (it only calls `System.exit` when NOT in test mode). The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure. Since the docs explicitly document exit code 4 as part of the scripting contract, this is a meaningful gap.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>quarkus/tests/integration/src/test/java/org/keycloak/it/cli/dist/UpdateCommandDistTest.java</file_path>\n<line_numbers>48</line_numbers>\n\n<comment>The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode (it only calls `System.exit` when NOT in test mode). The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure. Since the docs explicitly document exit code 4 as part of the scripting contract, this is a meaningful gap.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>quarkus/tests/integration/src/test/java/org/keycloak/it/cli/dist/UpdateCommandDistTest.java</file_path>\n<line_numbers>48</line_numbers>\n\n<comment>The `FEATURE_DISABLED` exit code (4) is never verified by any test. The `testFeatureNotEnabled` test uses `@Launch`, but `picocli.exit()` is a no-op in test launch mode (it only calls `System.exit` when NOT in test mode). The test only verifies the error message, not the exit code. No `distribution.run()` test exists for the feature-disabled scenario, so exit code 4 could be changed to any value without test failure. Since the docs explicitly document exit code 4 as part of the scripting contract, this is a meaningful gap.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:28:36Z"
+          },
+          {
+            "path": "quarkus/tests/integration/src/test/java/org/keycloak/it/cli/dist/UpdateCommandDistTest.java",
+            "line": 47,
+            "body": "P4: `testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand (`UpdateCompatibilityMetadata.NAME`), but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested. Since these are two separate code paths (lines 46-50 in both files), a bug in one wouldn't be caught.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>quarkus/tests/integration/src/test/java/org/keycloak/it/cli/dist/UpdateCommandDistTest.java</file_path>\n<line_numbers>47</line_numbers>\n\n<comment>`testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand (`UpdateCompatibilityMetadata.NAME`), but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested. Since these are two separate code paths (lines 46-50 in both files), a bug in one wouldn't be caught.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>quarkus/tests/integration/src/test/java/org/keycloak/it/cli/dist/UpdateCommandDistTest.java</file_path>\n<line_numbers>47</line_numbers>\n\n<comment>`testFeatureNotEnabled` only verifies the feature-disabled error for the `metadata` subcommand (`UpdateCompatibilityMetadata.NAME`), but both `UpdateCompatibilityCheck` and `UpdateCompatibilityMetadata` have independent feature checks in their `run()` methods. The `check` subcommand's feature-disabled path is never tested. Since these are two separate code paths (lines 46-50 in both files), a bug in one wouldn't be caught.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:28:36Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:28:36Z"
+          }
+        ]
       }
     ]
   },
@@ -5122,6 +5246,43 @@
             "created_at": "2026-03-18T20:26:46Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR36880__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36880__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java",
+            "line": 219,
+            "body": "P2: Potential NullPointerException: `getResourceTypeResource()` can return null (e.g., if the authorization schema resource hasn't been created yet), but the result is passed directly to `findByResource(server, resource)` without a null check. The `PolicyStore.findByResource` contract explicitly states the resource parameter cannot be null. If null is passed, it will NPE. A null check should be added after line 217, returning false if resource is null.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java</file_path>\n<line_numbers>219</line_numbers>\n\n<comment>Potential NullPointerException: `getResourceTypeResource()` can return null (e.g., if the authorization schema resource hasn't been created yet), but the result is passed directly to `findByResource(server, resource)` without a null check. The `PolicyStore.findByResource` contract explicitly states the resource parameter cannot be null. If null is passed, it will NPE. A null check should be added after line 217, returning false if resource is null.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java</file_path>\n<line_numbers>219</line_numbers>\n\n<comment>Potential NullPointerException: `getResourceTypeResource()` can return null (e.g., if the authorization schema resource hasn't been created yet), but the result is passed directly to `findByResource(server, resource)` without a null check. The `PolicyStore.findByResource` contract explicitly states the resource parameter cannot be null. If null is passed, it will NPE. A null check should be added after line 217, returning false if resource is null.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:37Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java",
+            "line": 138,
+            "body": "P2: getClientsWithPermission is broken for V2: `findByType(server, \"Clients\", ...)` only finds the all-clients type resource (which has type=\"Clients\" set during initialization), NOT individual per-client resources (created via `getOrCreateResource` which never calls `resource.setType()`). This causes two problems: (1) Per-client permissions are invisible \u2014 the method returns an empty set even when the user has permission on specific clients. (2) When all-clients permission exists, the type resource name \"Clients\" is added to the result set. The caller `AvailableRoleMappingResource.getRoleIdsWithPermissions` then calls `realm.getClientById(\"Clients\")` which returns null, causing an NPE on `.getRolesStream()`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java</file_path>\n<line_numbers>138</line_numbers>\n\n<comment>getClientsWithPermission is broken for V2: `findByType(server, \"Clients\", ...)` only finds the all-clients type resource (which has type=\"Clients\" set during initialization), NOT individual per-client resources (created via `getOrCreateResource` which never calls `resource.setType()`). This causes two problems: (1) Per-client permissions are invisible \u2014 the method returns an empty set even when the user has permission on specific clients. (2) When all-clients permission exists, the type resource name \"Clients\" is added to the result set. The caller `AvailableRoleMappingResource.getRoleIdsWithPermissions` then calls `realm.getClientById(\"Clients\")` which returns null, causing an NPE on `.getRolesStream()`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java</file_path>\n<line_numbers>138</line_numbers>\n\n<comment>getClientsWithPermission is broken for V2: `findByType(server, \"Clients\", ...)` only finds the all-clients type resource (which has type=\"Clients\" set during initialization), NOT individual per-client resources (created via `getOrCreateResource` which never calls `resource.setType()`). This causes two problems: (1) Per-client permissions are invisible \u2014 the method returns an empty set even when the user has permission on specific clients. (2) When all-clients permission exists, the type resource name \"Clients\" is added to the result set. The caller `AvailableRoleMappingResource.getRoleIdsWithPermissions` then calls `realm.getClientById(\"Clients\")` which returns null, causing an NPE on `.getRolesStream()`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:37Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java",
+            "line": 91,
+            "body": "P3: The `canMapRoles`, `canMapCompositeRoles`, and `canMapClientScopeRoles` overrides in ClientPermissionsV2 do not check the `MANAGE_CLIENTS` admin role, unlike `canManage(ClientModel)` and other methods which check `root.hasOneAdminRole(AdminRoles.MANAGE_CLIENTS)` first. This means a user with the `MANAGE_CLIENTS` realm admin role cannot map roles to/from clients unless they also have explicit fine-grained permissions. While V1 also doesn't check admin roles in these methods, the V2 `canManage(ClientModel)` establishes the pattern of checking admin roles first, and the interface doc for `canMapRoles` says V2 should check 'if the caller has a permission to MAP_ROLES for all clients' \u2014 but it doesn't check the admin role that grants full client management.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java</file_path>\n<line_numbers>91</line_numbers>\n\n<comment>The `canMapRoles`, `canMapCompositeRoles`, and `canMapClientScopeRoles` overrides in ClientPermissionsV2 do not check the `MANAGE_CLIENTS` admin role, unlike `canManage(ClientModel)` and other methods which check `root.hasOneAdminRole(AdminRoles.MANAGE_CLIENTS)` first. This means a user with the `MANAGE_CLIENTS` realm admin role cannot map roles to/from clients unless they also have explicit fine-grained permissions. While V1 also doesn't check admin roles in these methods, the V2 `canManage(ClientModel)` establishes the pattern of checking admin roles first, and the interface doc for `canMapRoles` says V2 should check 'if the caller has a permission to MAP_ROLES for all clients' \u2014 but it doesn't check the admin role that grants full client management.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionsV2.java</file_path>\n<line_numbers>91</line_numbers>\n\n<comment>The `canMapRoles`, `canMapCompositeRoles`, and `canMapClientScopeRoles` overrides in ClientPermissionsV2 do not check the `MANAGE_CLIENTS` admin role, unlike `canManage(ClientModel)` and other methods which check `root.hasOneAdminRole(AdminRoles.MANAGE_CLIENTS)` first. This means a user with the `MANAGE_CLIENTS` realm admin role cannot map roles to/from clients unless they also have explicit fine-grained permissions. While V1 also doesn't check admin roles in these methods, the V2 `canManage(ClientModel)` establishes the pattern of checking admin roles first, and the interface doc for `canMapRoles` says V2 should check 'if the caller has a permission to MAP_ROLES for all clients' \u2014 but it doesn't check the admin role that grants full client management.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:37Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionEvaluator.java",
+            "line": 99,
+            "body": "P4: Javadoc for `requireView()` says 'Returns {@code true} if canView() returns true' but the method is `void` and should throw ForbiddenException when the check fails. This is a copy-paste error from the `canView()` doc. The same error also applies to `requireViewClientScopes()`. Compare with the correctly documented `requireManage()` which says 'Throws ForbiddenException if canManage() returns false'.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionEvaluator.java</file_path>\n<line_numbers>99</line_numbers>\n\n<comment>Javadoc for `requireView()` says 'Returns {@code true} if canView() returns true' but the method is `void` and should throw ForbiddenException when the check fails. This is a copy-paste error from the `canView()` doc. The same error also applies to `requireViewClientScopes()`. Compare with the correctly documented `requireManage()` which says 'Throws ForbiddenException if canManage() returns false'.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/ClientPermissionEvaluator.java</file_path>\n<line_numbers>99</line_numbers>\n\n<comment>Javadoc for `requireView()` says 'Returns {@code true} if canView() returns true' but the method is `void` and should throw ForbiddenException when the check fails. This is a copy-paste error from the `canView()` doc. The same error also applies to `requireViewClientScopes()`. Compare with the correctly documented `requireManage()` which says 'Throws ForbiddenException if canManage() returns false'.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:37Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:22:37Z"
+          }
+        ]
       }
     ]
   },
@@ -6224,6 +6385,43 @@
             "created_at": "2026-03-18T20:26:43Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR37038__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37038__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionsV2.java",
+            "line": 70,
+            "body": "P1: GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, meaning a user with only VIEW permission on groups can pass canManage() checks. This gates group creation via GroupsResource.addTopLevelGroup() which calls auth.groups().requireManage(). The per-group canManage(GroupModel) correctly checks only MANAGE. This is a copy-paste from canView() where checking both scopes is correct.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionsV2.java</file_path>\n<line_numbers>70</line_numbers>\n\n<comment>GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, meaning a user with only VIEW permission on groups can pass canManage() checks. This gates group creation via GroupsResource.addTopLevelGroup() which calls auth.groups().requireManage(). The per-group canManage(GroupModel) correctly checks only MANAGE. This is a copy-paste from canView() where checking both scopes is correct.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionsV2.java</file_path>\n<line_numbers>70</line_numbers>\n\n<comment>GroupPermissionsV2.canManage() (no-arg) checks for both VIEW and MANAGE scopes, meaning a user with only VIEW permission on groups can pass canManage() checks. This gates group creation via GroupsResource.addTopLevelGroup() which calls auth.groups().requireManage(). The per-group canManage(GroupModel) correctly checks only MANAGE. This is a copy-paste from canView() where checking both scopes is correct.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:27Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionsV2.java",
+            "line": 122,
+            "body": "P1: getGroupIdsWithViewPermission() uses groupResource.getId() (the authorization resource server's internal UUID) instead of groupResource.getName() (which stores the actual group model ID). This causes two problems: (1) hasPermission() at line 122 calls resourceStore.findByName(server, <resource-internal-id>) which won't find the resource by name, so it always falls through to the 'all-groups' check, making per-group permission evaluation broken. (2) granted.add(groupResource.getId()) adds resource UUIDs, but downstream in JpaUserProvider.groupsWithPermissionsSubquery() they're compared against the groupId column in the user-group membership table, which stores group model IDs \u2014 so the query filter will never match.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionsV2.java</file_path>\n<line_numbers>122</line_numbers>\n\n<comment>getGroupIdsWithViewPermission() uses groupResource.getId() (the authorization resource server's internal UUID) instead of groupResource.getName() (which stores the actual group model ID). This causes two problems: (1) hasPermission() at line 122 calls resourceStore.findByName(server, <resource-internal-id>) which won't find the resource by name, so it always falls through to the 'all-groups' check, making per-group permission evaluation broken. (2) granted.add(groupResource.getId()) adds resource UUIDs, but downstream in JpaUserProvider.groupsWithPermissionsSubquery() they're compared against the groupId column in the user-group membership table, which stores group model IDs \u2014 so the query filter will never match.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionsV2.java</file_path>\n<line_numbers>122</line_numbers>\n\n<comment>getGroupIdsWithViewPermission() uses groupResource.getId() (the authorization resource server's internal UUID) instead of groupResource.getName() (which stores the actual group model ID). This causes two problems: (1) hasPermission() at line 122 calls resourceStore.findByName(server, <resource-internal-id>) which won't find the resource by name, so it always falls through to the 'all-groups' check, making per-group permission evaluation broken. (2) granted.add(groupResource.getId()) adds resource UUIDs, but downstream in JpaUserProvider.groupsWithPermissionsSubquery() they're compared against the groupId column in the user-group membership table, which stores group model IDs \u2014 so the query filter will never match.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:27Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionEvaluator.java",
+            "line": 73,
+            "body": "P3: The interface docstring for canManage() states 'Also if it has permission to VIEW or MANAGE groups', which codifies the privilege escalation as intentional. If canManage() is fixed to only check MANAGE, this docstring is misleading and should be updated. If the docstring is 'correct', then canManage() is intentionally granting manage privileges to view-only users, which is a design-level security issue.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionEvaluator.java</file_path>\n<line_numbers>73</line_numbers>\n\n<comment>The interface docstring for canManage() states 'Also if it has permission to VIEW or MANAGE groups', which codifies the privilege escalation as intentional. If canManage() is fixed to only check MANAGE, this docstring is misleading and should be updated. If the docstring is 'correct', then canManage() is intentionally granting manage privileges to view-only users, which is a design-level security issue.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/services/resources/admin/permissions/GroupPermissionEvaluator.java</file_path>\n<line_numbers>73</line_numbers>\n\n<comment>The interface docstring for canManage() states 'Also if it has permission to VIEW or MANAGE groups', which codifies the privilege escalation as intentional. If canManage() is fixed to only check MANAGE, this docstring is misleading and should be updated. If the docstring is 'correct', then canManage() is intentionally granting manage privileges to view-only users, which is a design-level security issue.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:27Z"
+          },
+          {
+            "path": "tests/base/src/test/java/org/keycloak/tests/admin/authz/fgap/GroupResourceTypeEvaluationTest.java",
+            "line": 75,
+            "body": "P4: In the test @BeforeEach method, `ApiUtil.handleCreatedResponse(response)` is called on a response that's wrapped in a try-with-resources. Since handleCreatedResponse was refactored to also use try-with-resources internally (closing the response), the response is asserted for status 201 twice (once by the test, once by handleCreatedResponse) and closed twice. More importantly, handleCreatedResponse may attempt to read Location headers from a response whose input stream state is uncertain after the assertion in handleCreatedResponse closes it, then the outer block closes again. This pattern is used at lines 73-77, 195-199, and 261-265.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/base/src/test/java/org/keycloak/tests/admin/authz/fgap/GroupResourceTypeEvaluationTest.java</file_path>\n<line_numbers>75</line_numbers>\n\n<comment>In the test @BeforeEach method, `ApiUtil.handleCreatedResponse(response)` is called on a response that's wrapped in a try-with-resources. Since handleCreatedResponse was refactored to also use try-with-resources internally (closing the response), the response is asserted for status 201 twice (once by the test, once by handleCreatedResponse) and closed twice. More importantly, handleCreatedResponse may attempt to read Location headers from a response whose input stream state is uncertain after the assertion in handleCreatedResponse closes it, then the outer block closes again. This pattern is used at lines 73-77, 195-199, and 261-265.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/base/src/test/java/org/keycloak/tests/admin/authz/fgap/GroupResourceTypeEvaluationTest.java</file_path>\n<line_numbers>75</line_numbers>\n\n<comment>In the test @BeforeEach method, `ApiUtil.handleCreatedResponse(response)` is called on a response that's wrapped in a try-with-resources. Since handleCreatedResponse was refactored to also use try-with-resources internally (closing the response), the response is asserted for status 201 twice (once by the test, once by handleCreatedResponse) and closed twice. More importantly, handleCreatedResponse may attempt to read Location headers from a response whose input stream state is uncertain after the assertion in handleCreatedResponse closes it, then the outer block closes again. This pattern is used at lines 73-77, 195-199, and 261-265.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:18:27Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:18:27Z"
+          }
+        ]
       }
     ]
   },
@@ -7255,6 +7453,43 @@
             "created_at": "2026-03-18T20:27:41Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR33832__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR33832__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "common/src/main/java/org/keycloak/common/crypto/CryptoProvider.java",
+            "line": 44,
+            "body": "P2: The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>common/src/main/java/org/keycloak/common/crypto/CryptoProvider.java</file_path>\n<line_numbers>44</line_numbers>\n\n<comment>The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>common/src/main/java/org/keycloak/common/crypto/CryptoProvider.java</file_path>\n<line_numbers>44</line_numbers>\n\n<comment>The `order()` method is added to the `CryptoProvider` interface without a `default` implementation. Since `CryptoProvider` is a public SPI loaded via `ServiceLoader`, any external implementation of this interface will fail to compile or load after this change. It should be `default int order() { return 0; }` so that existing third-party providers automatically get the lowest priority without breaking.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:33Z"
+          },
+          {
+            "path": "authz/client/src/test/java/org/keycloak/authorization/client/test/ECDSAAlgorithmTest.java",
+            "line": 42,
+            "body": "P3: The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures. This means bugs specific to handling full-size ES384/ES512 r/s values (e.g., off-by-one in `integerToBytes`) would not be caught.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>authz/client/src/test/java/org/keycloak/authorization/client/test/ECDSAAlgorithmTest.java</file_path>\n<line_numbers>42</line_numbers>\n\n<comment>The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures. This means bugs specific to handling full-size ES384/ES512 r/s values (e.g., off-by-one in `integerToBytes`) would not be caught.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>authz/client/src/test/java/org/keycloak/authorization/client/test/ECDSAAlgorithmTest.java</file_path>\n<line_numbers>42</line_numbers>\n\n<comment>The test generates a single EC key pair using the default curve (P-256/secp256r1) in the constructor, then reuses it for `testES384` and `testES512`. A P-256 key produces r/s values of ~32 bytes, but ES384 expects 48-byte values and ES512 expects 66-byte values. The test only verifies round-trip consistency of zero-padded values rather than validating the conversion with correctly-sized signatures. This means bugs specific to handling full-size ES384/ES512 r/s values (e.g., off-by-one in `integerToBytes`) would not be caught.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:33Z"
+          },
+          {
+            "path": "authz/client/src/main/java/org/keycloak/authorization/client/util/crypto/AuthzClientCryptoProvider.java",
+            "line": 114,
+            "body": "P4: Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>authz/client/src/main/java/org/keycloak/authorization/client/util/crypto/AuthzClientCryptoProvider.java</file_path>\n<line_numbers>114</line_numbers>\n\n<comment>Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>authz/client/src/main/java/org/keycloak/authorization/client/util/crypto/AuthzClientCryptoProvider.java</file_path>\n<line_numbers>114</line_numbers>\n\n<comment>Lines 114-115 create two `ASN1Encoder` instances, write integers to them, and immediately discard the results. These are dead statements \u2014 the actual encoding is performed on lines 117-121. This appears to be leftover debugging/development code that wastes allocations on every invocation of `concatenatedRSToASN1DER`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:33Z"
+          },
+          {
+            "path": "common/src/main/java/org/keycloak/common/crypto/CryptoIntegration.java",
+            "line": 67,
+            "body": "P4: In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name (e.g., `\"Ignored crypto providers: FooProvider, \"`). This produces a malformed log message.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>common/src/main/java/org/keycloak/common/crypto/CryptoIntegration.java</file_path>\n<line_numbers>67</line_numbers>\n\n<comment>In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name (e.g., `\"Ignored crypto providers: FooProvider, \"`). This produces a malformed log message.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>common/src/main/java/org/keycloak/common/crypto/CryptoIntegration.java</file_path>\n<line_numbers>67</line_numbers>\n\n<comment>In `CryptoIntegration.detectProvider`, when multiple providers are found, the \"Ignored crypto providers\" log message unconditionally appends a trailing \", \" after the last provider name (e.g., `\"Ignored crypto providers: FooProvider, \"`). This produces a malformed log message.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:22:33Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:22:33Z"
+          }
+        ]
       }
     ]
   },
@@ -8190,6 +8425,31 @@
             "created_at": "2026-03-18T20:26:48Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR40940__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR40940__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/GroupAdapter.java",
+            "line": 271,
+            "body": "P2: _Note: Issue is at line 256, shown here on nearest changed line._\n\nIncomplete NPE fix: `getSubGroupsStream(String, Integer, Integer)` at line 256 calls `modelSupplier.get().getSubGroupsStream(...)` without a null check, which will throw an NPE under the same concurrent-deletion scenario that this PR fixes in `getSubGroupsCount()`. The same applies to the overloads at lines 262 and 268. All three methods should guard against `modelSupplier.get()` returning null, just as the fixed `getSubGroupsCount()` now does.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/GroupAdapter.java</file_path>\n<line_numbers>271</line_numbers>\n\n<comment>_Note: Issue is at line 256, shown here on nearest changed line._\n\nIncomplete NPE fix: `getSubGroupsStream(String, Integer, Integer)` at line 256 calls `modelSupplier.get().getSubGroupsStream(...)` without a null check, which will throw an NPE under the same concurrent-deletion scenario that this PR fixes in `getSubGroupsCount()`. The same applies to the overloads at lines 262 and 268. All three methods should guard against `modelSupplier.get()` returning null, just as the fixed `getSubGroupsCount()` now does.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/GroupAdapter.java</file_path>\n<line_numbers>271</line_numbers>\n\n<comment>_Note: Issue is at line 256, shown here on nearest changed line._\n\nIncomplete NPE fix: `getSubGroupsStream(String, Integer, Integer)` at line 256 calls `modelSupplier.get().getSubGroupsStream(...)` without a null check, which will throw an NPE under the same concurrent-deletion scenario that this PR fixes in `getSubGroupsCount()`. The same applies to the overloads at lines 262 and 268. All three methods should guard against `modelSupplier.get()` returning null, just as the fixed `getSubGroupsCount()` now does.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:11:17Z"
+          },
+          {
+            "path": "tests/base/src/test/java/org/keycloak/tests/admin/group/GroupTest.java",
+            "line": 157,
+            "body": "P3: The reader thread is never joined before the assertion. After `deletedAll.set(true)` on line 155, the test asserts `caughtExceptions` is empty on line 157, but the reader thread may still be in the middle of a `groups()` call that was started before `deletedAll` was observed. If that in-flight call throws an exception, it will be added to `caughtExceptions` after the assertion has already passed, causing the test to silently miss the very failures it's designed to detect.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/base/src/test/java/org/keycloak/tests/admin/group/GroupTest.java</file_path>\n<line_numbers>157</line_numbers>\n\n<comment>The reader thread is never joined before the assertion. After `deletedAll.set(true)` on line 155, the test asserts `caughtExceptions` is empty on line 157, but the reader thread may still be in the middle of a `groups()` call that was started before `deletedAll` was observed. If that in-flight call throws an exception, it will be added to `caughtExceptions` after the assertion has already passed, causing the test to silently miss the very failures it's designed to detect.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/base/src/test/java/org/keycloak/tests/admin/group/GroupTest.java</file_path>\n<line_numbers>157</line_numbers>\n\n<comment>The reader thread is never joined before the assertion. After `deletedAll.set(true)` on line 155, the test asserts `caughtExceptions` is empty on line 157, but the reader thread may still be in the middle of a `groups()` call that was started before `deletedAll` was observed. If that in-flight call throws an exception, it will be added to `caughtExceptions` after the assertion has already passed, causing the test to silently miss the very failures it's designed to detect.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:11:17Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 2 issues",
+            "created_at": "2026-04-12T02:11:17Z"
+          }
+        ]
       }
     ]
   },
@@ -9239,6 +9499,43 @@
             "created_at": "2026-03-18T20:26:47Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak-greptile__tusk__PR1__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak-greptile__tusk__PR1__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernameForm.java",
+            "line": 47,
+            "body": "P1: Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the second commit (3214b18) changed the method signature in the parent class UsernamePasswordForm to `isConditionalPasskeysEnabled(UserModel user)`, requiring one parameter. This code will not compile. The call should be `isConditionalPasskeysEnabled(context.getUser())`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernameForm.java</file_path>\n<line_numbers>47</line_numbers>\n\n<comment>Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the second commit (3214b18) changed the method signature in the parent class UsernamePasswordForm to `isConditionalPasskeysEnabled(UserModel user)`, requiring one parameter. This code will not compile. The call should be `isConditionalPasskeysEnabled(context.getUser())`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernameForm.java</file_path>\n<line_numbers>47</line_numbers>\n\n<comment>Compilation error: UsernameForm.java calls `isConditionalPasskeysEnabled()` with no arguments, but the second commit (3214b18) changed the method signature in the parent class UsernamePasswordForm to `isConditionalPasskeysEnabled(UserModel user)`, requiring one parameter. This code will not compile. The call should be `isConditionalPasskeysEnabled(context.getUser())`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:26:36Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernamePasswordForm.java",
+            "line": 115,
+            "body": "P2: Regression: passkeys no longer set up during initial login. In `authenticate()`, the old code called `fillContextForm` when `context.getUser() == null` (inside the else branch), which is the initial login case. The new code calls it only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This means the WebAuthn challenge, rpId, and conditional UI attributes are never set on the login form for first-time visitors, breaking passkey conditional UI (autocomplete=\"webauthn\") on the initial login page.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernamePasswordForm.java</file_path>\n<line_numbers>115</line_numbers>\n\n<comment>Regression: passkeys no longer set up during initial login. In `authenticate()`, the old code called `fillContextForm` when `context.getUser() == null` (inside the else branch), which is the initial login case. The new code calls it only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This means the WebAuthn challenge, rpId, and conditional UI attributes are never set on the login form for first-time visitors, breaking passkey conditional UI (autocomplete=\"webauthn\") on the initial login page.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernamePasswordForm.java</file_path>\n<line_numbers>115</line_numbers>\n\n<comment>Regression: passkeys no longer set up during initial login. In `authenticate()`, the old code called `fillContextForm` when `context.getUser() == null` (inside the else branch), which is the initial login case. The new code calls it only when `isConditionalPasskeysEnabled(context.getUser())` returns true, which requires `user != null`. This means the WebAuthn challenge, rpId, and conditional UI attributes are never set on the login form for first-time visitors, breaking passkey conditional UI (autocomplete=\"webauthn\") on the initial login page.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:26:36Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernamePasswordForm.java",
+            "line": 137,
+            "body": "P3: Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null` (initial login error recovery). The new code calls it only when `user != null`. This means if a user enters an invalid username during initial login with passkeys enabled, the passkey UI elements will not be re-rendered on the error page, preventing the user from switching to passkey login after an initial password error.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernamePasswordForm.java</file_path>\n<line_numbers>137</line_numbers>\n\n<comment>Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null` (initial login error recovery). The new code calls it only when `user != null`. This means if a user enters an invalid username during initial login with passkeys enabled, the passkey UI elements will not be re-rendered on the error page, preventing the user from switching to passkey login after an initial password error.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/UsernamePasswordForm.java</file_path>\n<line_numbers>137</line_numbers>\n\n<comment>Behavior reversal in error challenge path: the old code in `challenge(context, error, field)` called `fillContextForm` when `context.getUser() == null` (initial login error recovery). The new code calls it only when `user != null`. This means if a user enters an invalid username during initial login with passkeys enabled, the passkey UI elements will not be re-rendered on the error page, preventing the user from switching to passkey login after an initial password error.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:26:36Z"
+          },
+          {
+            "path": "services/src/main/java/org/keycloak/authentication/authenticators/browser/WebAuthnAuthenticator.java",
+            "line": 103,
+            "body": "P4: Potential NullPointerException in `fillContextForm`: The refactored code in `WebAuthnAuthenticator.fillContextForm()` at line 103 passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. The base implementation of `shouldShowWebAuthnAuthenticators` checks `context.getUser() != null`, which is safe. However, the method is now `protected` and designed for override (WebAuthnConditionalUIAuthenticator already overrides it to return `false`). A future subclass could override it to return `true` without a null-user guard, causing NPE on line 103 where `user` would be null. The old code had a direct `if (user != null)` check that was inherently safe.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/WebAuthnAuthenticator.java</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>Potential NullPointerException in `fillContextForm`: The refactored code in `WebAuthnAuthenticator.fillContextForm()` at line 103 passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. The base implementation of `shouldShowWebAuthnAuthenticators` checks `context.getUser() != null`, which is safe. However, the method is now `protected` and designed for override (WebAuthnConditionalUIAuthenticator already overrides it to return `false`). A future subclass could override it to return `true` without a null-user guard, causing NPE on line 103 where `user` would be null. The old code had a direct `if (user != null)` check that was inherently safe.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>services/src/main/java/org/keycloak/authentication/authenticators/browser/WebAuthnAuthenticator.java</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>Potential NullPointerException in `fillContextForm`: The refactored code in `WebAuthnAuthenticator.fillContextForm()` at line 103 passes `user` to `WebAuthnAuthenticatorsBean` constructor inside the `if (shouldShowWebAuthnAuthenticators(context))` block. The base implementation of `shouldShowWebAuthnAuthenticators` checks `context.getUser() != null`, which is safe. However, the method is now `protected` and designed for override (WebAuthnConditionalUIAuthenticator already overrides it to return `false`). A future subclass could override it to return `true` without a null-user guard, causing NPE on line 103 where `user` would be null. The old code had a direct `if (user != null)` check that was inherently safe.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:26:36Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:26:36Z"
+          }
+        ]
       }
     ]
   },
@@ -10318,6 +10615,37 @@
             "created_at": "2026-03-18T20:27:05Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry__tusk__PR93824__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR93824__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/spans/consumers/process/flusher.py",
+            "line": 199,
+            "body": "P3: Inconsistent metrics tag key: `\"shards\"` (plural) is used here, while all other metrics in the same function use `\"shard\"` (singular) as the tag key. This will cause this metric to be silently missing from any dashboards or alerts that filter on the `shard` tag.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/spans/consumers/process/flusher.py</file_path>\n<line_numbers>199</line_numbers>\n\n<comment>Inconsistent metrics tag key: `\"shards\"` (plural) is used here, while all other metrics in the same function use `\"shard\"` (singular) as the tag key. This will cause this metric to be silently missing from any dashboards or alerts that filter on the `shard` tag.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/spans/consumers/process/flusher.py</file_path>\n<line_numbers>199</line_numbers>\n\n<comment>Inconsistent metrics tag key: `\"shards\"` (plural) is used here, while all other metrics in the same function use `\"shard\"` (singular) as the tag key. This will cause this metric to be silently missing from any dashboards or alerts that filter on the `shard` tag.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:57:14Z"
+          },
+          {
+            "path": "src/sentry/spans/consumers/process/flusher.py",
+            "line": 340,
+            "body": "P3: When the deadline is exceeded during `join()`, the `break` on line 341 exits the loop entirely, skipping `.terminate()` for all remaining processes. In the old single-process code, terminate was always called. With multiple processes, if process 0 takes the entire timeout, processes 1..N never get terminated. While `stopped.value=True` is set, a hung process (e.g., blocked on Redis/Kafka I/O) won't observe the flag and will remain alive as a zombie until the parent exits.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/spans/consumers/process/flusher.py</file_path>\n<line_numbers>340</line_numbers>\n\n<comment>When the deadline is exceeded during `join()`, the `break` on line 341 exits the loop entirely, skipping `.terminate()` for all remaining processes. In the old single-process code, terminate was always called. With multiple processes, if process 0 takes the entire timeout, processes 1..N never get terminated. While `stopped.value=True` is set, a hung process (e.g., blocked on Redis/Kafka I/O) won't observe the flag and will remain alive as a zombie until the parent exits.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/spans/consumers/process/flusher.py</file_path>\n<line_numbers>340</line_numbers>\n\n<comment>When the deadline is exceeded during `join()`, the `break` on line 341 exits the loop entirely, skipping `.terminate()` for all remaining processes. In the old single-process code, terminate was always called. With multiple processes, if process 0 takes the entire timeout, processes 1..N never get terminated. While `stopped.value=True` is set, a hung process (e.g., blocked on Redis/Kafka I/O) won't observe the flag and will remain alive as a zombie until the parent exits.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:57:14Z"
+          },
+          {
+            "path": "src/sentry/spans/consumers/process/flusher.py",
+            "line": 300,
+            "body": "P4: Each sub-buffer's `get_memory_info()` calls `iter_cluster_memory_usage(self.client)`, which queries Redis `INFO` on ALL nodes in the cluster. Since all sub-buffers share the same Redis cluster (via the module-level `get_redis_client()`), calling this N times (once per sub-buffer) makes N redundant Redis INFO roundtrips. The ratio `used/available` is mathematically unchanged (both numerator and denominator are multiplied by N), so the check is correct, but the extra INFO calls add unnecessary latency to each `submit()` call on the hot path.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/spans/consumers/process/flusher.py</file_path>\n<line_numbers>300</line_numbers>\n\n<comment>Each sub-buffer's `get_memory_info()` calls `iter_cluster_memory_usage(self.client)`, which queries Redis `INFO` on ALL nodes in the cluster. Since all sub-buffers share the same Redis cluster (via the module-level `get_redis_client()`), calling this N times (once per sub-buffer) makes N redundant Redis INFO roundtrips. The ratio `used/available` is mathematically unchanged (both numerator and denominator are multiplied by N), so the check is correct, but the extra INFO calls add unnecessary latency to each `submit()` call on the hot path.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/spans/consumers/process/flusher.py</file_path>\n<line_numbers>300</line_numbers>\n\n<comment>Each sub-buffer's `get_memory_info()` calls `iter_cluster_memory_usage(self.client)`, which queries Redis `INFO` on ALL nodes in the cluster. Since all sub-buffers share the same Redis cluster (via the module-level `get_redis_client()`), calling this N times (once per sub-buffer) makes N redundant Redis INFO roundtrips. The ratio `used/available` is mathematically unchanged (both numerator and denominator are multiplied by N), so the check is correct, but the extra INFO calls add unnecessary latency to each `submit()` call on the hot path.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:57:15Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T01:57:14Z"
+          }
+        ]
       }
     ]
   },
@@ -11718,6 +12046,31 @@
             "created_at": "2026-03-18T20:57:26Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry-greptile__tusk__PR5__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR5__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/replays/endpoints/project_replay_summarize_breadcrumbs.py",
+            "line": 118,
+            "body": "P2: fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` in nodestore/base.py:205 appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list. This silently pairs wrong error IDs with wrong event data (title, message, timestamp), corrupting the error context sent to the AI summarizer. The fix should iterate over `node_ids` and look up results by key: `for node_id, error_id in zip(node_ids, error_ids): data = events.get(node_id); if data is not None: ...`\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/replays/endpoints/project_replay_summarize_breadcrumbs.py</file_path>\n<line_numbers>118</line_numbers>\n\n<comment>fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` in nodestore/base.py:205 appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list. This silently pairs wrong error IDs with wrong event data (title, message, timestamp), corrupting the error context sent to the AI summarizer. The fix should iterate over `node_ids` and look up results by key: `for node_id, error_id in zip(node_ids, error_ids): data = events.get(node_id); if data is not None: ...`</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/replays/endpoints/project_replay_summarize_breadcrumbs.py</file_path>\n<line_numbers>118</line_numbers>\n\n<comment>fetch_error_details uses `zip(error_ids, events.values())` but `nodestore.backend.get_multi()` returns a dict keyed by node_ids, not error_ids. When some items are cached and others aren't, `items.update(cache_items)` in nodestore/base.py:205 appends cached entries at the end of the dict, breaking insertion-order alignment with the original `error_ids` list. This silently pairs wrong error IDs with wrong event data (title, message, timestamp), corrupting the error context sent to the AI summarizer. The fix should iterate over `node_ids` and look up results by key: `for node_id, error_id in zip(node_ids, error_ids): data = events.get(node_id); if data is not None: ...`</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:58:05Z"
+          },
+          {
+            "path": "src/sentry/replays/endpoints/project_replay_summarize_breadcrumbs.py",
+            "line": 153,
+            "body": "P3: In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds like 1713000000.0) are compared directly against replay segment event timestamps (`event.get('timestamp', 0)`). Standard rrweb replay events use Unix epoch milliseconds (like 1713000000000). This unit mismatch means all error events will always appear to have timestamps far earlier than replay events, causing them to be yielded before any replay breadcrumbs rather than interleaved in chronological order. The unit test passes only because it uses the same small float values for both.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/replays/endpoints/project_replay_summarize_breadcrumbs.py</file_path>\n<line_numbers>153</line_numbers>\n\n<comment>In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds like 1713000000.0) are compared directly against replay segment event timestamps (`event.get('timestamp', 0)`). Standard rrweb replay events use Unix epoch milliseconds (like 1713000000000). This unit mismatch means all error events will always appear to have timestamps far earlier than replay events, causing them to be yielded before any replay breadcrumbs rather than interleaved in chronological order. The unit test passes only because it uses the same small float values for both.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/replays/endpoints/project_replay_summarize_breadcrumbs.py</file_path>\n<line_numbers>153</line_numbers>\n\n<comment>In `gen_request_data`, error event timestamps (from nodestore, stored as Unix epoch seconds like 1713000000.0) are compared directly against replay segment event timestamps (`event.get('timestamp', 0)`). Standard rrweb replay events use Unix epoch milliseconds (like 1713000000000). This unit mismatch means all error events will always appear to have timestamps far earlier than replay events, causing them to be yielded before any replay breadcrumbs rather than interleaved in chronological order. The unit test passes only because it uses the same small float values for both.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:58:05Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 2 issues",
+            "created_at": "2026-04-12T01:58:05Z"
+          }
+        ]
       }
     ]
   },
@@ -12973,6 +13326,49 @@
             "created_at": "2026-03-18T20:59:23Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry-greptile__tusk__PR1__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR1__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/api/paginator.py",
+            "line": 840,
+            "body": "P1: OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The only caller (organization_auditlogs.py) uses order_by='-datetime', meaning self.key is 'datetime' and getattr(item, self.key) returns a datetime. Compare with DateTimePaginator.get_item_key (line 235) which correctly converts via float(value.strftime('%s.%f')) before calling math.floor().\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>840</line_numbers>\n\n<comment>OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The only caller (organization_auditlogs.py) uses order_by='-datetime', meaning self.key is 'datetime' and getattr(item, self.key) returns a datetime. Compare with DateTimePaginator.get_item_key (line 235) which correctly converts via float(value.strftime('%s.%f')) before calling math.floor().</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>840</line_numbers>\n\n<comment>OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil directly on a datetime object, which raises TypeError. The only caller (organization_auditlogs.py) uses order_by='-datetime', meaning self.key is 'datetime' and getattr(item, self.key) returns a datetime. Compare with DateTimePaginator.get_item_key (line 235) which correctly converts via float(value.strftime('%s.%f')) before calling math.floor().</comment>\n```\n\n</details>",
+            "created_at": "2026-04-10T17:35:45Z"
+          },
+          {
+            "path": "src/sentry/api/paginator.py",
+            "line": 882,
+            "body": "P1: Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), `queryset[start_offset:stop]` will raise AssertionError('Negative indexing is not supported.'). The comment 'The underlying Django ORM properly handles negative slicing automatically' is incorrect.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>882</line_numbers>\n\n<comment>Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), `queryset[start_offset:stop]` will raise AssertionError('Negative indexing is not supported.'). The comment 'The underlying Django ORM properly handles negative slicing automatically' is incorrect.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>882</line_numbers>\n\n<comment>Django querysets do not support negative indexing. When enable_advanced_features is True and cursor.offset < 0 (achievable via user-supplied cursor string like '0:-5:0'), `queryset[start_offset:stop]` will raise AssertionError('Negative indexing is not supported.'). The comment 'The underlying Django ORM properly handles negative slicing automatically' is incorrect.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-10T17:35:45Z"
+          },
+          {
+            "path": "src/sentry/api/paginator.py",
+            "line": 843,
+            "body": "P1: OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter (e.g., `datetime__lte=<int>`), producing incorrect results or a database type error. Compare with DateTimePaginator.value_from_cursor (line 238-241) which converts the numeric cursor value back to a timezone-aware datetime.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>843</line_numbers>\n\n<comment>OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter (e.g., `datetime__lte=<int>`), producing incorrect results or a database type error. Compare with DateTimePaginator.value_from_cursor (line 238-241) which converts the numeric cursor value back to a timezone-aware datetime.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>843</line_numbers>\n\n<comment>OptimizedCursorPaginator.value_from_cursor returns the raw cursor.value (an int/float) without converting it back to a datetime. When used with order_by='-datetime', build_queryset will compare this raw number against a DateTimeField in a Django filter (e.g., `datetime__lte=<int>`), producing incorrect results or a database type error. Compare with DateTimePaginator.value_from_cursor (line 238-241) which converts the numeric cursor value back to a timezone-aware datetime.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-10T17:35:46Z"
+          },
+          {
+            "path": "src/sentry/api/endpoints/organization_auditlogs.py",
+            "line": 71,
+            "body": "P2: organization_context.member can be None (declared as `RpcOrganizationMember | None = None` at model.py:346) when the user has no membership in the org. Accessing `.has_global_access` on None raises AttributeError. This executes on EVERY request to this endpoint (not just optimized ones) because the `enable_advanced` check runs unconditionally before the `use_optimized` guard.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/endpoints/organization_auditlogs.py</file_path>\n<line_numbers>71</line_numbers>\n\n<comment>organization_context.member can be None (declared as `RpcOrganizationMember | None = None` at model.py:346) when the user has no membership in the org. Accessing `.has_global_access` on None raises AttributeError. This executes on EVERY request to this endpoint (not just optimized ones) because the `enable_advanced` check runs unconditionally before the `use_optimized` guard.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/endpoints/organization_auditlogs.py</file_path>\n<line_numbers>71</line_numbers>\n\n<comment>organization_context.member can be None (declared as `RpcOrganizationMember | None = None` at model.py:346) when the user has no membership in the org. Accessing `.has_global_access` on None raises AttributeError. This executes on EVERY request to this endpoint (not just optimized ones) because the `enable_advanced` check runs unconditionally before the `use_optimized` guard.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-10T17:35:46Z"
+          },
+          {
+            "path": "src/sentry/api/paginator.py",
+            "line": 182,
+            "body": "P2: BasePaginator (the base class used by ALL paginators across Sentry) now allows negative start_offset when cursor.is_prev is True: `start_offset = offset` without clamping. A malicious or malformed cursor string with a negative offset (e.g., '0:-3:1') will cause Django's AssertionError on queryset slicing. The original code used `stop = offset + limit + extra` with offset always starting from the parsed cursor, but the queryset slice `queryset[offset:stop]` was safe because Django's non-negative assertion would catch it immediately at the same line. Now the variable rename obscures that the safety semantics changed for is_prev paths.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>182</line_numbers>\n\n<comment>BasePaginator (the base class used by ALL paginators across Sentry) now allows negative start_offset when cursor.is_prev is True: `start_offset = offset` without clamping. A malicious or malformed cursor string with a negative offset (e.g., '0:-3:1') will cause Django's AssertionError on queryset slicing. The original code used `stop = offset + limit + extra` with offset always starting from the parsed cursor, but the queryset slice `queryset[offset:stop]` was safe because Django's non-negative assertion would catch it immediately at the same line. Now the variable rename obscures that the safety semantics changed for is_prev paths.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>182</line_numbers>\n\n<comment>BasePaginator (the base class used by ALL paginators across Sentry) now allows negative start_offset when cursor.is_prev is True: `start_offset = offset` without clamping. A malicious or malformed cursor string with a negative offset (e.g., '0:-3:1') will cause Django's AssertionError on queryset slicing. The original code used `stop = offset + limit + extra` with offset always starting from the parsed cursor, but the queryset slice `queryset[offset:stop]` was safe because Django's non-negative assertion would catch it immediately at the same line. Now the variable rename obscures that the safety semantics changed for is_prev paths.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-10T17:35:46Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-10T17:35:45Z"
+          }
+        ]
       }
     ]
   },
@@ -13924,6 +14320,37 @@
             "created_at": "2026-03-18T20:26:55Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR97529__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR97529__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/storage/unified/search/bleve.go",
+            "line": 137,
+            "body": "P2: Race condition from lock scope reduction: The old code held `cacheMu.Lock()` for the entire `BuildIndex` function, serializing all concurrent calls. The new code only locks the final cache write. This means two concurrent callers (e.g., a Search request and a handleEvent goroutine both hitting `getOrCreateIndex` for the same unindexed key) can simultaneously build an index for the same key. For file-based indexes, the second `bleve.New(dir, mapper)` call will fail because the directory already exists. For memory indexes, duplicate work is done and the first result is silently overwritten.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/storage/unified/search/bleve.go</file_path>\n<line_numbers>137</line_numbers>\n\n<comment>Race condition from lock scope reduction: The old code held `cacheMu.Lock()` for the entire `BuildIndex` function, serializing all concurrent calls. The new code only locks the final cache write. This means two concurrent callers (e.g., a Search request and a handleEvent goroutine both hitting `getOrCreateIndex` for the same unindexed key) can simultaneously build an index for the same key. For file-based indexes, the second `bleve.New(dir, mapper)` call will fail because the directory already exists. For memory indexes, duplicate work is done and the first result is silently overwritten.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/storage/unified/search/bleve.go</file_path>\n<line_numbers>137</line_numbers>\n\n<comment>Race condition from lock scope reduction: The old code held `cacheMu.Lock()` for the entire `BuildIndex` function, serializing all concurrent calls. The new code only locks the final cache write. This means two concurrent callers (e.g., a Search request and a handleEvent goroutine both hitting `getOrCreateIndex` for the same unindexed key) can simultaneously build an index for the same key. For file-based indexes, the second `bleve.New(dir, mapper)` call will fail because the directory already exists. For memory indexes, duplicate work is done and the first result is silently overwritten.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:33Z"
+          },
+          {
+            "path": "pkg/storage/unified/resource/search.go",
+            "line": 176,
+            "body": "P2: _Note: Issue is at line 190, shown here on nearest changed line._\n\nData race on shared `err` variable in concurrent goroutines: The `err` variable from `stats, err := s.storage.GetResourceStats(...)` on line 181 is captured by reference in all `group.Go` closures. Multiple goroutines concurrently write to this same `err` via `_, _, err = s.build(...)` and then read it via `return err`. This is both a data race (detected by `-race`) and a correctness bug \u2014 a goroutine may return another goroutine's error (or nil) instead of its own. Pre-existing but in modified function.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/storage/unified/resource/search.go</file_path>\n<line_numbers>176</line_numbers>\n\n<comment>_Note: Issue is at line 190, shown here on nearest changed line._\n\nData race on shared `err` variable in concurrent goroutines: The `err` variable from `stats, err := s.storage.GetResourceStats(...)` on line 181 is captured by reference in all `group.Go` closures. Multiple goroutines concurrently write to this same `err` via `_, _, err = s.build(...)` and then read it via `return err`. This is both a data race (detected by `-race`) and a correctness bug \u2014 a goroutine may return another goroutine's error (or nil) instead of its own. Pre-existing but in modified function.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/storage/unified/resource/search.go</file_path>\n<line_numbers>176</line_numbers>\n\n<comment>_Note: Issue is at line 190, shown here on nearest changed line._\n\nData race on shared `err` variable in concurrent goroutines: The `err` variable from `stats, err := s.storage.GetResourceStats(...)` on line 181 is captured by reference in all `group.Go` closures. Multiple goroutines concurrently write to this same `err` via `_, _, err = s.build(...)` and then read it via `return err`. This is both a data race (detected by `-race`) and a correctness bug \u2014 a goroutine may return another goroutine's error (or nil) instead of its own. Pre-existing but in modified function.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:33Z"
+          },
+          {
+            "path": "pkg/storage/unified/resource/search.go",
+            "line": 176,
+            "body": "P4: _Note: Issue is at line 189, shown here on nearest changed line._\n\nData race on `totalBatchesIndexed`: This variable is incremented from multiple goroutines launched via `errgroup` without any synchronization. With `initWorkers > 1`, concurrent increments are a data race. Pre-existing but in modified function.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/storage/unified/resource/search.go</file_path>\n<line_numbers>176</line_numbers>\n\n<comment>_Note: Issue is at line 189, shown here on nearest changed line._\n\nData race on `totalBatchesIndexed`: This variable is incremented from multiple goroutines launched via `errgroup` without any synchronization. With `initWorkers > 1`, concurrent increments are a data race. Pre-existing but in modified function.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/storage/unified/resource/search.go</file_path>\n<line_numbers>176</line_numbers>\n\n<comment>_Note: Issue is at line 189, shown here on nearest changed line._\n\nData race on `totalBatchesIndexed`: This variable is incremented from multiple goroutines launched via `errgroup` without any synchronization. With `initWorkers > 1`, concurrent increments are a data race. Pre-existing but in modified function.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:33Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:00:33Z"
+          }
+        ]
       }
     ]
   },
@@ -14930,6 +15357,43 @@
             "created_at": "2026-03-18T20:27:09Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry__tusk__PR80168__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80168__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/incidents/grouptype.py",
+            "line": 11,
+            "body": "P1: MetricAlertDetectorHandler now extends StatefulDetectorHandler (which has abstract methods: counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data) but implements none of them \u2014 it just has `pass`. Python will raise `TypeError: Can't instantiate abstract class MetricAlertDetectorHandler with abstract methods...` when `Detector.detector_handler` (detector.py:86) calls `MetricAlertDetectorHandler(self)`. Previously it extended `DetectorHandler` with a concrete `evaluate()` returning `[]`, which was a safe no-op stub.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/incidents/grouptype.py</file_path>\n<line_numbers>11</line_numbers>\n\n<comment>MetricAlertDetectorHandler now extends StatefulDetectorHandler (which has abstract methods: counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data) but implements none of them \u2014 it just has `pass`. Python will raise `TypeError: Can't instantiate abstract class MetricAlertDetectorHandler with abstract methods...` when `Detector.detector_handler` (detector.py:86) calls `MetricAlertDetectorHandler(self)`. Previously it extended `DetectorHandler` with a concrete `evaluate()` returning `[]`, which was a safe no-op stub.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/incidents/grouptype.py</file_path>\n<line_numbers>11</line_numbers>\n\n<comment>MetricAlertDetectorHandler now extends StatefulDetectorHandler (which has abstract methods: counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data) but implements none of them \u2014 it just has `pass`. Python will raise `TypeError: Can't instantiate abstract class MetricAlertDetectorHandler with abstract methods...` when `Detector.detector_handler` (detector.py:86) calls `MetricAlertDetectorHandler(self)`. Previously it extended `DetectorHandler` with a concrete `evaluate()` returning `[]`, which was a safe no-op stub.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:59:12Z"
+          },
+          {
+            "path": "tests/sentry/workflow_engine/processors/test_detector.py",
+            "line": 537,
+            "body": "P3: In TestEvaluateGroupKeyValue.test_dedupe, the expected occurrence is built with group_key=\"val1\" (line 537) but evaluate_group_key_value is called with group_key=\"group_key\" (line 548). The actual code path calls build_occurrence_and_event_data with group_key=\"group_key\", producing a fingerprint like `[\"<id>:group_key\"]`, while the expected occurrence has fingerprint `[\"<id>:val1\"]`. The test only passes because IssueOccurrence.__eq__ compares solely by the hardcoded `id` field, meaning the test does not actually verify that the correct occurrence (with correct fingerprint/group_key) was produced.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/sentry/workflow_engine/processors/test_detector.py</file_path>\n<line_numbers>537</line_numbers>\n\n<comment>In TestEvaluateGroupKeyValue.test_dedupe, the expected occurrence is built with group_key=\"val1\" (line 537) but evaluate_group_key_value is called with group_key=\"group_key\" (line 548). The actual code path calls build_occurrence_and_event_data with group_key=\"group_key\", producing a fingerprint like `[\"<id>:group_key\"]`, while the expected occurrence has fingerprint `[\"<id>:val1\"]`. The test only passes because IssueOccurrence.__eq__ compares solely by the hardcoded `id` field, meaning the test does not actually verify that the correct occurrence (with correct fingerprint/group_key) was produced.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/sentry/workflow_engine/processors/test_detector.py</file_path>\n<line_numbers>537</line_numbers>\n\n<comment>In TestEvaluateGroupKeyValue.test_dedupe, the expected occurrence is built with group_key=\"val1\" (line 537) but evaluate_group_key_value is called with group_key=\"group_key\" (line 548). The actual code path calls build_occurrence_and_event_data with group_key=\"group_key\", producing a fingerprint like `[\"<id>:group_key\"]`, while the expected occurrence has fingerprint `[\"<id>:val1\"]`. The test only passes because IssueOccurrence.__eq__ compares solely by the hardcoded `id` field, meaning the test does not actually verify that the correct occurrence (with correct fingerprint/group_key) was produced.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:59:12Z"
+          },
+          {
+            "path": "tests/sentry/workflow_engine/processors/test_detector.py",
+            "line": 191,
+            "body": "P4: In test_state_results_multi_group, the expected occurrence for group_2 is built with value=6 (line 191: `build_mock_occurrence_and_event(detector.detector_handler, \"group_2\", 6, PriorityLevel.HIGH)`) but the actual data packet has group_2=10. The real code path passes value=10 to build_occurrence_and_event_data. This passes only because the mock function doesn't vary output based on `value` and IssueOccurrence.__eq__ compares by id. The test gives false confidence that the correct value is being forwarded.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/sentry/workflow_engine/processors/test_detector.py</file_path>\n<line_numbers>191</line_numbers>\n\n<comment>In test_state_results_multi_group, the expected occurrence for group_2 is built with value=6 (line 191: `build_mock_occurrence_and_event(detector.detector_handler, \"group_2\", 6, PriorityLevel.HIGH)`) but the actual data packet has group_2=10. The real code path passes value=10 to build_occurrence_and_event_data. This passes only because the mock function doesn't vary output based on `value` and IssueOccurrence.__eq__ compares by id. The test gives false confidence that the correct value is being forwarded.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/sentry/workflow_engine/processors/test_detector.py</file_path>\n<line_numbers>191</line_numbers>\n\n<comment>In test_state_results_multi_group, the expected occurrence for group_2 is built with value=6 (line 191: `build_mock_occurrence_and_event(detector.detector_handler, \"group_2\", 6, PriorityLevel.HIGH)`) but the actual data packet has group_2=10. The real code path passes value=10 to build_occurrence_and_event_data. This passes only because the mock function doesn't vary output based on `value` and IssueOccurrence.__eq__ compares by id. The test gives false confidence that the correct value is being forwarded.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:59:12Z"
+          },
+          {
+            "path": "tests/sentry/workflow_engine/processors/test_detector.py",
+            "line": 151,
+            "body": "P4: In test_state_results, the expected occurrence is constructed by calling `build_mock_occurrence_and_event(detector.detector_handler, ...)` (line 150-151), but `detector.detector_handler` creates a NEW handler instance each time the property is accessed (detector.py:86). The handler used inside `process_detectors` is a different instance from the one used to build the expected occurrence. If build_mock_occurrence_and_event ever uses handler-instance-specific state (e.g., detector id is fine since it's shared, but any mutable handler state would diverge), the expected and actual would silently differ. Currently this works by coincidence since the mock only uses `handler.detector` and `handler.build_fingerprint`, but it's fragile.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/sentry/workflow_engine/processors/test_detector.py</file_path>\n<line_numbers>151</line_numbers>\n\n<comment>In test_state_results, the expected occurrence is constructed by calling `build_mock_occurrence_and_event(detector.detector_handler, ...)` (line 150-151), but `detector.detector_handler` creates a NEW handler instance each time the property is accessed (detector.py:86). The handler used inside `process_detectors` is a different instance from the one used to build the expected occurrence. If build_mock_occurrence_and_event ever uses handler-instance-specific state (e.g., detector id is fine since it's shared, but any mutable handler state would diverge), the expected and actual would silently differ. Currently this works by coincidence since the mock only uses `handler.detector` and `handler.build_fingerprint`, but it's fragile.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/sentry/workflow_engine/processors/test_detector.py</file_path>\n<line_numbers>151</line_numbers>\n\n<comment>In test_state_results, the expected occurrence is constructed by calling `build_mock_occurrence_and_event(detector.detector_handler, ...)` (line 150-151), but `detector.detector_handler` creates a NEW handler instance each time the property is accessed (detector.py:86). The handler used inside `process_detectors` is a different instance from the one used to build the expected occurrence. If build_mock_occurrence_and_event ever uses handler-instance-specific state (e.g., detector id is fine since it's shared, but any mutable handler state would diverge), the expected and actual would silently differ. Currently this works by coincidence since the mock only uses `handler.detector` and `handler.build_fingerprint`, but it's fragile.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:59:12Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T01:59:11Z"
+          }
+        ]
       }
     ]
   },
@@ -15853,6 +16317,25 @@
             "created_at": "2026-03-18T20:27:17Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry__tusk__PR80528__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80528__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/monitors/logic/incident_occurrence.py",
+            "line": 168,
+            "body": "P3: In `get_monitor_environment_context`, a copy of `config` is made on line 160 and the `schedule_type` is overridden to a human-readable display value on line 162. However, line 168 returns `monitor_environment.monitor.config` (the original, unmodified config) instead of the local `config` variable. This means the `schedule_type` override is silently discarded and the monitor context always contains the raw `schedule_type` value. This is a pre-existing bug carried over during the refactor.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/monitors/logic/incident_occurrence.py</file_path>\n<line_numbers>168</line_numbers>\n\n<comment>In `get_monitor_environment_context`, a copy of `config` is made on line 160 and the `schedule_type` is overridden to a human-readable display value on line 162. However, line 168 returns `monitor_environment.monitor.config` (the original, unmodified config) instead of the local `config` variable. This means the `schedule_type` override is silently discarded and the monitor context always contains the raw `schedule_type` value. This is a pre-existing bug carried over during the refactor.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/monitors/logic/incident_occurrence.py</file_path>\n<line_numbers>168</line_numbers>\n\n<comment>In `get_monitor_environment_context`, a copy of `config` is made on line 160 and the `schedule_type` is overridden to a human-readable display value on line 162. However, line 168 returns `monitor_environment.monitor.config` (the original, unmodified config) instead of the local `config` variable. This means the `schedule_type` override is silently discarded and the monitor context always contains the raw `schedule_type` value. This is a pre-existing bug carried over during the refactor.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:56:26Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 1 issue",
+            "created_at": "2026-04-12T01:56:26Z"
+          }
+        ]
       }
     ]
   },
@@ -16795,6 +17278,43 @@
             "created_at": "2026-03-18T20:27:23Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry__tusk__PR77754__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR77754__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/integrations/services/assignment_source.py",
+            "line": 18,
+            "body": "P2: Classic Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition (module import) time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` (which doesn't pass `queued`) will share the same timestamp \u2014 the time the module was first imported. The fix is `queued: datetime = field(default_factory=timezone.now)`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/services/assignment_source.py</file_path>\n<line_numbers>18</line_numbers>\n\n<comment>Classic Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition (module import) time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` (which doesn't pass `queued`) will share the same timestamp \u2014 the time the module was first imported. The fix is `queued: datetime = field(default_factory=timezone.now)`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/services/assignment_source.py</file_path>\n<line_numbers>18</line_numbers>\n\n<comment>Classic Python dataclass default-value bug: `timezone.now()` is evaluated once at class definition (module import) time, not at instance creation time. Every `AssignmentSource` created via `from_integration()` (which doesn't pass `queued`) will share the same timestamp \u2014 the time the module was first imported. The fix is `queued: datetime = field(default_factory=timezone.now)`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:09Z"
+          },
+          {
+            "path": "src/sentry/integrations/mixins/issues.py",
+            "line": 416,
+            "body": "P3: The `assignment_source` parameter was added to the abstract `sync_status_outbound` method, but this PR is about assignee sync cycles \u2014 status sync is unrelated. Furthermore, `ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` (line 161) does not accept `**kwargs`, so any future caller passing `assignment_source` to it will get a TypeError. The other concrete implementations (jira, jira_server) use `**kwargs` so they silently absorb it, but ExampleIntegration will crash.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/mixins/issues.py</file_path>\n<line_numbers>416</line_numbers>\n\n<comment>The `assignment_source` parameter was added to the abstract `sync_status_outbound` method, but this PR is about assignee sync cycles \u2014 status sync is unrelated. Furthermore, `ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` (line 161) does not accept `**kwargs`, so any future caller passing `assignment_source` to it will get a TypeError. The other concrete implementations (jira, jira_server) use `**kwargs` so they silently absorb it, but ExampleIntegration will crash.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/mixins/issues.py</file_path>\n<line_numbers>416</line_numbers>\n\n<comment>The `assignment_source` parameter was added to the abstract `sync_status_outbound` method, but this PR is about assignee sync cycles \u2014 status sync is unrelated. Furthermore, `ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` (line 161) does not accept `**kwargs`, so any future caller passing `assignment_source` to it will get a TypeError. The other concrete implementations (jira, jira_server) use `**kwargs` so they silently absorb it, but ExampleIntegration will crash.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:09Z"
+          },
+          {
+            "path": "tests/sentry/integrations/services/test_assignment_source.py",
+            "line": 8,
+            "body": "P4: Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array. The name is misleading and suggests the wrong data structure is being tested.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/sentry/integrations/services/test_assignment_source.py</file_path>\n<line_numbers>8</line_numbers>\n\n<comment>Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array. The name is misleading and suggests the wrong data structure is being tested.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/sentry/integrations/services/test_assignment_source.py</file_path>\n<line_numbers>8</line_numbers>\n\n<comment>Test method is named `test_from_dict_empty_array` but it tests an empty dict (`{}`), not an empty array. The name is misleading and suggests the wrong data structure is being tested.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:09Z"
+          },
+          {
+            "path": "tests/sentry/integrations/services/test_assignment_source.py",
+            "line": 13,
+            "body": "P4: Test method name `test_from_dict_inalid_data` has a typo \u2014 missing 'v' in 'invalid'. While the test still runs (Python doesn't validate spelling), this makes it harder to find via search and indicates insufficient review.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/sentry/integrations/services/test_assignment_source.py</file_path>\n<line_numbers>13</line_numbers>\n\n<comment>Test method name `test_from_dict_inalid_data` has a typo \u2014 missing 'v' in 'invalid'. While the test still runs (Python doesn't validate spelling), this makes it harder to find via search and indicates insufficient review.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/sentry/integrations/services/test_assignment_source.py</file_path>\n<line_numbers>13</line_numbers>\n\n<comment>Test method name `test_from_dict_inalid_data` has a typo \u2014 missing 'v' in 'invalid'. While the test still runs (Python doesn't validate spelling), this makes it harder to find via search and indicates insufficient review.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:09Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T01:55:09Z"
+          }
+        ]
       }
     ]
   },
@@ -17835,6 +18355,43 @@
             "created_at": "2026-03-18T20:27:00Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry__tusk__PR95633__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR95633__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/remote_subscriptions/consumers/queue_consumer.py",
+            "line": 335,
+            "body": "P3: close() joins the commit thread BEFORE shutting down the worker pool. After shutdown_event is set, the commit thread does one final commit and exits. Then queue_pool.shutdown() drains remaining items from worker queues (queue.shutdown(immediate=False) lets workers process remaining items). These items are processed (side effects occur) but the commit thread is already dead, so their offsets are never committed. On restart, these items will be re-consumed and processed again, causing duplicate side effects. The fix is to drain workers first, then do a final commit, then stop the commit thread.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/remote_subscriptions/consumers/queue_consumer.py</file_path>\n<line_numbers>335</line_numbers>\n\n<comment>close() joins the commit thread BEFORE shutting down the worker pool. After shutdown_event is set, the commit thread does one final commit and exits. Then queue_pool.shutdown() drains remaining items from worker queues (queue.shutdown(immediate=False) lets workers process remaining items). These items are processed (side effects occur) but the commit thread is already dead, so their offsets are never committed. On restart, these items will be re-consumed and processed again, causing duplicate side effects. The fix is to drain workers first, then do a final commit, then stop the commit thread.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/remote_subscriptions/consumers/queue_consumer.py</file_path>\n<line_numbers>335</line_numbers>\n\n<comment>close() joins the commit thread BEFORE shutting down the worker pool. After shutdown_event is set, the commit thread does one final commit and exits. Then queue_pool.shutdown() drains remaining items from worker queues (queue.shutdown(immediate=False) lets workers process remaining items). These items are processed (side effects occur) but the commit thread is already dead, so their offsets are never committed. On restart, these items will be re-consumed and processed again, causing duplicate side effects. The fix is to drain workers first, then do a final commit, then stop the commit thread.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:58Z"
+          },
+          {
+            "path": "tests/sentry/uptime/consumers/test_results_consumer.py",
+            "line": 1922,
+            "body": "P3: test_thread_queue_parallel_error_handling's docstring says 'Test that errors in processing don't block offset commits for other messages' but its assertion checks the OPPOSITE: `assert len(committed_offsets) == 0 or test_partition not in committed_offsets` (asserts no commits happened). Furthermore, the code always marks offsets as complete via the `finally` block in OrderedQueueWorker.run() (line 149), so offsets WILL be committed for both the failed and successful message. The test only passes due to timing: the commit loop waits 1 second between iterations and the test only waits 0.2s after processing completes, so the commit hasn't fired yet. This test is inherently flaky.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>tests/sentry/uptime/consumers/test_results_consumer.py</file_path>\n<line_numbers>1922</line_numbers>\n\n<comment>test_thread_queue_parallel_error_handling's docstring says 'Test that errors in processing don't block offset commits for other messages' but its assertion checks the OPPOSITE: `assert len(committed_offsets) == 0 or test_partition not in committed_offsets` (asserts no commits happened). Furthermore, the code always marks offsets as complete via the `finally` block in OrderedQueueWorker.run() (line 149), so offsets WILL be committed for both the failed and successful message. The test only passes due to timing: the commit loop waits 1 second between iterations and the test only waits 0.2s after processing completes, so the commit hasn't fired yet. This test is inherently flaky.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>tests/sentry/uptime/consumers/test_results_consumer.py</file_path>\n<line_numbers>1922</line_numbers>\n\n<comment>test_thread_queue_parallel_error_handling's docstring says 'Test that errors in processing don't block offset commits for other messages' but its assertion checks the OPPOSITE: `assert len(committed_offsets) == 0 or test_partition not in committed_offsets` (asserts no commits happened). Furthermore, the code always marks offsets as complete via the `finally` block in OrderedQueueWorker.run() (line 149), so offsets WILL be committed for both the failed and successful message. The test only passes due to timing: the commit loop waits 1 second between iterations and the test only waits 0.2s after processing completes, so the commit hasn't fired yet. This test is inherently flaky.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:58Z"
+          },
+          {
+            "path": "src/sentry/remote_subscriptions/consumers/queue_consumer.py",
+            "line": 185,
+            "body": "P3: The SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' and 'Only commits offsets after successful processing', but neither is true. The queues are created as `queue.Queue()` without a maxsize parameter (line 185), so they are unbounded and never 'fill up' - there is no backpressure mechanism. If workers are slow and the producer is fast, memory will grow without bound. Additionally, the finally block in OrderedQueueWorker.run() always calls complete_offset regardless of whether processing succeeded or raised an exception.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/remote_subscriptions/consumers/queue_consumer.py</file_path>\n<line_numbers>185</line_numbers>\n\n<comment>The SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' and 'Only commits offsets after successful processing', but neither is true. The queues are created as `queue.Queue()` without a maxsize parameter (line 185), so they are unbounded and never 'fill up' - there is no backpressure mechanism. If workers are slow and the producer is fast, memory will grow without bound. Additionally, the finally block in OrderedQueueWorker.run() always calls complete_offset regardless of whether processing succeeded or raised an exception.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/remote_subscriptions/consumers/queue_consumer.py</file_path>\n<line_numbers>185</line_numbers>\n\n<comment>The SimpleQueueProcessingStrategy docstring claims 'Natural backpressure when queues fill up' and 'Only commits offsets after successful processing', but neither is true. The queues are created as `queue.Queue()` without a maxsize parameter (line 185), so they are unbounded and never 'fill up' - there is no backpressure mechanism. If workers are slow and the producer is fast, memory will grow without bound. Additionally, the finally block in OrderedQueueWorker.run() always calls complete_offset regardless of whether processing succeeded or raised an exception.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:58Z"
+          },
+          {
+            "path": "src/sentry/remote_subscriptions/consumers/queue_consumer.py",
+            "line": 49,
+            "body": "P4: OffsetTracker._get_partition_lock and all methods that use it (add_offset, complete_offset, get_committable_offsets) are accessed concurrently from the main thread (submit), worker threads (complete_offset), and the commit thread (get_committable_offsets). The _get_partition_lock method accesses self.partition_locks without synchronization - two threads calling it with the same new partition simultaneously will both create a Lock() but setdefault ensures only one is stored. This is safe under CPython's GIL but not guaranteed by the Python language spec. More critically, get_committable_offsets iterates `list(self.all_offsets.keys())` without holding any lock, while worker threads may be modifying all_offsets via defaultdict auto-creation. This relies on CPython GIL atomicity guarantees that are not part of the language specification.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/remote_subscriptions/consumers/queue_consumer.py</file_path>\n<line_numbers>49</line_numbers>\n\n<comment>OffsetTracker._get_partition_lock and all methods that use it (add_offset, complete_offset, get_committable_offsets) are accessed concurrently from the main thread (submit), worker threads (complete_offset), and the commit thread (get_committable_offsets). The _get_partition_lock method accesses self.partition_locks without synchronization - two threads calling it with the same new partition simultaneously will both create a Lock() but setdefault ensures only one is stored. This is safe under CPython's GIL but not guaranteed by the Python language spec. More critically, get_committable_offsets iterates `list(self.all_offsets.keys())` without holding any lock, while worker threads may be modifying all_offsets via defaultdict auto-creation. This relies on CPython GIL atomicity guarantees that are not part of the language specification.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/remote_subscriptions/consumers/queue_consumer.py</file_path>\n<line_numbers>49</line_numbers>\n\n<comment>OffsetTracker._get_partition_lock and all methods that use it (add_offset, complete_offset, get_committable_offsets) are accessed concurrently from the main thread (submit), worker threads (complete_offset), and the commit thread (get_committable_offsets). The _get_partition_lock method accesses self.partition_locks without synchronization - two threads calling it with the same new partition simultaneously will both create a Lock() but setdefault ensures only one is stored. This is safe under CPython's GIL but not guaranteed by the Python language spec. More critically, get_committable_offsets iterates `list(self.all_offsets.keys())` without holding any lock, while worker threads may be modifying all_offsets via defaultdict auto-creation. This relies on CPython GIL atomicity guarantees that are not part of the language specification.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:58Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:03:58Z"
+          }
+        ]
       }
     ]
   },
@@ -19066,6 +19623,37 @@
             "created_at": "2026-03-18T20:27:44Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry-greptile__tusk__PR2__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR2__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/api/paginator.py",
+            "line": 840,
+            "body": "P1: OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. This paginator is used with order_by=\"-datetime\" on AuditLogEntry, so getattr(item, self.key) returns a datetime. Compare to DateTimePaginator.get_item_key which correctly converts to float first via `float(value.strftime(\"%s.%f\")) * self.multiplier` before applying math.floor/ceil.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>840</line_numbers>\n\n<comment>OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. This paginator is used with order_by=\"-datetime\" on AuditLogEntry, so getattr(item, self.key) returns a datetime. Compare to DateTimePaginator.get_item_key which correctly converts to float first via `float(value.strftime(\"%s.%f\")) * self.multiplier` before applying math.floor/ceil.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>840</line_numbers>\n\n<comment>OptimizedCursorPaginator.get_item_key calls math.floor()/math.ceil() directly on a datetime object, which will raise TypeError. This paginator is used with order_by=\"-datetime\" on AuditLogEntry, so getattr(item, self.key) returns a datetime. Compare to DateTimePaginator.get_item_key which correctly converts to float first via `float(value.strftime(\"%s.%f\")) * self.multiplier` before applying math.floor/ceil.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:45:42Z"
+          },
+          {
+            "path": "src/sentry/api/paginator.py",
+            "line": 882,
+            "body": "P2: The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing \u2014 it raises AssertionError(\"Negative indexing is not supported.\"). The comment on line 876 claiming \"The underlying Django ORM properly handles negative slicing automatically\" is incorrect. Any request to the audit logs endpoint with optimized_pagination=true and a crafted cursor with a negative offset will crash with a 500 error.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>882</line_numbers>\n\n<comment>The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing \u2014 it raises AssertionError(\"Negative indexing is not supported.\"). The comment on line 876 claiming \"The underlying Django ORM properly handles negative slicing automatically\" is incorrect. Any request to the audit logs endpoint with optimized_pagination=true and a crafted cursor with a negative offset will crash with a 500 error.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/paginator.py</file_path>\n<line_numbers>882</line_numbers>\n\n<comment>The 'advanced features' negative offset path does queryset[negative_offset:stop], but Django ORM does not support negative indexing \u2014 it raises AssertionError(\"Negative indexing is not supported.\"). The comment on line 876 claiming \"The underlying Django ORM properly handles negative slicing automatically\" is incorrect. Any request to the audit logs endpoint with optimized_pagination=true and a crafted cursor with a negative offset will crash with a 500 error.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:45:42Z"
+          },
+          {
+            "path": "src/sentry/spans/buffer.py",
+            "line": 449,
+            "body": "P3: The old code had a safety check that skipped segments exceeding max_segment_spans (default 1001) during flush, with metrics and error logging. This check was completely removed. While the Lua script now caps spans at 1000 during insert, the flush-side guard protected against edge cases (e.g., race conditions between concurrent inserts, or spans inserted before the eviction logic was deployed). The max_segment_spans attribute is still initialized on the class but is now dead code.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/spans/buffer.py</file_path>\n<line_numbers>449</line_numbers>\n\n<comment>The old code had a safety check that skipped segments exceeding max_segment_spans (default 1001) during flush, with metrics and error logging. This check was completely removed. While the Lua script now caps spans at 1000 during insert, the flush-side guard protected against edge cases (e.g., race conditions between concurrent inserts, or spans inserted before the eviction logic was deployed). The max_segment_spans attribute is still initialized on the class but is now dead code.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/spans/buffer.py</file_path>\n<line_numbers>449</line_numbers>\n\n<comment>The old code had a safety check that skipped segments exceeding max_segment_spans (default 1001) during flush, with metrics and error logging. This check was completely removed. While the Lua script now caps spans at 1000 during insert, the flush-side guard protected against edge cases (e.g., race conditions between concurrent inserts, or spans inserted before the eviction logic was deployed). The max_segment_spans attribute is still initialized on the class but is now dead code.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:45:43Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T01:45:42Z"
+          }
+        ]
       }
     ]
   },
@@ -20334,6 +20922,49 @@
             "created_at": "2026-03-18T20:26:39Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry-greptile__tusk__PR3__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR3__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/api/helpers/error_upsampling.py",
+            "line": 27,
+            "body": "P2: Python's built-in `hash()` is randomized per-process (via PYTHONHASHSEED). Since the cache backend is shared Redis/Memcached, different Sentry worker processes will compute different hash values for the same project IDs, causing every cross-process cache lookup to miss. This completely defeats the caching optimization. Process A writes under key `...:{hash_A}`, but process B looks up `...:{hash_B}` and gets a miss.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>27</line_numbers>\n\n<comment>Python's built-in `hash()` is randomized per-process (via PYTHONHASHSEED). Since the cache backend is shared Redis/Memcached, different Sentry worker processes will compute different hash values for the same project IDs, causing every cross-process cache lookup to miss. This completely defeats the caching optimization. Process A writes under key `...:{hash_A}`, but process B looks up `...:{hash_B}` and gets a miss.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>27</line_numbers>\n\n<comment>Python's built-in `hash()` is randomized per-process (via PYTHONHASHSEED). Since the cache backend is shared Redis/Memcached, different Sentry worker processes will compute different hash values for the same project IDs, causing every cross-process cache lookup to miss. This completely defeats the caching optimization. Process A writes under key `...:{hash_A}`, but process B looks up `...:{hash_B}` and gets a miss.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:56:21Z"
+          },
+          {
+            "path": "src/sentry/api/helpers/error_upsampling.py",
+            "line": 73,
+            "body": "P2: The `invalidate_upsampling_cache` function uses the same non-deterministic `hash()` in its cache key construction. Even if a caller existed, invalidation from one process would not clear the cache entry written by a different process, since hash values differ across processes.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>73</line_numbers>\n\n<comment>The `invalidate_upsampling_cache` function uses the same non-deterministic `hash()` in its cache key construction. Even if a caller existed, invalidation from one process would not clear the cache entry written by a different process, since hash values differ across processes.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>73</line_numbers>\n\n<comment>The `invalidate_upsampling_cache` function uses the same non-deterministic `hash()` in its cache key construction. Even if a caller existed, invalidation from one process would not clear the cache entry written by a different process, since hash values differ across processes.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:56:21Z"
+          },
+          {
+            "path": "src/sentry/api/helpers/error_upsampling.py",
+            "line": 27,
+            "body": "P3: The caching logic caches a boolean `False` value. When `cache.get(cache_key)` returns `False` (a cached negative result), the check `if cached_result is not None` passes and `cached_result and _should_apply_sample_weight_transform(...)` short-circuits to `False` without calling the second function \u2014 this is correct. However, if the cache stores `True` but a later call changes `dataset` or `request` (different query context), the cached `True` combined with a fresh `_should_apply_sample_weight_transform` call could return inconsistent results vs. what was cached. This is minor but worth noting: the cache key doesn't include dataset or request context, so the same cache entry is reused across different query types for the same org+projects.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>27</line_numbers>\n\n<comment>The caching logic caches a boolean `False` value. When `cache.get(cache_key)` returns `False` (a cached negative result), the check `if cached_result is not None` passes and `cached_result and _should_apply_sample_weight_transform(...)` short-circuits to `False` without calling the second function \u2014 this is correct. However, if the cache stores `True` but a later call changes `dataset` or `request` (different query context), the cached `True` combined with a fresh `_should_apply_sample_weight_transform` call could return inconsistent results vs. what was cached. This is minor but worth noting: the cache key doesn't include dataset or request context, so the same cache entry is reused across different query types for the same org+projects.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>27</line_numbers>\n\n<comment>The caching logic caches a boolean `False` value. When `cache.get(cache_key)` returns `False` (a cached negative result), the check `if cached_result is not None` passes and `cached_result and _should_apply_sample_weight_transform(...)` short-circuits to `False` without calling the second function \u2014 this is correct. However, if the cache stores `True` but a later call changes `dataset` or `request` (different query context), the cached `True` combined with a fresh `_should_apply_sample_weight_transform` call could return inconsistent results vs. what was cached. This is minor but worth noting: the cache key doesn't include dataset or request context, so the same cache entry is reused across different query types for the same org+projects.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:56:21Z"
+          },
+          {
+            "path": "src/sentry/api/helpers/error_upsampling.py",
+            "line": 63,
+            "body": "P3: The behavior was changed from `any(project_id in allowlist ...)` to `all(project_id in allowlist ...)` and the docstring was updated to match ('ANY' -> 'ALL'). This is a significant semantic change: previously, if *any* project was allowlisted, upsampling was applied; now *all* projects must be allowlisted. For a query spanning both allowlisted and non-allowlisted projects, `count()` will no longer be transformed to `upsampled_count()`, meaning the returned counts will be raw sample counts rather than upsampled estimates. This could silently return incorrect (underestimated) data for mixed-project queries where some projects are allowlisted.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>63</line_numbers>\n\n<comment>The behavior was changed from `any(project_id in allowlist ...)` to `all(project_id in allowlist ...)` and the docstring was updated to match ('ANY' -> 'ALL'). This is a significant semantic change: previously, if *any* project was allowlisted, upsampling was applied; now *all* projects must be allowlisted. For a query spanning both allowlisted and non-allowlisted projects, `count()` will no longer be transformed to `upsampled_count()`, meaning the returned counts will be raw sample counts rather than upsampled estimates. This could silently return incorrect (underestimated) data for mixed-project queries where some projects are allowlisted.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/helpers/error_upsampling.py</file_path>\n<line_numbers>63</line_numbers>\n\n<comment>The behavior was changed from `any(project_id in allowlist ...)` to `all(project_id in allowlist ...)` and the docstring was updated to match ('ANY' -> 'ALL'). This is a significant semantic change: previously, if *any* project was allowlisted, upsampling was applied; now *all* projects must be allowlisted. For a query spanning both allowlisted and non-allowlisted projects, `count()` will no longer be transformed to `upsampled_count()`, meaning the returned counts will be raw sample counts rather than upsampled estimates. This could silently return incorrect (underestimated) data for mixed-project queries where some projects are allowlisted.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:56:21Z"
+          },
+          {
+            "path": "src/sentry/api/endpoints/organization_events_stats.py",
+            "line": 229,
+            "body": "P3: In the `top_events > 0` branch of `organization_events_stats.py`, the old code validated that 'timestamp' was not in `raw_groupby` and raised `ParseError` if so. This validation was removed during the refactoring (the `ParseError` import was also removed). This means users can now group by 'timestamp' which was previously explicitly disallowed, potentially causing unexpected query behavior or errors deeper in the stack.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/api/endpoints/organization_events_stats.py</file_path>\n<line_numbers>229</line_numbers>\n\n<comment>In the `top_events > 0` branch of `organization_events_stats.py`, the old code validated that 'timestamp' was not in `raw_groupby` and raised `ParseError` if so. This validation was removed during the refactoring (the `ParseError` import was also removed). This means users can now group by 'timestamp' which was previously explicitly disallowed, potentially causing unexpected query behavior or errors deeper in the stack.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/api/endpoints/organization_events_stats.py</file_path>\n<line_numbers>229</line_numbers>\n\n<comment>In the `top_events > 0` branch of `organization_events_stats.py`, the old code validated that 'timestamp' was not in `raw_groupby` and raised `ParseError` if so. This validation was removed during the refactoring (the `ParseError` import was also removed). This means users can now group by 'timestamp' which was previously explicitly disallowed, potentially causing unexpected query behavior or errors deeper in the stack.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:56:21Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T01:56:21Z"
+          }
+        ]
       }
     ]
   },
@@ -21299,6 +21930,43 @@
             "created_at": "2026-03-18T20:26:53Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR103633__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR103633__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/services/authz/rbac/cache.go",
+            "line": 31,
+            "body": "P3: Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/authz/rbac/cache.go</file_path>\n<line_numbers>31</line_numbers>\n\n<comment>Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/authz/rbac/cache.go</file_path>\n<line_numbers>31</line_numbers>\n\n<comment>Cache key collision in `userPermDenialCacheKey`: the key joins `name` and `parent` with `_`, but both fields can contain underscores (dashboard/folder UIDs). For example, `userPermDenialCacheKey(ns, uid, act, \"a_b\", \"c\")` and `userPermDenialCacheKey(ns, uid, act, \"a\", \"b_c\")` both produce `ns.perm_uid_act_a_b_c`. This means a denial cached for one resource could incorrectly deny access to a completely different resource.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:52Z"
+          },
+          {
+            "path": "pkg/services/authz/rbac/service_test.go",
+            "line": 982,
+            "body": "P3: Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. Line 981 comments \"Allow access to the dashboard to prove this is not checked\" but line 982 sets the permission cache value to `false` (`map[string]bool{\"dashboards:uid:dash1\": false}`). Since `checkPermission` checks `scopeMap[t.scope(req.Name)]` and `false` is falsy, the permission cache would ALSO deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/authz/rbac/service_test.go</file_path>\n<line_numbers>982</line_numbers>\n\n<comment>Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. Line 981 comments \"Allow access to the dashboard to prove this is not checked\" but line 982 sets the permission cache value to `false` (`map[string]bool{\"dashboards:uid:dash1\": false}`). Since `checkPermission` checks `scopeMap[t.scope(req.Name)]` and `false` is falsy, the permission cache would ALSO deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/authz/rbac/service_test.go</file_path>\n<line_numbers>982</line_numbers>\n\n<comment>Test \"Should deny on explicit cache deny entry\" doesn't verify what it claims. Line 981 comments \"Allow access to the dashboard to prove this is not checked\" but line 982 sets the permission cache value to `false` (`map[string]bool{\"dashboards:uid:dash1\": false}`). Since `checkPermission` checks `scopeMap[t.scope(req.Name)]` and `false` is falsy, the permission cache would ALSO deny access. The test passes regardless of whether the denial cache is consulted, so it doesn't prove the denial cache takes priority over the permission cache.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:52Z"
+          },
+          {
+            "path": "pkg/services/authz/rbac/cache.go",
+            "line": 30,
+            "body": "P4: The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance (both passed the same `cache` in `NewService`) but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix `ns.perm_UID_ACTION_NAME_PARENT` overlaps with the permission cache key prefix `ns.perm_UID_ACTION`. If a denial key accidentally matches a permission key (e.g., if a user UID contains underscore-delimited segments resembling `UID_ACTION_NAME_PARENT`), `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/authz/rbac/cache.go</file_path>\n<line_numbers>30</line_numbers>\n\n<comment>The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance (both passed the same `cache` in `NewService`) but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix `ns.perm_UID_ACTION_NAME_PARENT` overlaps with the permission cache key prefix `ns.perm_UID_ACTION`. If a denial key accidentally matches a permission key (e.g., if a user UID contains underscore-delimited segments resembling `UID_ACTION_NAME_PARENT`), `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/authz/rbac/cache.go</file_path>\n<line_numbers>30</line_numbers>\n\n<comment>The `permDenialCache` and `permCache` share the same underlying `cache.Cache` instance (both passed the same `cache` in `NewService`) but store different Go types (`bool` vs `map[string]bool`). The denial cache key prefix `ns.perm_UID_ACTION_NAME_PARENT` overlaps with the permission cache key prefix `ns.perm_UID_ACTION`. If a denial key accidentally matches a permission key (e.g., if a user UID contains underscore-delimited segments resembling `UID_ACTION_NAME_PARENT`), `json.Unmarshal` will fail when the wrong cacheWrap tries to deserialize the other type's data, causing spurious cache misses and warning logs.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:53Z"
+          },
+          {
+            "path": "pkg/services/authz/rbac/service.go",
+            "line": 137,
+            "body": "P4: After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"` (line 137), suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates two semantically different situations (cache miss vs. cache hit with stale data) into the same metric label, making it difficult to distinguish genuine cache misses from outdated-cache fallbacks when monitoring.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/authz/rbac/service.go</file_path>\n<line_numbers>137</line_numbers>\n\n<comment>After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"` (line 137), suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates two semantically different situations (cache miss vs. cache hit with stale data) into the same metric label, making it difficult to distinguish genuine cache misses from outdated-cache fallbacks when monitoring.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/authz/rbac/service.go</file_path>\n<line_numbers>137</line_numbers>\n\n<comment>After the permission cache is hit but the specific resource is not found (allowed=false), the `permissionCacheUsage` metric is incremented with label `\"false\"` (line 137), suggesting a cache miss. But the cache WAS hit \u2014 the resource simply wasn't in the cached permission set. This conflates two semantically different situations (cache miss vs. cache hit with stale data) into the same metric label, making it difficult to distinguish genuine cache misses from outdated-cache fallbacks when monitoring.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:53Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:07:52Z"
+          }
+        ]
       }
     ]
   },
@@ -22417,6 +23085,49 @@
             "created_at": "2026-03-18T20:26:52Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "sentry__sentry__tusk__PR67876__20260410",
+        "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR67876__20260410/pull/1",
+        "review_comments": [
+          {
+            "path": "src/sentry/integrations/github/integration.py",
+            "line": 402,
+            "body": "P1: The OAuth `state` parameter used to prevent CSRF is `pipeline.signature`, which is a deterministic MD5 hash of the pipeline view class names. This value is identical for every user and every session \u2014 it's essentially a static string (e.g., `9cae5e88803f35ed7970fc131e6e65d3` as seen in tests). An attacker who knows the pipeline configuration (open-source code) can compute this value and forge the `state` parameter, defeating the CSRF protection that the `state` parameter is supposed to provide in the OAuth2 flow. OAuth `state` should include a cryptographic random nonce tied to the user's session.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>402</line_numbers>\n\n<comment>The OAuth `state` parameter used to prevent CSRF is `pipeline.signature`, which is a deterministic MD5 hash of the pipeline view class names. This value is identical for every user and every session \u2014 it's essentially a static string (e.g., `9cae5e88803f35ed7970fc131e6e65d3` as seen in tests). An attacker who knows the pipeline configuration (open-source code) can compute this value and forge the `state` parameter, defeating the CSRF protection that the `state` parameter is supposed to provide in the OAuth2 flow. OAuth `state` should include a cryptographic random nonce tied to the user's session.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>402</line_numbers>\n\n<comment>The OAuth `state` parameter used to prevent CSRF is `pipeline.signature`, which is a deterministic MD5 hash of the pipeline view class names. This value is identical for every user and every session \u2014 it's essentially a static string (e.g., `9cae5e88803f35ed7970fc131e6e65d3` as seen in tests). An attacker who knows the pipeline configuration (open-source code) can compute this value and forge the `state` parameter, defeating the CSRF protection that the `state` parameter is supposed to provide in the OAuth2 flow. OAuth `state` should include a cryptographic random nonce tied to the user's session.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:01Z"
+          },
+          {
+            "path": "src/sentry/integrations/github/integration.py",
+            "line": 503,
+            "body": "P2: Accessing `integration.metadata[\"sender\"][\"login\"]` will raise a `KeyError` if the `sender` key is not present in the integration's metadata. The `sender` field is only set when the integration is created via a GitHub webhook installation event (in `webhook.py`). Integrations created before this code change, or through other paths, will not have a `sender` key in metadata. This will cause a 500 error instead of the intended user-facing error page.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>503</line_numbers>\n\n<comment>Accessing `integration.metadata[\"sender\"][\"login\"]` will raise a `KeyError` if the `sender` key is not present in the integration's metadata. The `sender` field is only set when the integration is created via a GitHub webhook installation event (in `webhook.py`). Integrations created before this code change, or through other paths, will not have a `sender` key in metadata. This will cause a 500 error instead of the intended user-facing error page.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>503</line_numbers>\n\n<comment>Accessing `integration.metadata[\"sender\"][\"login\"]` will raise a `KeyError` if the `sender` key is not present in the integration's metadata. The `sender` field is only set when the integration is created via a GitHub webhook installation event (in `webhook.py`). Integrations created before this code change, or through other paths, will not have a `sender` key in metadata. This will cause a 500 error instead of the intended user-facing error page.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:01Z"
+          },
+          {
+            "path": "src/sentry/integrations/github/integration.py",
+            "line": 482,
+            "body": "P1: When the `Integration` does not exist yet (line 481-482, `Integration.DoesNotExist`), the code calls `pipeline.next_step()` without performing the authenticated user check. This means the first installation of a GitHub app completely bypasses the sender/user validation that this PR is supposed to add. An attacker could install a GitHub app on a victim's Sentry org without being the GitHub user who initiated the installation, as long as the Integration record hasn't been created yet by the webhook.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>482</line_numbers>\n\n<comment>When the `Integration` does not exist yet (line 481-482, `Integration.DoesNotExist`), the code calls `pipeline.next_step()` without performing the authenticated user check. This means the first installation of a GitHub app completely bypasses the sender/user validation that this PR is supposed to add. An attacker could install a GitHub app on a victim's Sentry org without being the GitHub user who initiated the installation, as long as the Integration record hasn't been created yet by the webhook.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>482</line_numbers>\n\n<comment>When the `Integration` does not exist yet (line 481-482, `Integration.DoesNotExist`), the code calls `pipeline.next_step()` without performing the authenticated user check. This means the first installation of a GitHub app completely bypasses the sender/user validation that this PR is supposed to add. An attacker could install a GitHub app on a victim's Sentry org without being the GitHub user who initiated the installation, as long as the Integration record hasn't been created yet by the webhook.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:01Z"
+          },
+          {
+            "path": "src/sentry/integrations/github/integration.py",
+            "line": 416,
+            "body": "P3: The `redirect_uri` is included in the authorization request (line 408) but is missing from the token exchange request (line 416-419). Per the OAuth2 spec (RFC 6749, Section 4.1.3), if a `redirect_uri` was included in the authorization request, the same `redirect_uri` MUST be included in the access token request. Some OAuth providers (including GitHub in certain configurations) may reject the token exchange or behave unexpectedly when the redirect_uri is omitted from the token request but was present in the authorization request.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>416</line_numbers>\n\n<comment>The `redirect_uri` is included in the authorization request (line 408) but is missing from the token exchange request (line 416-419). Per the OAuth2 spec (RFC 6749, Section 4.1.3), if a `redirect_uri` was included in the authorization request, the same `redirect_uri` MUST be included in the access token request. Some OAuth providers (including GitHub in certain configurations) may reject the token exchange or behave unexpectedly when the redirect_uri is omitted from the token request but was present in the authorization request.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>416</line_numbers>\n\n<comment>The `redirect_uri` is included in the authorization request (line 408) but is missing from the token exchange request (line 416-419). Per the OAuth2 spec (RFC 6749, Section 4.1.3), if a `redirect_uri` was included in the authorization request, the same `redirect_uri` MUST be included in the access token request. Some OAuth providers (including GitHub in certain configurations) may reject the token exchange or behave unexpectedly when the redirect_uri is omitted from the token request but was present in the authorization request.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:01Z"
+          },
+          {
+            "path": "src/sentry/integrations/github/integration.py",
+            "line": 150,
+            "body": "P4: The `error()` helper function at line 129 receives `org` (which is `self.active_organization`) and calls `org.organization` (line 150). However, `self.active_organization` is set by `determine_active_organization()` and can be `None`. While there is a `if org and ...` guard in `get_document_origin`, the `org` parameter is typed loosely and callers in both `OAuthLoginView` and `GitHubInstallation` pass `self.active_organization` which may not have been set if `determine_active_organization` wasn't called yet. In `OAuthLoginView.dispatch()`, `determine_active_organization` is called at line 391, but the redirect at line 407 returns before the error paths \u2014 so error paths at lines 412-436 are fine. However, the `error()` function accesses `org.organization` which assumes a specific object structure \u2014 if the active_organization is a different type than expected, this would crash.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>150</line_numbers>\n\n<comment>The `error()` helper function at line 129 receives `org` (which is `self.active_organization`) and calls `org.organization` (line 150). However, `self.active_organization` is set by `determine_active_organization()` and can be `None`. While there is a `if org and ...` guard in `get_document_origin`, the `org` parameter is typed loosely and callers in both `OAuthLoginView` and `GitHubInstallation` pass `self.active_organization` which may not have been set if `determine_active_organization` wasn't called yet. In `OAuthLoginView.dispatch()`, `determine_active_organization` is called at line 391, but the redirect at line 407 returns before the error paths \u2014 so error paths at lines 412-436 are fine. However, the `error()` function accesses `org.organization` which assumes a specific object structure \u2014 if the active_organization is a different type than expected, this would crash.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>src/sentry/integrations/github/integration.py</file_path>\n<line_numbers>150</line_numbers>\n\n<comment>The `error()` helper function at line 129 receives `org` (which is `self.active_organization`) and calls `org.organization` (line 150). However, `self.active_organization` is set by `determine_active_organization()` and can be `None`. While there is a `if org and ...` guard in `get_document_origin`, the `org` parameter is typed loosely and callers in both `OAuthLoginView` and `GitHubInstallation` pass `self.active_organization` which may not have been set if `determine_active_organization` wasn't called yet. In `OAuthLoginView.dispatch()`, `determine_active_organization` is called at line 391, but the redirect at line 407 returns before the error paths \u2014 so error paths at lines 412-436 are fine. However, the `error()` function accesses `org.organization` which assumes a specific object structure \u2014 if the active_organization is a different type than expected, this would crash.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:55:01Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T01:55:01Z"
+          }
+        ]
       }
     ]
   },
@@ -23351,6 +24062,37 @@
             "created_at": "2026-03-18T20:27:47Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "keycloak__keycloak__tusk__PR32918__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR32918__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "testsuite/integration-arquillian/tests/base/src/test/java/org/keycloak/testsuite/organization/cache/OrganizationCacheTest.java",
+            "line": 381,
+            "body": "P3: Test cleanup registers removal for a literal `\"alias\"` instead of the actual IDP alias (e.g., `\"idp-alias-\" + i`). This means all 20 IDPs created in the loop are never cleaned up after the test, potentially causing state leakage to other tests in the suite. The same bug exists at line 425 for `idp-alias-20`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>testsuite/integration-arquillian/tests/base/src/test/java/org/keycloak/testsuite/organization/cache/OrganizationCacheTest.java</file_path>\n<line_numbers>381</line_numbers>\n\n<comment>Test cleanup registers removal for a literal `\"alias\"` instead of the actual IDP alias (e.g., `\"idp-alias-\" + i`). This means all 20 IDPs created in the loop are never cleaned up after the test, potentially causing state leakage to other tests in the suite. The same bug exists at line 425 for `idp-alias-20`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>testsuite/integration-arquillian/tests/base/src/test/java/org/keycloak/testsuite/organization/cache/OrganizationCacheTest.java</file_path>\n<line_numbers>381</line_numbers>\n\n<comment>Test cleanup registers removal for a literal `\"alias\"` instead of the actual IDP alias (e.g., `\"idp-alias-\" + i`). This means all 20 IDPs created in the loop are never cleaned up after the test, potentially causing state leakage to other tests in the suite. The same bug exists at line 425 for `idp-alias-20`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:36Z"
+          },
+          {
+            "path": "model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/idp/InfinispanIdentityProviderStorageProvider.java",
+            "line": 244,
+            "body": "P4: The `getForLogin` cached path returns results from a `HashSet<IdentityProviderModel>` which does not preserve any ordering from the delegate's query. While current callers sort the results, this differs from the behavior of the non-cached path (which preserves delegate ordering) and the `getByOrganization` method (which also uses HashSet but documents no ordering guarantee). Any future caller relying on delegate ordering would get inconsistent behavior between cached and non-cached paths.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/idp/InfinispanIdentityProviderStorageProvider.java</file_path>\n<line_numbers>244</line_numbers>\n\n<comment>The `getForLogin` cached path returns results from a `HashSet<IdentityProviderModel>` which does not preserve any ordering from the delegate's query. While current callers sort the results, this differs from the behavior of the non-cached path (which preserves delegate ordering) and the `getByOrganization` method (which also uses HashSet but documents no ordering guarantee). Any future caller relying on delegate ordering would get inconsistent behavior between cached and non-cached paths.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/idp/InfinispanIdentityProviderStorageProvider.java</file_path>\n<line_numbers>244</line_numbers>\n\n<comment>The `getForLogin` cached path returns results from a `HashSet<IdentityProviderModel>` which does not preserve any ordering from the delegate's query. While current callers sort the results, this differs from the behavior of the non-cached path (which preserves delegate ordering) and the `getByOrganization` method (which also uses HashSet but documents no ordering guarantee). Any future caller relying on delegate ordering would get inconsistent behavior between cached and non-cached paths.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:36Z"
+          },
+          {
+            "path": "model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/idp/InfinispanIdentityProviderStorageProvider.java",
+            "line": 405,
+            "body": "P4: In `registerIDPLoginInvalidationOnUpdate`, the `original` model is fetched via `getById()` which wraps it with `createOrganizationAwareIdentityProviderModel`, overriding `isEnabled()` to dynamically check org state. The `updated` model is raw (not wrapped). When comparing both with `getLoginPredicate()`, the `isEnabled()` check may yield different results for `original` vs `updated` even if the IDP's own `enabled` field hasn't changed (e.g., if the org is disabled). This causes unnecessary cache invalidation (over-invalidation) whenever the org-aware `isEnabled()` differs from the raw value, but will never cause missed invalidation.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/idp/InfinispanIdentityProviderStorageProvider.java</file_path>\n<line_numbers>405</line_numbers>\n\n<comment>In `registerIDPLoginInvalidationOnUpdate`, the `original` model is fetched via `getById()` which wraps it with `createOrganizationAwareIdentityProviderModel`, overriding `isEnabled()` to dynamically check org state. The `updated` model is raw (not wrapped). When comparing both with `getLoginPredicate()`, the `isEnabled()` check may yield different results for `original` vs `updated` even if the IDP's own `enabled` field hasn't changed (e.g., if the org is disabled). This causes unnecessary cache invalidation (over-invalidation) whenever the org-aware `isEnabled()` differs from the raw value, but will never cause missed invalidation.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>model/infinispan/src/main/java/org/keycloak/models/cache/infinispan/idp/InfinispanIdentityProviderStorageProvider.java</file_path>\n<line_numbers>405</line_numbers>\n\n<comment>In `registerIDPLoginInvalidationOnUpdate`, the `original` model is fetched via `getById()` which wraps it with `createOrganizationAwareIdentityProviderModel`, overriding `isEnabled()` to dynamically check org state. The `updated` model is raw (not wrapped). When comparing both with `getLoginPredicate()`, the `isEnabled()` check may yield different results for `original` vs `updated` even if the IDP's own `enabled` field hasn't changed (e.g., if the org is disabled). This causes unnecessary cache invalidation (over-invalidation) whenever the org-aware `isEnabled()` differs from the raw value, but will never cause missed invalidation.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:36Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:30:36Z"
+          }
+        ]
       }
     ]
   },
@@ -24237,6 +24979,31 @@
             "created_at": "2026-03-18T20:27:01Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR94942__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR94942__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/expr/reader.go",
+            "line": 194,
+            "body": "P3: The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code. When `FlagSqlExpressions` is enabled globally, `!IsEnabledGlobally()` is `false`, so `enabled` is `false`, and the function falls through to `return false`. When the flag is disabled, `enabled` is `true`, and the `if enabled` branch explicitly `return false`. Both paths return `false`. While this achieves the PR's goal of disabling SQL expressions entirely, the convoluted logic with a meaningless feature flag check is misleading \u2014 it suggests the flag could re-enable the feature when it cannot. A simple `return false` would be clearer and wouldn't mislead future maintainers into thinking the feature flag is load-bearing.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/expr/reader.go</file_path>\n<line_numbers>194</line_numbers>\n\n<comment>The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code. When `FlagSqlExpressions` is enabled globally, `!IsEnabledGlobally()` is `false`, so `enabled` is `false`, and the function falls through to `return false`. When the flag is disabled, `enabled` is `true`, and the `if enabled` branch explicitly `return false`. Both paths return `false`. While this achieves the PR's goal of disabling SQL expressions entirely, the convoluted logic with a meaningless feature flag check is misleading \u2014 it suggests the flag could re-enable the feature when it cannot. A simple `return false` would be clearer and wouldn't mislead future maintainers into thinking the feature flag is load-bearing.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/expr/reader.go</file_path>\n<line_numbers>194</line_numbers>\n\n<comment>The `enableSqlExpressions` function always returns `false` regardless of the feature flag value, making the feature flag check dead code. When `FlagSqlExpressions` is enabled globally, `!IsEnabledGlobally()` is `false`, so `enabled` is `false`, and the function falls through to `return false`. When the flag is disabled, `enabled` is `true`, and the `if enabled` branch explicitly `return false`. Both paths return `false`. While this achieves the PR's goal of disabling SQL expressions entirely, the convoluted logic with a meaningless feature flag check is misleading \u2014 it suggests the flag could re-enable the feature when it cannot. A simple `return false` would be clearer and wouldn't mislead future maintainers into thinking the feature flag is load-bearing.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:58:29Z"
+          },
+          {
+            "path": "pkg/expr/sql_command.go",
+            "line": 96,
+            "body": "P4: The stub `DB.QueryFramesInto` method always returns an error (\"not implemented\"), but the caller in `sql_command.go:Execute` (line ~96-99) treats this error as a query failure and sets `rsp.Error = err`, then continues to build a result with `mathexp.TableData{Frame: frame}` and appends it to `rsp.Values`. This means even though `QueryFramesInto` fails, an empty frame is still added to the results. However, since `enableSqlExpressions` now always returns false (blocking at the reader level), this code path is unreachable in practice \u2014 the SQL command can never be constructed. This is a latent issue only if the guard is ever removed.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/expr/sql_command.go</file_path>\n<line_numbers>96</line_numbers>\n\n<comment>The stub `DB.QueryFramesInto` method always returns an error (\"not implemented\"), but the caller in `sql_command.go:Execute` (line ~96-99) treats this error as a query failure and sets `rsp.Error = err`, then continues to build a result with `mathexp.TableData{Frame: frame}` and appends it to `rsp.Values`. This means even though `QueryFramesInto` fails, an empty frame is still added to the results. However, since `enableSqlExpressions` now always returns false (blocking at the reader level), this code path is unreachable in practice \u2014 the SQL command can never be constructed. This is a latent issue only if the guard is ever removed.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/expr/sql_command.go</file_path>\n<line_numbers>96</line_numbers>\n\n<comment>The stub `DB.QueryFramesInto` method always returns an error (\"not implemented\"), but the caller in `sql_command.go:Execute` (line ~96-99) treats this error as a query failure and sets `rsp.Error = err`, then continues to build a result with `mathexp.TableData{Frame: frame}` and appends it to `rsp.Values`. This means even though `QueryFramesInto` fails, an empty frame is still added to the results. However, since `enableSqlExpressions` now always returns false (blocking at the reader level), this code path is unreachable in practice \u2014 the SQL command can never be constructed. This is a latent issue only if the guard is ever removed.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:58:29Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 2 issues",
+            "created_at": "2026-04-12T01:58:28Z"
+          }
+        ]
       }
     ]
   },
@@ -25034,6 +25801,25 @@
             "created_at": "2026-03-18T20:27:02Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR90939__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90939__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/api/webassets/webassets.go",
+            "line": 48,
+            "body": "P3: Classic double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether `entryPointAssetsCache` was already populated by another goroutine. Between RUnlock and Lock, another goroutine can acquire the Lock first and populate the cache. The current goroutine then unnecessarily re-reads and re-parses the manifest file, performing redundant I/O. In non-dev mode with many concurrent requests at startup, this causes a thundering herd of file reads. The fix is to add a second check after acquiring the write lock: `if cfg.Env != setting.Dev && entryPointAssetsCache != nil { return entryPointAssetsCache, nil }`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/api/webassets/webassets.go</file_path>\n<line_numbers>48</line_numbers>\n\n<comment>Classic double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether `entryPointAssetsCache` was already populated by another goroutine. Between RUnlock and Lock, another goroutine can acquire the Lock first and populate the cache. The current goroutine then unnecessarily re-reads and re-parses the manifest file, performing redundant I/O. In non-dev mode with many concurrent requests at startup, this causes a thundering herd of file reads. The fix is to add a second check after acquiring the write lock: `if cfg.Env != setting.Dev && entryPointAssetsCache != nil { return entryPointAssetsCache, nil }`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/api/webassets/webassets.go</file_path>\n<line_numbers>48</line_numbers>\n\n<comment>Classic double-checked locking bug: after releasing RLock (line 43) and acquiring Lock (line 48), the code does not re-check whether `entryPointAssetsCache` was already populated by another goroutine. Between RUnlock and Lock, another goroutine can acquire the Lock first and populate the cache. The current goroutine then unnecessarily re-reads and re-parses the manifest file, performing redundant I/O. In non-dev mode with many concurrent requests at startup, this causes a thundering herd of file reads. The fix is to add a second check after acquiring the write lock: `if cfg.Env != setting.Dev && entryPointAssetsCache != nil { return entryPointAssetsCache, nil }`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T01:59:35Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 1 issue",
+            "created_at": "2026-04-12T01:59:34Z"
+          }
+        ]
       }
     ]
   },
@@ -26092,6 +26878,43 @@
             "created_at": "2026-03-18T20:27:03Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR80329__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR80329__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/services/cleanup/cleanup.go",
+            "line": 77,
+            "body": "P1: The cleanup service ticker was changed from 10 minutes to 1 minute. This appears to be an accidental debug leftover. The `clean()` method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running. This 10x increase in cleanup frequency will also increase database load significantly in production.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/cleanup/cleanup.go</file_path>\n<line_numbers>77</line_numbers>\n\n<comment>The cleanup service ticker was changed from 10 minutes to 1 minute. This appears to be an accidental debug leftover. The `clean()` method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running. This 10x increase in cleanup frequency will also increase database load significantly in production.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/cleanup/cleanup.go</file_path>\n<line_numbers>77</line_numbers>\n\n<comment>The cleanup service ticker was changed from 10 minutes to 1 minute. This appears to be an accidental debug leftover. The `clean()` method uses a 9-minute timeout, so a 1-minute ticker means a new cleanup cycle could be triggered while the previous one is still running. This 10x increase in cleanup frequency will also increase database load significantly in production.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:36Z"
+          },
+          {
+            "path": "pkg/services/annotations/annotationsimpl/xorm_store.go",
+            "line": 534,
+            "body": "P2: All diagnostic log lines in the cleanup path use `r.log.Error()` instead of `r.log.Debug()` or `r.log.Info()`. These are clearly informational messages (e.g., \"Annotations to clean by time\", \"cleaned annotations by time\") that fire on every successful cleanup cycle. In production this will flood error-level logs and likely trigger monitoring/alerting systems, making real errors harder to find.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/annotations/annotationsimpl/xorm_store.go</file_path>\n<line_numbers>534</line_numbers>\n\n<comment>All diagnostic log lines in the cleanup path use `r.log.Error()` instead of `r.log.Debug()` or `r.log.Info()`. These are clearly informational messages (e.g., \"Annotations to clean by time\", \"cleaned annotations by time\") that fire on every successful cleanup cycle. In production this will flood error-level logs and likely trigger monitoring/alerting systems, making real errors harder to find.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/annotations/annotationsimpl/xorm_store.go</file_path>\n<line_numbers>534</line_numbers>\n\n<comment>All diagnostic log lines in the cleanup path use `r.log.Error()` instead of `r.log.Debug()` or `r.log.Info()`. These are clearly informational messages (e.g., \"Annotations to clean by time\", \"cleaned annotations by time\") that fire on every successful cleanup cycle. In production this will flood error-level logs and likely trigger monitoring/alerting systems, making real errors harder to find.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:36Z"
+          },
+          {
+            "path": "pkg/services/annotations/annotationsimpl/xorm_store.go",
+            "line": 608,
+            "body": "P3: The SQLite parameter limit guard checks `r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit` but should check `len(ids) > sqliteParameterLimit`. The batch size is the *maximum* number of IDs that can be fetched, but the actual number returned may be less. More importantly, if `AnnotationCleanupJobBatchSize` is e.g. 1000 but only 5 IDs are returned, the code will unnecessarily build an inline SQL string instead of using parameterized queries, which is the safer approach. The guard should reflect the actual parameter count, not the configured ceiling.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/annotations/annotationsimpl/xorm_store.go</file_path>\n<line_numbers>608</line_numbers>\n\n<comment>The SQLite parameter limit guard checks `r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit` but should check `len(ids) > sqliteParameterLimit`. The batch size is the *maximum* number of IDs that can be fetched, but the actual number returned may be less. More importantly, if `AnnotationCleanupJobBatchSize` is e.g. 1000 but only 5 IDs are returned, the code will unnecessarily build an inline SQL string instead of using parameterized queries, which is the safer approach. The guard should reflect the actual parameter count, not the configured ceiling.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/annotations/annotationsimpl/xorm_store.go</file_path>\n<line_numbers>608</line_numbers>\n\n<comment>The SQLite parameter limit guard checks `r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit` but should check `len(ids) > sqliteParameterLimit`. The batch size is the *maximum* number of IDs that can be fetched, but the actual number returned may be less. More importantly, if `AnnotationCleanupJobBatchSize` is e.g. 1000 but only 5 IDs are returned, the code will unnecessarily build an inline SQL string instead of using parameterized queries, which is the safer approach. The guard should reflect the actual parameter count, not the configured ceiling.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:36Z"
+          },
+          {
+            "path": "pkg/services/annotations/annotationsimpl/cleanup_test.go",
+            "line": 241,
+            "body": "P4: Test helper `createTestAnnotations` hardcodes annotation IDs as `int64(i + 1)` via `InsertMulti`. Since tests share a single `db.InitTestDB(t)` and SQLite's auto-increment counter is not reset by `DELETE FROM annotation`, subsequent test cases that also insert IDs starting at 1 will collide with the auto-increment sequence on databases that enforce uniqueness on the primary key. This works today because explicit IDs are provided, but if any test case inserts *without* explicit IDs (like `TestIntegrationOldAnnotationsAreDeletedFirst` at line 178), the auto-generated IDs may collide with IDs from a previous `createTestAnnotations` call that didn't fully clean up yet.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/annotations/annotationsimpl/cleanup_test.go</file_path>\n<line_numbers>241</line_numbers>\n\n<comment>Test helper `createTestAnnotations` hardcodes annotation IDs as `int64(i + 1)` via `InsertMulti`. Since tests share a single `db.InitTestDB(t)` and SQLite's auto-increment counter is not reset by `DELETE FROM annotation`, subsequent test cases that also insert IDs starting at 1 will collide with the auto-increment sequence on databases that enforce uniqueness on the primary key. This works today because explicit IDs are provided, but if any test case inserts *without* explicit IDs (like `TestIntegrationOldAnnotationsAreDeletedFirst` at line 178), the auto-generated IDs may collide with IDs from a previous `createTestAnnotations` call that didn't fully clean up yet.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/annotations/annotationsimpl/cleanup_test.go</file_path>\n<line_numbers>241</line_numbers>\n\n<comment>Test helper `createTestAnnotations` hardcodes annotation IDs as `int64(i + 1)` via `InsertMulti`. Since tests share a single `db.InitTestDB(t)` and SQLite's auto-increment counter is not reset by `DELETE FROM annotation`, subsequent test cases that also insert IDs starting at 1 will collide with the auto-increment sequence on databases that enforce uniqueness on the primary key. This works today because explicit IDs are provided, but if any test case inserts *without* explicit IDs (like `TestIntegrationOldAnnotationsAreDeletedFirst` at line 178), the auto-generated IDs may collide with IDs from a previous `createTestAnnotations` call that didn't fully clean up yet.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:36Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:00:36Z"
+          }
+        ]
       }
     ]
   },
@@ -27475,6 +28298,49 @@
             "created_at": "2026-03-18T20:27:12Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR90045__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90045__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/apiserver/rest/dualwriter_mode3.go",
+            "line": 45,
+            "body": "P2: In Create's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. The storage operation failed, but the metric is recorded against the legacy histogram, corrupting both metrics.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>45</line_numbers>\n\n<comment>In Create's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. The storage operation failed, but the metric is recorded against the legacy histogram, corrupting both metrics.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>45</line_numbers>\n\n<comment>In Create's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. The storage operation failed, but the metric is recorded against the legacy histogram, corrupting both metrics.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:42Z"
+          },
+          {
+            "path": "pkg/apiserver/rest/dualwriter_mode3.go",
+            "line": 129,
+            "body": "P2: In Update's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. Same issue as Create \u2014 the storage call failed but the latency is attributed to legacy.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>129</line_numbers>\n\n<comment>In Update's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. Same issue as Create \u2014 the storage call failed but the latency is attributed to legacy.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>129</line_numbers>\n\n<comment>In Update's error path, `recordLegacyDuration` is called instead of `recordStorageDuration`. Same issue as Create \u2014 the storage call failed but the latency is attributed to legacy.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:42Z"
+          },
+          {
+            "path": "pkg/apiserver/rest/dualwriter_mode3.go",
+            "line": 166,
+            "body": "P2: In DeleteCollection's goroutine, `recordStorageDuration` is called instead of `recordLegacyDuration` for the legacy write. This is the inverse mistake \u2014 legacy latency is recorded in the storage histogram.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>166</line_numbers>\n\n<comment>In DeleteCollection's goroutine, `recordStorageDuration` is called instead of `recordLegacyDuration` for the legacy write. This is the inverse mistake \u2014 legacy latency is recorded in the storage histogram.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>166</line_numbers>\n\n<comment>In DeleteCollection's goroutine, `recordStorageDuration` is called instead of `recordLegacyDuration` for the legacy write. This is the inverse mistake \u2014 legacy latency is recorded in the storage histogram.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:42Z"
+          },
+          {
+            "path": "pkg/apiserver/rest/dualwriter_mode3.go",
+            "line": 106,
+            "body": "P2: In Delete's success path, the third argument to `recordStorageDuration` is `name` (the object name, e.g. \"my-playlist\") instead of `options.Kind`. Every other call site in this file and in mode 1 passes `options.Kind` as the metric label. This will create high-cardinality metric labels (one time series per object name), which can cause performance issues in Prometheus.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>106</line_numbers>\n\n<comment>In Delete's success path, the third argument to `recordStorageDuration` is `name` (the object name, e.g. \"my-playlist\") instead of `options.Kind`. Every other call site in this file and in mode 1 passes `options.Kind` as the metric label. This will create high-cardinality metric labels (one time series per object name), which can cause performance issues in Prometheus.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>106</line_numbers>\n\n<comment>In Delete's success path, the third argument to `recordStorageDuration` is `name` (the object name, e.g. \"my-playlist\") instead of `options.Kind`. Every other call site in this file and in mode 1 passes `options.Kind` as the metric label. This will create high-cardinality metric labels (one time series per object name), which can cause performance issues in Prometheus.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:42Z"
+          },
+          {
+            "path": "pkg/apiserver/rest/dualwriter_mode3.go",
+            "line": 97,
+            "body": "P3: In the Delete method, `klog.NewContext(ctx, d.Log)` uses `d.Log` (the base logger) instead of the enriched `log` variable that includes the `name`, `kind`, and `method` values. All other methods correctly pass `log` to `klog.NewContext`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>97</line_numbers>\n\n<comment>In the Delete method, `klog.NewContext(ctx, d.Log)` uses `d.Log` (the base logger) instead of the enriched `log` variable that includes the `name`, `kind`, and `method` values. All other methods correctly pass `log` to `klog.NewContext`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/apiserver/rest/dualwriter_mode3.go</file_path>\n<line_numbers>97</line_numbers>\n\n<comment>In the Delete method, `klog.NewContext(ctx, d.Log)` uses `d.Log` (the base logger) instead of the enriched `log` variable that includes the `name`, `kind`, and `method` values. All other methods correctly pass `log` to `klog.NewContext`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:00:42Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:00:42Z"
+          }
+        ]
       }
     ]
   },
@@ -28591,6 +29457,49 @@
             "created_at": "2026-03-18T20:26:57Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR106778__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR106778__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "public/app/features/alerting/unified/rule-list/FilterView.tsx",
+            "line": 157,
+            "body": "P2: Missing React `key` prop on GrafanaRuleListItem in the filter view's list rendering. The old `GrafanaRuleLoader` had `key={key}`, but the replacement `GrafanaRuleListItem` omits it. The other branches (`DataSourceRuleLoader`, `UnknownRuleListItem`) all have `key={key}`. This will cause React reconciliation issues when the list updates (e.g., items reordering, stale state, performance degradation).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/features/alerting/unified/rule-list/FilterView.tsx</file_path>\n<line_numbers>157</line_numbers>\n\n<comment>Missing React `key` prop on GrafanaRuleListItem in the filter view's list rendering. The old `GrafanaRuleLoader` had `key={key}`, but the replacement `GrafanaRuleListItem` omits it. The other branches (`DataSourceRuleLoader`, `UnknownRuleListItem`) all have `key={key}`. This will cause React reconciliation issues when the list updates (e.g., items reordering, stale state, performance degradation).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/features/alerting/unified/rule-list/FilterView.tsx</file_path>\n<line_numbers>157</line_numbers>\n\n<comment>Missing React `key` prop on GrafanaRuleListItem in the filter view's list rendering. The old `GrafanaRuleLoader` had `key={key}`, but the replacement `GrafanaRuleListItem` omits it. The other branches (`DataSourceRuleLoader`, `UnknownRuleListItem`) all have `key={key}`. This will cause React reconciliation issues when the list updates (e.g., items reordering, stale state, performance degradation).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:25Z"
+          },
+          {
+            "path": "public/app/features/alerting/unified/components/rule-viewer/AlertRuleMenu.tsx",
+            "line": 103,
+            "body": "P2: _Note: Issue is at line 165, shown here on nearest changed line._\n\nDelete menu item will never render in the new list view. `AlertRuleMenu` at line 165 guards the delete menu item with `canDelete && rulerRule && (...)`. Since the new list view no longer fetches ruler data, `rulerRule` is always `undefined` when coming from `GrafanaRuleListItem` \u2192 `RuleActionsButtons.V2` \u2192 `AlertRuleMenu`. Even though `canDelete` becomes true via the new `grafanaDeleteAbility` path, the `rulerRule` guard prevents the delete option from appearing.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/features/alerting/unified/components/rule-viewer/AlertRuleMenu.tsx</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>_Note: Issue is at line 165, shown here on nearest changed line._\n\nDelete menu item will never render in the new list view. `AlertRuleMenu` at line 165 guards the delete menu item with `canDelete && rulerRule && (...)`. Since the new list view no longer fetches ruler data, `rulerRule` is always `undefined` when coming from `GrafanaRuleListItem` \u2192 `RuleActionsButtons.V2` \u2192 `AlertRuleMenu`. Even though `canDelete` becomes true via the new `grafanaDeleteAbility` path, the `rulerRule` guard prevents the delete option from appearing.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/features/alerting/unified/components/rule-viewer/AlertRuleMenu.tsx</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>_Note: Issue is at line 165, shown here on nearest changed line._\n\nDelete menu item will never render in the new list view. `AlertRuleMenu` at line 165 guards the delete menu item with `canDelete && rulerRule && (...)`. Since the new list view no longer fetches ruler data, `rulerRule` is always `undefined` when coming from `GrafanaRuleListItem` \u2192 `RuleActionsButtons.V2` \u2192 `AlertRuleMenu`. Even though `canDelete` becomes true via the new `grafanaDeleteAbility` path, the `rulerRule` guard prevents the delete option from appearing.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:25Z"
+          },
+          {
+            "path": "public/app/features/alerting/unified/components/rule-viewer/AlertRuleMenu.tsx",
+            "line": 103,
+            "body": "P2: _Note: Issue is at line 123, shown here on nearest changed line._\n\nPause menu item will never render in the new list view. At line 123, `canPause && rulerRuleType.grafana.rule(rulerRule)` requires `rulerRule` to be a grafana ruler rule. Since `rulerRule` is undefined when coming from the new list view, `rulerRuleType.grafana.rule(undefined)` returns false, so the pause option never appears \u2014 even though the user has permission via the grafana prom abilities path.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/features/alerting/unified/components/rule-viewer/AlertRuleMenu.tsx</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>_Note: Issue is at line 123, shown here on nearest changed line._\n\nPause menu item will never render in the new list view. At line 123, `canPause && rulerRuleType.grafana.rule(rulerRule)` requires `rulerRule` to be a grafana ruler rule. Since `rulerRule` is undefined when coming from the new list view, `rulerRuleType.grafana.rule(undefined)` returns false, so the pause option never appears \u2014 even though the user has permission via the grafana prom abilities path.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/features/alerting/unified/components/rule-viewer/AlertRuleMenu.tsx</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>_Note: Issue is at line 123, shown here on nearest changed line._\n\nPause menu item will never render in the new list view. At line 123, `canPause && rulerRuleType.grafana.rule(rulerRule)` requires `rulerRule` to be a grafana ruler rule. Since `rulerRule` is undefined when coming from the new list view, `rulerRuleType.grafana.rule(undefined)` returns false, so the pause option never appears \u2014 even though the user has permission via the grafana prom abilities path.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:25Z"
+          },
+          {
+            "path": "public/app/features/alerting/unified/rule-list/components/RuleActionsButtons.V2.tsx",
+            "line": 114,
+            "body": "P2: _Note: Issue is at line 104, shown here on nearest changed line._\n\nSilence drawer won't open in the new list view. In `RuleActionsButtons.V2.tsx`, the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` at line 104. When `rule` (the RulerRuleDTO) is undefined (coming from GrafanaRuleListItem which only passes `promRule`), this check is false. So even though the \"Silence notifications\" menu item appears (via `canSilence` from the grafana prom path), clicking it sets `showSilenceDrawer` to true but the drawer never renders.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/features/alerting/unified/rule-list/components/RuleActionsButtons.V2.tsx</file_path>\n<line_numbers>114</line_numbers>\n\n<comment>_Note: Issue is at line 104, shown here on nearest changed line._\n\nSilence drawer won't open in the new list view. In `RuleActionsButtons.V2.tsx`, the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` at line 104. When `rule` (the RulerRuleDTO) is undefined (coming from GrafanaRuleListItem which only passes `promRule`), this check is false. So even though the \"Silence notifications\" menu item appears (via `canSilence` from the grafana prom path), clicking it sets `showSilenceDrawer` to true but the drawer never renders.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/features/alerting/unified/rule-list/components/RuleActionsButtons.V2.tsx</file_path>\n<line_numbers>114</line_numbers>\n\n<comment>_Note: Issue is at line 104, shown here on nearest changed line._\n\nSilence drawer won't open in the new list view. In `RuleActionsButtons.V2.tsx`, the silence drawer rendering is guarded by `rulerRuleType.grafana.alertingRule(rule)` at line 104. When `rule` (the RulerRuleDTO) is undefined (coming from GrafanaRuleListItem which only passes `promRule`), this check is false. So even though the \"Silence notifications\" menu item appears (via `canSilence` from the grafana prom path), clicking it sets `showSilenceDrawer` to true but the drawer never renders.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:25Z"
+          },
+          {
+            "path": "public/app/features/alerting/unified/rule-list/components/RuleActionsButtons.V2.tsx",
+            "line": 57,
+            "body": "P3: The `useGrafanaPromRuleAbility` hook in `RuleActionsButtons.V2.tsx` is called with `promRule` typed as `Rule` (from unified-alerting.ts), but the hook expects `GrafanaPromRuleDTO | SkipToken`. While `prometheusRuleType.grafana.rule()` does a structural check, it returns `boolean` (not a type predicate narrowing to `GrafanaPromRuleDTO`), so TypeScript won't narrow the type. This is likely a compile error, or if it passes due to structural compatibility, the `folderUid` field used inside `useIsGrafanaPromRuleEditable` may be missing on non-Grafana `Rule` objects (where `folderUid` is optional), leading to incorrect permission checks.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/features/alerting/unified/rule-list/components/RuleActionsButtons.V2.tsx</file_path>\n<line_numbers>57</line_numbers>\n\n<comment>The `useGrafanaPromRuleAbility` hook in `RuleActionsButtons.V2.tsx` is called with `promRule` typed as `Rule` (from unified-alerting.ts), but the hook expects `GrafanaPromRuleDTO | SkipToken`. While `prometheusRuleType.grafana.rule()` does a structural check, it returns `boolean` (not a type predicate narrowing to `GrafanaPromRuleDTO`), so TypeScript won't narrow the type. This is likely a compile error, or if it passes due to structural compatibility, the `folderUid` field used inside `useIsGrafanaPromRuleEditable` may be missing on non-Grafana `Rule` objects (where `folderUid` is optional), leading to incorrect permission checks.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/features/alerting/unified/rule-list/components/RuleActionsButtons.V2.tsx</file_path>\n<line_numbers>57</line_numbers>\n\n<comment>The `useGrafanaPromRuleAbility` hook in `RuleActionsButtons.V2.tsx` is called with `promRule` typed as `Rule` (from unified-alerting.ts), but the hook expects `GrafanaPromRuleDTO | SkipToken`. While `prometheusRuleType.grafana.rule()` does a structural check, it returns `boolean` (not a type predicate narrowing to `GrafanaPromRuleDTO`), so TypeScript won't narrow the type. This is likely a compile error, or if it passes due to structural compatibility, the `folderUid` field used inside `useIsGrafanaPromRuleEditable` may be missing on non-Grafana `Rule` objects (where `folderUid` is optional), leading to incorrect permission checks.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:25Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:04:25Z"
+          }
+        ]
       }
     ]
   },
@@ -29444,6 +30353,37 @@
             "created_at": "2026-03-18T20:26:58Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR107534__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR107534__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "public/app/plugins/datasource/loki/querySplitting.ts",
+            "line": 299,
+            "body": "P2: Double interpolation of queries in the shard splitting path. `runShardSplitQuery()` calls `applyTemplateVariables()` at line 52, then passes queries to `splitQueriesByStreamShard()`, which calls `runSplitQuery()` at line 163. `runSplitQuery()` now also calls `applyTemplateVariables()` at line 299. While template variable replacement is idempotent (already-replaced text won't match), `addAdHocFilters()` in `datasource.ts:1080-1088` unconditionally appends label matchers to the expression via `addLabelToQuery()`. This means ad-hoc filters (e.g., `env=\"prod\"`) will be added to the query expression twice. The test at line 112 expects `applyTemplateVariables` to be called 5 times (1 from `runShardSplitQuery` + 4 from `runSplitQuery` calls), confirming the double-call pattern, but the test mock uses simple string replacement which is idempotent, masking the real bug.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/plugins/datasource/loki/querySplitting.ts</file_path>\n<line_numbers>299</line_numbers>\n\n<comment>Double interpolation of queries in the shard splitting path. `runShardSplitQuery()` calls `applyTemplateVariables()` at line 52, then passes queries to `splitQueriesByStreamShard()`, which calls `runSplitQuery()` at line 163. `runSplitQuery()` now also calls `applyTemplateVariables()` at line 299. While template variable replacement is idempotent (already-replaced text won't match), `addAdHocFilters()` in `datasource.ts:1080-1088` unconditionally appends label matchers to the expression via `addLabelToQuery()`. This means ad-hoc filters (e.g., `env=\"prod\"`) will be added to the query expression twice. The test at line 112 expects `applyTemplateVariables` to be called 5 times (1 from `runShardSplitQuery` + 4 from `runSplitQuery` calls), confirming the double-call pattern, but the test mock uses simple string replacement which is idempotent, masking the real bug.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/plugins/datasource/loki/querySplitting.ts</file_path>\n<line_numbers>299</line_numbers>\n\n<comment>Double interpolation of queries in the shard splitting path. `runShardSplitQuery()` calls `applyTemplateVariables()` at line 52, then passes queries to `splitQueriesByStreamShard()`, which calls `runSplitQuery()` at line 163. `runSplitQuery()` now also calls `applyTemplateVariables()` at line 299. While template variable replacement is idempotent (already-replaced text won't match), `addAdHocFilters()` in `datasource.ts:1080-1088` unconditionally appends label matchers to the expression via `addLabelToQuery()`. This means ad-hoc filters (e.g., `env=\"prod\"`) will be added to the query expression twice. The test at line 112 expects `applyTemplateVariables` to be called 5 times (1 from `runShardSplitQuery` + 4 from `runSplitQuery` calls), confirming the double-call pattern, but the test mock uses simple string replacement which is idempotent, masking the real bug.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:22Z"
+          },
+          {
+            "path": "public/app/plugins/datasource/loki/shardQuerySplitting.test.ts",
+            "line": 57,
+            "body": "P3: Test mock for `applyTemplateVariables` mutates the input query object directly (`query.expr = query.expr.replace(...)`) instead of returning a new object as the real implementation does (which returns `{ ...target, expr, step, legendFormat }`). This means the mock modifies the original `request.targets` entries in place, which hides potential bugs where code relies on the original targets remaining unmodified (e.g., `request.targets` is still used by `runSplitQuery` after shard interpolation). If the mock returned a new object like the real code, the test assertions for `runQuery` call arguments might fail, revealing the double-interpolation issue.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/plugins/datasource/loki/shardQuerySplitting.test.ts</file_path>\n<line_numbers>57</line_numbers>\n\n<comment>Test mock for `applyTemplateVariables` mutates the input query object directly (`query.expr = query.expr.replace(...)`) instead of returning a new object as the real implementation does (which returns `{ ...target, expr, step, legendFormat }`). This means the mock modifies the original `request.targets` entries in place, which hides potential bugs where code relies on the original targets remaining unmodified (e.g., `request.targets` is still used by `runSplitQuery` after shard interpolation). If the mock returned a new object like the real code, the test assertions for `runQuery` call arguments might fail, revealing the double-interpolation issue.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/plugins/datasource/loki/shardQuerySplitting.test.ts</file_path>\n<line_numbers>57</line_numbers>\n\n<comment>Test mock for `applyTemplateVariables` mutates the input query object directly (`query.expr = query.expr.replace(...)`) instead of returning a new object as the real implementation does (which returns `{ ...target, expr, step, legendFormat }`). This means the mock modifies the original `request.targets` entries in place, which hides potential bugs where code relies on the original targets remaining unmodified (e.g., `request.targets` is still used by `runSplitQuery` after shard interpolation). If the mock returned a new object like the real code, the test assertions for `runQuery` call arguments might fail, revealing the double-interpolation issue.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:22Z"
+          },
+          {
+            "path": "public/app/plugins/datasource/loki/shardQuerySplitting.ts",
+            "line": 49,
+            "body": "P4: Filter ordering change in `runShardSplitQuery` may cause behavioral difference. The old code called `interpolateVariablesInQueries` first, then filtered with `.filter((query) => query.expr)`. The new code filters first, then interpolates. If a query's `expr` is a template variable that resolves to an empty string (e.g., `expr: '$optional_query'` \u2192 `''`), the old code would correctly filter it out after interpolation, but the new code passes it through the filter (since `'$optional_query'` is truthy) and then interpolates it to empty, resulting in a query with an empty expression being sent to `splitQueriesByStreamShard`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>public/app/plugins/datasource/loki/shardQuerySplitting.ts</file_path>\n<line_numbers>49</line_numbers>\n\n<comment>Filter ordering change in `runShardSplitQuery` may cause behavioral difference. The old code called `interpolateVariablesInQueries` first, then filtered with `.filter((query) => query.expr)`. The new code filters first, then interpolates. If a query's `expr` is a template variable that resolves to an empty string (e.g., `expr: '$optional_query'` \u2192 `''`), the old code would correctly filter it out after interpolation, but the new code passes it through the filter (since `'$optional_query'` is truthy) and then interpolates it to empty, resulting in a query with an empty expression being sent to `splitQueriesByStreamShard`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>public/app/plugins/datasource/loki/shardQuerySplitting.ts</file_path>\n<line_numbers>49</line_numbers>\n\n<comment>Filter ordering change in `runShardSplitQuery` may cause behavioral difference. The old code called `interpolateVariablesInQueries` first, then filtered with `.filter((query) => query.expr)`. The new code filters first, then interpolates. If a query's `expr` is a template variable that resolves to an empty string (e.g., `expr: '$optional_query'` \u2192 `''`), the old code would correctly filter it out after interpolation, but the new code passes it through the filter (since `'$optional_query'` is truthy) and then interpolates it to empty, resulting in a query with an empty expression being sent to `splitQueriesByStreamShard`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:04:22Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:04:22Z"
+          }
+        ]
       }
     ]
   },
@@ -30553,6 +31493,43 @@
             "created_at": "2026-03-18T20:27:43Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR79265__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR79265__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/services/anonymous/anonimpl/client.go",
+            "line": 44,
+            "body": "P2: TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/anonymous/anonimpl/client.go</file_path>\n<line_numbers>44</line_numbers>\n\n<comment>TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/anonymous/anonimpl/client.go</file_path>\n<line_numbers>44</line_numbers>\n\n<comment>TagDevice was changed from fire-and-forget (goroutine with independent context/timeout) to synchronous on the request path. Previously, a slow or failing DB operation would not block authentication. Now, every anonymous request waits for the DB round-trip (CountDevices + CreateOrUpdateDevice). This adds latency to every anonymous request and means transient DB issues will slow down (though not fail) anonymous auth. The old code had a 2-minute timeout on a background context; the new code inherits the HTTP request context which may have a much shorter timeout.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:20Z"
+          },
+          {
+            "path": "pkg/services/anonymous/anonimpl/anonstore/database.go",
+            "line": 109,
+            "body": "P3: TOCTOU race in device limit enforcement. `CreateOrUpdateDevice` first counts devices, then conditionally calls `updateDevice`. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit. With high concurrency this could allow significantly more devices than intended.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/anonymous/anonimpl/anonstore/database.go</file_path>\n<line_numbers>109</line_numbers>\n\n<comment>TOCTOU race in device limit enforcement. `CreateOrUpdateDevice` first counts devices, then conditionally calls `updateDevice`. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit. With high concurrency this could allow significantly more devices than intended.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/anonymous/anonimpl/anonstore/database.go</file_path>\n<line_numbers>109</line_numbers>\n\n<comment>TOCTOU race in device limit enforcement. `CreateOrUpdateDevice` first counts devices, then conditionally calls `updateDevice`. Between the count check and the update/insert, concurrent requests can each pass the count check and create new devices, exceeding the configured limit. With high concurrency this could allow significantly more devices than intended.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:20Z"
+          },
+          {
+            "path": "pkg/services/anonymous/anonimpl/anonstore/database.go",
+            "line": 95,
+            "body": "P3: The `updateDevice` method returns `ErrDeviceLimitReached` when `rowsAffected == 0`, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its `updated_at` falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with `ErrDeviceLimitReached` instead of being re-created or handled differently. The error is misleading for the second case.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/anonymous/anonimpl/anonstore/database.go</file_path>\n<line_numbers>95</line_numbers>\n\n<comment>The `updateDevice` method returns `ErrDeviceLimitReached` when `rowsAffected == 0`, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its `updated_at` falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with `ErrDeviceLimitReached` instead of being re-created or handled differently. The error is misleading for the second case.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/anonymous/anonimpl/anonstore/database.go</file_path>\n<line_numbers>95</line_numbers>\n\n<comment>The `updateDevice` method returns `ErrDeviceLimitReached` when `rowsAffected == 0`, but this conflates two distinct cases: (1) the device doesn't exist (new device, limit should apply), and (2) the device exists but its `updated_at` falls outside the BETWEEN range (e.g., device expired). In case 2, a returning known device that expired more than 30 days ago is incorrectly rejected with `ErrDeviceLimitReached` instead of being re-created or handled differently. The error is misleading for the second case.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:21Z"
+          },
+          {
+            "path": "pkg/api/dtos/frontend_settings.go",
+            "line": 195,
+            "body": "P4: The `AnonymousDeviceLimit` field in `FrontendSettingsDTO` is `int64` with `json:\"anonymousDeviceLimit\"` \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as `number | undefined`. The Go DTO has no `omitempty` tag, so the field is never `undefined` in practice. Any frontend code checking `config.anonymousDeviceLimit === undefined` to detect 'no limit configured' will never match; it will see `0` instead.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/api/dtos/frontend_settings.go</file_path>\n<line_numbers>195</line_numbers>\n\n<comment>The `AnonymousDeviceLimit` field in `FrontendSettingsDTO` is `int64` with `json:\"anonymousDeviceLimit\"` \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as `number | undefined`. The Go DTO has no `omitempty` tag, so the field is never `undefined` in practice. Any frontend code checking `config.anonymousDeviceLimit === undefined` to detect 'no limit configured' will never match; it will see `0` instead.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/api/dtos/frontend_settings.go</file_path>\n<line_numbers>195</line_numbers>\n\n<comment>The `AnonymousDeviceLimit` field in `FrontendSettingsDTO` is `int64` with `json:\"anonymousDeviceLimit\"` \u2014 it always serializes as a number (0 when unset). However, the TypeScript interface declares the type as `number | undefined`. The Go DTO has no `omitempty` tag, so the field is never `undefined` in practice. Any frontend code checking `config.anonymousDeviceLimit === undefined` to detect 'no limit configured' will never match; it will see `0` instead.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:03:21Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:03:20Z"
+          }
+        ]
       }
     ]
   },
@@ -31439,6 +32416,37 @@
             "created_at": "2026-03-18T20:27:11Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR9__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR9__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "lib/freedom_patches/translate_accelerator.rb",
+            "line": 69,
+            "body": "P3: _Note: Issue is at line 72, shown here on nearest changed line._\n\nThe LRU translation cache key is `\"#{key}#{config.locale}#{config.backend.object_id}\"` but does not include `SiteSetting.default_locale`, which is now part of the fallback chain. When an admin changes the site's default locale, cached translations that were resolved through the old fallback chain will continue to be served (up to 300 entries in the LRU). For example: user locale `:fr` missing a key falls back to `:en` (default) and caches \"Hello\". Admin then changes default to `:de`. The cache still returns \"Hello\" for that key instead of trying `:de` \u2192 \"Hallo\". There is no `I18n.reload!` triggered on `default_locale` change to clear the cache.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>lib/freedom_patches/translate_accelerator.rb</file_path>\n<line_numbers>69</line_numbers>\n\n<comment>_Note: Issue is at line 72, shown here on nearest changed line._\n\nThe LRU translation cache key is `\"#{key}#{config.locale}#{config.backend.object_id}\"` but does not include `SiteSetting.default_locale`, which is now part of the fallback chain. When an admin changes the site's default locale, cached translations that were resolved through the old fallback chain will continue to be served (up to 300 entries in the LRU). For example: user locale `:fr` missing a key falls back to `:en` (default) and caches \"Hello\". Admin then changes default to `:de`. The cache still returns \"Hello\" for that key instead of trying `:de` \u2192 \"Hallo\". There is no `I18n.reload!` triggered on `default_locale` change to clear the cache.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>lib/freedom_patches/translate_accelerator.rb</file_path>\n<line_numbers>69</line_numbers>\n\n<comment>_Note: Issue is at line 72, shown here on nearest changed line._\n\nThe LRU translation cache key is `\"#{key}#{config.locale}#{config.backend.object_id}\"` but does not include `SiteSetting.default_locale`, which is now part of the fallback chain. When an admin changes the site's default locale, cached translations that were resolved through the old fallback chain will continue to be served (up to 300 entries in the LRU). For example: user locale `:fr` missing a key falls back to `:en` (default) and caches \"Hello\". Admin then changes default to `:de`. The cache still returns \"Hello\" for that key instead of trying `:de` \u2192 \"Hallo\". There is no `I18n.reload!` triggered on `default_locale` change to clear the cache.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:28:04Z"
+          },
+          {
+            "path": "lib/freedom_patches/translate_accelerator.rb",
+            "line": 68,
+            "body": "P3: The `translate` method only loads `config.locale` (line 68), but with `I18n::Backend::Fallbacks` now included in the backend, the backend's `translate` iterates through the fallback chain (`[locale, default_locale, :en]`) and calls `lookup` for each. If fallback locale YML files haven't been loaded, lookups return nil and fallbacks silently fail. The `ensure_loaded!` call in `set_locale` handles web requests, but any code path outside a controller (Sidekiq background jobs, rake tasks, console) will only have the current locale loaded \u2014 the fallback feature introduced by this PR is silently non-functional there.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>lib/freedom_patches/translate_accelerator.rb</file_path>\n<line_numbers>68</line_numbers>\n\n<comment>The `translate` method only loads `config.locale` (line 68), but with `I18n::Backend::Fallbacks` now included in the backend, the backend's `translate` iterates through the fallback chain (`[locale, default_locale, :en]`) and calls `lookup` for each. If fallback locale YML files haven't been loaded, lookups return nil and fallbacks silently fail. The `ensure_loaded!` call in `set_locale` handles web requests, but any code path outside a controller (Sidekiq background jobs, rake tasks, console) will only have the current locale loaded \u2014 the fallback feature introduced by this PR is silently non-functional there.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>lib/freedom_patches/translate_accelerator.rb</file_path>\n<line_numbers>68</line_numbers>\n\n<comment>The `translate` method only loads `config.locale` (line 68), but with `I18n::Backend::Fallbacks` now included in the backend, the backend's `translate` iterates through the fallback chain (`[locale, default_locale, :en]`) and calls `lookup` for each. If fallback locale YML files haven't been loaded, lookups return nil and fallbacks silently fail. The `ensure_loaded!` call in `set_locale` handles web requests, but any code path outside a controller (Sidekiq background jobs, rake tasks, console) will only have the current locale loaded \u2014 the fallback feature introduced by this PR is silently non-functional there.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:28:04Z"
+          },
+          {
+            "path": "lib/freedom_patches/translate_accelerator.rb",
+            "line": 63,
+            "body": "P4: The `ensure_loaded!` method initializes `@loaded_locales ||= []` outside the `LOAD_MUTEX`. In a multi-threaded server (Puma), two threads hitting `ensure_loaded!` concurrently before any locale is loaded could both evaluate `@loaded_locales` as nil and each assign a new `[]`. One thread's array (and any locales appended to it inside `load_locale`) would be silently discarded when the second thread overwrites `@loaded_locales`. This could cause a locale to appear unloaded and trigger redundant `load_locale` calls, or in an unlucky interleaving with `translate` (line 68), skip loading a locale entirely because the check `@loaded_locales.include?` reads a stale array reference.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>lib/freedom_patches/translate_accelerator.rb</file_path>\n<line_numbers>63</line_numbers>\n\n<comment>The `ensure_loaded!` method initializes `@loaded_locales ||= []` outside the `LOAD_MUTEX`. In a multi-threaded server (Puma), two threads hitting `ensure_loaded!` concurrently before any locale is loaded could both evaluate `@loaded_locales` as nil and each assign a new `[]`. One thread's array (and any locales appended to it inside `load_locale`) would be silently discarded when the second thread overwrites `@loaded_locales`. This could cause a locale to appear unloaded and trigger redundant `load_locale` calls, or in an unlucky interleaving with `translate` (line 68), skip loading a locale entirely because the check `@loaded_locales.include?` reads a stale array reference.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>lib/freedom_patches/translate_accelerator.rb</file_path>\n<line_numbers>63</line_numbers>\n\n<comment>The `ensure_loaded!` method initializes `@loaded_locales ||= []` outside the `LOAD_MUTEX`. In a multi-threaded server (Puma), two threads hitting `ensure_loaded!` concurrently before any locale is loaded could both evaluate `@loaded_locales` as nil and each assign a new `[]`. One thread's array (and any locales appended to it inside `load_locale`) would be silently discarded when the second thread overwrites `@loaded_locales`. This could cause a locale to appear unloaded and trigger redundant `load_locale` calls, or in an unlucky interleaving with `translate` (line 68), skip loading a locale entirely because the check `@loaded_locales.include?` reads a stale array reference.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:28:04Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:28:04Z"
+          }
+        ]
       }
     ]
   },
@@ -32326,6 +33334,31 @@
             "created_at": "2026-03-18T20:27:31Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "grafana__grafana__tusk__PR76186__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR76186__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "pkg/plugins/log/fake.go",
+            "line": 46,
+            "body": "P3: TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(). Since LoggerMiddleware.logRequest now calls m.logger.FromContext(ctx).Info(...), any test using TestLogger to verify logged output will silently lose all log messages \u2014 the .Info() call goes to a throwaway instance that nobody holds a reference to. Unlike the production grafanaInfraLogWrapper.FromContext (which correctly returns a logger enriched with contextual attributes), this fake makes it impossible to test the LoggerMiddleware logging behavior. The method should return the receiver (f) to preserve log capture, consistent with how the real implementation preserves the logger identity.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/plugins/log/fake.go</file_path>\n<line_numbers>46</line_numbers>\n\n<comment>TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(). Since LoggerMiddleware.logRequest now calls m.logger.FromContext(ctx).Info(...), any test using TestLogger to verify logged output will silently lose all log messages \u2014 the .Info() call goes to a throwaway instance that nobody holds a reference to. Unlike the production grafanaInfraLogWrapper.FromContext (which correctly returns a logger enriched with contextual attributes), this fake makes it impossible to test the LoggerMiddleware logging behavior. The method should return the receiver (f) to preserve log capture, consistent with how the real implementation preserves the logger identity.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/plugins/log/fake.go</file_path>\n<line_numbers>46</line_numbers>\n\n<comment>TestLogger.FromContext returns a brand new disconnected TestLogger instance via NewTestLogger(). Since LoggerMiddleware.logRequest now calls m.logger.FromContext(ctx).Info(...), any test using TestLogger to verify logged output will silently lose all log messages \u2014 the .Info() call goes to a throwaway instance that nobody holds a reference to. Unlike the production grafanaInfraLogWrapper.FromContext (which correctly returns a logger enriched with contextual attributes), this fake makes it impossible to test the LoggerMiddleware logging behavior. The method should return the receiver (f) to preserve log capture, consistent with how the real implementation preserves the logger identity.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:05:13Z"
+          },
+          {
+            "path": "pkg/services/pluginsintegration/clientmiddleware/contextual_logger_middleware.go",
+            "line": 39,
+            "body": "P4: ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic. While the Decorator currently prevents nil requests from reaching middlewares, the LoggerMiddleware explicitly handles nil requests with guard clauses (e.g., `if req == nil { return m.next.QueryData(ctx, req) }`). This inconsistency means the new middleware is less defensive than the existing pattern in the same package. If the middleware is ever used outside the Decorator (e.g., in tests), it will panic.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>pkg/services/pluginsintegration/clientmiddleware/contextual_logger_middleware.go</file_path>\n<line_numbers>39</line_numbers>\n\n<comment>ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic. While the Decorator currently prevents nil requests from reaching middlewares, the LoggerMiddleware explicitly handles nil requests with guard clauses (e.g., `if req == nil { return m.next.QueryData(ctx, req) }`). This inconsistency means the new middleware is less defensive than the existing pattern in the same package. If the middleware is ever used outside the Decorator (e.g., in tests), it will panic.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>pkg/services/pluginsintegration/clientmiddleware/contextual_logger_middleware.go</file_path>\n<line_numbers>39</line_numbers>\n\n<comment>ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) access req.PluginContext without nil-checking req, which would cause a nil pointer dereference panic. While the Decorator currently prevents nil requests from reaching middlewares, the LoggerMiddleware explicitly handles nil requests with guard clauses (e.g., `if req == nil { return m.next.QueryData(ctx, req) }`). This inconsistency means the new middleware is less defensive than the existing pattern in the same package. If the middleware is ever used outside the Decorator (e.g., in tests), it will panic.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:05:13Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 2 issues",
+            "created_at": "2026-04-12T02:05:13Z"
+          }
+        ]
       }
     ]
   },
@@ -33826,6 +34859,55 @@
             "created_at": "2026-03-18T20:27:07Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR10__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR10__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "db/migrate/20150818190757_create_embeddable_hosts.rb",
+            "line": 11,
+            "body": "P1: Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist (e.g., fresh installs or installs that never configured embedding), the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>db/migrate/20150818190757_create_embeddable_hosts.rb</file_path>\n<line_numbers>11</line_numbers>\n\n<comment>Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist (e.g., fresh installs or installs that never configured embedding), the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>db/migrate/20150818190757_create_embeddable_hosts.rb</file_path>\n<line_numbers>11</line_numbers>\n\n<comment>Migration crashes on fresh installs. If the 'embed_category' site setting doesn't exist (e.g., fresh installs or installs that never configured embedding), the SQL query returns 0 rows. Accessing `[0]` returns nil, then `['id']` raises NoMethodError. The fallback `if category_id == 0` is never reached.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:25:12Z"
+          },
+          {
+            "path": "db/migrate/20150818190757_create_embeddable_hosts.rb",
+            "line": 25,
+            "body": "P1: SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote (e.g., from a corrupted setting) will break the SQL or allow injection.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>db/migrate/20150818190757_create_embeddable_hosts.rb</file_path>\n<line_numbers>25</line_numbers>\n\n<comment>SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote (e.g., from a corrupted setting) will break the SQL or allow injection.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>db/migrate/20150818190757_create_embeddable_hosts.rb</file_path>\n<line_numbers>25</line_numbers>\n\n<comment>SQL injection in migration. Host values from the old `embeddable_hosts` site setting are interpolated directly into a SQL INSERT via string interpolation (`'#{h}'`). A host containing a single quote (e.g., from a corrupted setting) will break the SQL or allow injection.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:25:12Z"
+          },
+          {
+            "path": "app/controllers/admin/embeddable_hosts_controller.rb",
+            "line": 15,
+            "body": "P2: Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist (e.g., already deleted, invalid id). In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/controllers/admin/embeddable_hosts_controller.rb</file_path>\n<line_numbers>15</line_numbers>\n\n<comment>Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist (e.g., already deleted, invalid id). In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/controllers/admin/embeddable_hosts_controller.rb</file_path>\n<line_numbers>15</line_numbers>\n\n<comment>Null pointer crash in `destroy` and `update` actions. `EmbeddableHost.where(id: params[:id]).first` returns nil if the record doesn't exist (e.g., already deleted, invalid id). In `destroy`, `host.destroy` will raise NoMethodError on nil. In `update`, `save_host(host)` will crash at `host.host = ...`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:25:12Z"
+          },
+          {
+            "path": "app/models/embeddable_host.rb",
+            "line": 6,
+            "body": "P2: `before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value (host is nil), `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it. The controller's `save_host` could pass nil if `params[:embeddable_host][:host]` is missing.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/models/embeddable_host.rb</file_path>\n<line_numbers>6</line_numbers>\n\n<comment>`before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value (host is nil), `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it. The controller's `save_host` could pass nil if `params[:embeddable_host][:host]` is missing.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/models/embeddable_host.rb</file_path>\n<line_numbers>6</line_numbers>\n\n<comment>`before_validation` crashes when `host` is nil. If an `EmbeddableHost` is created without setting a host value (host is nil), `self.host.sub!` raises `NoMethodError: undefined method 'sub!' for nil:NilClass` before validation can reject it. The controller's `save_host` could pass nil if `params[:embeddable_host][:host]` is missing.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:25:12Z"
+          },
+          {
+            "path": "spec/fabricators/category_fabricator.rb",
+            "line": 1,
+            "body": "P3: Fabricator file contents are swapped. `category_fabricator.rb` now contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators (`:category`, `:diff_category`, `:happy_category`, `:private_category`). The file names don't match their contents. While Fabricator loads by name not filename so tests may still pass, this will confuse every developer who looks for a fabricator by filename.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>spec/fabricators/category_fabricator.rb</file_path>\n<line_numbers>1</line_numbers>\n\n<comment>Fabricator file contents are swapped. `category_fabricator.rb` now contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators (`:category`, `:diff_category`, `:happy_category`, `:private_category`). The file names don't match their contents. While Fabricator loads by name not filename so tests may still pass, this will confuse every developer who looks for a fabricator by filename.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>spec/fabricators/category_fabricator.rb</file_path>\n<line_numbers>1</line_numbers>\n\n<comment>Fabricator file contents are swapped. `category_fabricator.rb` now contains the `Fabricator(:embeddable_host)` definition, while `embeddable_host_fabricator.rb` contains all the category fabricators (`:category`, `:diff_category`, `:happy_category`, `:private_category`). The file names don't match their contents. While Fabricator loads by name not filename so tests may still pass, this will confuse every developer who looks for a fabricator by filename.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:25:12Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/adapters/rest.js.es6",
+            "line": 22,
+            "body": "P4: JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores (e.g., `'some_admin_model'` becomes `'some-admin_model'` instead of `'some-admin-model'`).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/adapters/rest.js.es6</file_path>\n<line_numbers>22</line_numbers>\n\n<comment>JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores (e.g., `'some_admin_model'` becomes `'some-admin_model'` instead of `'some-admin-model'`).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/adapters/rest.js.es6</file_path>\n<line_numbers>22</line_numbers>\n\n<comment>JavaScript `String.replace` with a string argument only replaces the first occurrence. `type.replace('_', '-')` in `basePath` will fail to match ADMIN_MODELS for any future type with multiple underscores (e.g., `'some_admin_model'` becomes `'some-admin_model'` instead of `'some-admin-model'`).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:25:12Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 6 issues",
+            "created_at": "2026-04-12T02:25:11Z"
+          }
+        ]
       }
     ]
   },
@@ -34896,6 +35978,43 @@
             "created_at": "2026-03-18T20:36:43Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR7__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR7__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/assets/stylesheets/desktop/topic-post.scss",
+            "line": 291,
+            "body": "P3: Light-theme lightness value changed from 30% to 70% (swapped with dark value). The old code was `scale-color($primary, $lightness: 30%)` producing a relatively dark color for `.embedded-posts` link text. The new code uses 70% for the light theme, making it much lighter/more washed out. The dark-theme and light-theme values appear to be accidentally swapped.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/desktop/topic-post.scss</file_path>\n<line_numbers>291</line_numbers>\n\n<comment>Light-theme lightness value changed from 30% to 70% (swapped with dark value). The old code was `scale-color($primary, $lightness: 30%)` producing a relatively dark color for `.embedded-posts` link text. The new code uses 70% for the light theme, making it much lighter/more washed out. The dark-theme and light-theme values appear to be accidentally swapped.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/desktop/topic-post.scss</file_path>\n<line_numbers>291</line_numbers>\n\n<comment>Light-theme lightness value changed from 30% to 70% (swapped with dark value). The old code was `scale-color($primary, $lightness: 30%)` producing a relatively dark color for `.embedded-posts` link text. The new code uses 70% for the light theme, making it much lighter/more washed out. The dark-theme and light-theme values appear to be accidentally swapped.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:27:12Z"
+          },
+          {
+            "path": "app/assets/stylesheets/mobile/modal.scss",
+            "line": 102,
+            "body": "P3: Light-theme lightness value changed from 70% to 30% \u2014 opposite of the desktop version. Desktop `modal.scss` correctly wraps the same `.custom-message-length` as `dark-light-choose(...70%..., ...30%...)`, but the mobile version swaps them to `dark-light-choose(...30%..., ...70%...)`. This makes the character count text much darker on mobile light themes and much lighter on mobile dark themes \u2014 the inverse of what's intended.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/mobile/modal.scss</file_path>\n<line_numbers>102</line_numbers>\n\n<comment>Light-theme lightness value changed from 70% to 30% \u2014 opposite of the desktop version. Desktop `modal.scss` correctly wraps the same `.custom-message-length` as `dark-light-choose(...70%..., ...30%...)`, but the mobile version swaps them to `dark-light-choose(...30%..., ...70%...)`. This makes the character count text much darker on mobile light themes and much lighter on mobile dark themes \u2014 the inverse of what's intended.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/mobile/modal.scss</file_path>\n<line_numbers>102</line_numbers>\n\n<comment>Light-theme lightness value changed from 70% to 30% \u2014 opposite of the desktop version. Desktop `modal.scss` correctly wraps the same `.custom-message-length` as `dark-light-choose(...70%..., ...30%...)`, but the mobile version swaps them to `dark-light-choose(...30%..., ...70%...)`. This makes the character count text much darker on mobile light themes and much lighter on mobile dark themes \u2014 the inverse of what's intended.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:27:12Z"
+          },
+          {
+            "path": "app/assets/stylesheets/mobile/topic-post.scss",
+            "line": 182,
+            "body": "P3: Light-theme lightness for `h3` changed from 20% to 50%. The old code used `$lightness: 20%` which produces a near-black heading. The new light-theme value of 50% makes headings significantly more faded/grey, which is a visual regression on light themes. The dark-theme value should have been 80% (complementary to 20%), not 50%.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/mobile/topic-post.scss</file_path>\n<line_numbers>182</line_numbers>\n\n<comment>Light-theme lightness for `h3` changed from 20% to 50%. The old code used `$lightness: 20%` which produces a near-black heading. The new light-theme value of 50% makes headings significantly more faded/grey, which is a visual regression on light themes. The dark-theme value should have been 80% (complementary to 20%), not 50%.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/mobile/topic-post.scss</file_path>\n<line_numbers>182</line_numbers>\n\n<comment>Light-theme lightness for `h3` changed from 20% to 50%. The old code used `$lightness: 20%` which produces a near-black heading. The new light-theme value of 50% makes headings significantly more faded/grey, which is a visual regression on light themes. The dark-theme value should have been 80% (complementary to 20%), not 50%.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:27:12Z"
+          },
+          {
+            "path": "app/assets/stylesheets/desktop/user.scss",
+            "line": 522,
+            "body": "P4: Light-theme lightness for `.name` changed from 30% to 50% in both desktop and mobile user stylesheets. The old code used `$lightness: 30%` producing a fairly dark name. The new light-theme value of 50% makes user names noticeably lighter/more washed out. This appears to be an intentional-looking but inconsistent change \u2014 the nearby `.username a` in common/base/user.scss correctly preserves 30% for light theme.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/desktop/user.scss</file_path>\n<line_numbers>522</line_numbers>\n\n<comment>Light-theme lightness for `.name` changed from 30% to 50% in both desktop and mobile user stylesheets. The old code used `$lightness: 30%` producing a fairly dark name. The new light-theme value of 50% makes user names noticeably lighter/more washed out. This appears to be an intentional-looking but inconsistent change \u2014 the nearby `.username a` in common/base/user.scss correctly preserves 30% for light theme.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/desktop/user.scss</file_path>\n<line_numbers>522</line_numbers>\n\n<comment>Light-theme lightness for `.name` changed from 30% to 50% in both desktop and mobile user stylesheets. The old code used `$lightness: 30%` producing a fairly dark name. The new light-theme value of 50% makes user names noticeably lighter/more washed out. This appears to be an intentional-looking but inconsistent change \u2014 the nearby `.username a` in common/base/user.scss correctly preserves 30% for light theme.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:27:12Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:27:11Z"
+          }
+        ]
       }
     ]
   },
@@ -36309,6 +37428,49 @@
             "created_at": "2026-03-18T20:27:14Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR8__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR8__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/controllers/groups_controller.rb",
+            "line": 28,
+            "body": "P2: Breaking change in the public `GroupsController#members` endpoint response format. The old code used `render_serialized(members.to_a, GroupUserSerializer)` which returned a flat JSON array. The new code returns `{ members: [...], meta: { total, limit, offset } }`. Any existing consumer of this public (non-admin) endpoint expecting a flat array will break. The old code also had no limit/offset for non-automatic groups (returning all members); the new code always paginates with a default limit of 50, so large groups will silently return truncated results to existing callers that don't pass pagination params.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/controllers/groups_controller.rb</file_path>\n<line_numbers>28</line_numbers>\n\n<comment>Breaking change in the public `GroupsController#members` endpoint response format. The old code used `render_serialized(members.to_a, GroupUserSerializer)` which returned a flat JSON array. The new code returns `{ members: [...], meta: { total, limit, offset } }`. Any existing consumer of this public (non-admin) endpoint expecting a flat array will break. The old code also had no limit/offset for non-automatic groups (returning all members); the new code always paginates with a default limit of 50, so large groups will silently return truncated results to existing callers that don't pass pagination params.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/controllers/groups_controller.rb</file_path>\n<line_numbers>28</line_numbers>\n\n<comment>Breaking change in the public `GroupsController#members` endpoint response format. The old code used `render_serialized(members.to_a, GroupUserSerializer)` which returned a flat JSON array. The new code returns `{ members: [...], meta: { total, limit, offset } }`. Any existing consumer of this public (non-admin) endpoint expecting a flat array will break. The old code also had no limit/offset for non-automatic groups (returning all members); the new code always paginates with a default limit of 50, so large groups will silently return truncated results to existing callers that don't pass pagination params.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:54Z"
+          },
+          {
+            "path": "app/assets/javascripts/admin/controllers/admin-group.js.es6",
+            "line": 13,
+            "body": "P3: Off-by-one in `totalPages` calculation. `Math.floor(user_count / limit) + 1` yields one extra page when user_count is exactly divisible by limit. For example, 100 users with limit 50 gives `Math.floor(100/50) + 1 = 3` pages instead of the correct 2. This allows the user to paginate to an empty final page. Should use `Math.ceil(user_count / limit)`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/admin/controllers/admin-group.js.es6</file_path>\n<line_numbers>13</line_numbers>\n\n<comment>Off-by-one in `totalPages` calculation. `Math.floor(user_count / limit) + 1` yields one extra page when user_count is exactly divisible by limit. For example, 100 users with limit 50 gives `Math.floor(100/50) + 1 = 3` pages instead of the correct 2. This allows the user to paginate to an empty final page. Should use `Math.ceil(user_count / limit)`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/admin/controllers/admin-group.js.es6</file_path>\n<line_numbers>13</line_numbers>\n\n<comment>Off-by-one in `totalPages` calculation. `Math.floor(user_count / limit) + 1` yields one extra page when user_count is exactly divisible by limit. For example, 100 users with limit 50 gives `Math.floor(100/50) + 1 = 3` pages instead of the correct 2. This allows the user to paginate to an empty final page. Should use `Math.ceil(user_count / limit)`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:54Z"
+          },
+          {
+            "path": "spec/controllers/admin/groups_controller_spec.rb",
+            "line": 115,
+            "body": "P3: Test for 'cannot remove members from automatic groups' uses `xhr :put` but the route is defined as `delete \"members\" => \"groups#remove_member\"`. In RSpec controller specs the action is dispatched directly so the test passes, but it doesn't validate the actual HTTP method. More importantly, the test is not exercising the real request path that clients will use (DELETE), so route misconfiguration bugs would go undetected.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>spec/controllers/admin/groups_controller_spec.rb</file_path>\n<line_numbers>115</line_numbers>\n\n<comment>Test for 'cannot remove members from automatic groups' uses `xhr :put` but the route is defined as `delete \"members\" => \"groups#remove_member\"`. In RSpec controller specs the action is dispatched directly so the test passes, but it doesn't validate the actual HTTP method. More importantly, the test is not exercising the real request path that clients will use (DELETE), so route misconfiguration bugs would go undetected.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>spec/controllers/admin/groups_controller_spec.rb</file_path>\n<line_numbers>115</line_numbers>\n\n<comment>Test for 'cannot remove members from automatic groups' uses `xhr :put` but the route is defined as `delete \"members\" => \"groups#remove_member\"`. In RSpec controller specs the action is dispatched directly so the test passes, but it doesn't validate the actual HTTP method. More importantly, the test is not exercising the real request path that clients will use (DELETE), so route misconfiguration bugs would go undetected.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:54Z"
+          },
+          {
+            "path": "app/controllers/admin/groups_controller.rb",
+            "line": 38,
+            "body": "P3: In `update`, `group.visible = params[:visible] == \"true\"` is executed unconditionally. If a caller sends only `alias_level` or `name` without a `visible` parameter, `params[:visible]` is nil and `visible` gets set to false. The old code had the same line but read from a nested `params[:group]` hash, so clients always sent the full group payload. The new flattened params make it easy for partial updates to accidentally unset `visible`. The JS `save()` always sends all fields via `asJSON()` so the admin UI works, but any other caller (API consumers, scripts) doing a partial PUT will reset visibility.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/controllers/admin/groups_controller.rb</file_path>\n<line_numbers>38</line_numbers>\n\n<comment>In `update`, `group.visible = params[:visible] == \"true\"` is executed unconditionally. If a caller sends only `alias_level` or `name` without a `visible` parameter, `params[:visible]` is nil and `visible` gets set to false. The old code had the same line but read from a nested `params[:group]` hash, so clients always sent the full group payload. The new flattened params make it easy for partial updates to accidentally unset `visible`. The JS `save()` always sends all fields via `asJSON()` so the admin UI works, but any other caller (API consumers, scripts) doing a partial PUT will reset visibility.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/controllers/admin/groups_controller.rb</file_path>\n<line_numbers>38</line_numbers>\n\n<comment>In `update`, `group.visible = params[:visible] == \"true\"` is executed unconditionally. If a caller sends only `alias_level` or `name` without a `visible` parameter, `params[:visible]` is nil and `visible` gets set to false. The old code had the same line but read from a nested `params[:group]` hash, so clients always sent the full group payload. The new flattened params make it easy for partial updates to accidentally unset `visible`. The JS `save()` always sends all fields via `asJSON()` so the admin UI works, but any other caller (API consumers, scripts) doing a partial PUT will reset visibility.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:54Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/models/group.js",
+            "line": 66,
+            "body": "P4: The `create` action reads `params[:name]` directly, but `asJSON()` in the JS Group model sends `visible: !!this.get('visible')` which will be the boolean `true` or `false`. When sent as form data, Rails converts booleans to strings, so `params[:visible] == \"true\"` works. However, `asJSON()` sends `visible: false` for a new group (default), and the `create` action sets `group.visible = params[:visible] == \"true\"`, so new groups will always be created with `visible: false` regardless of the checkbox state, because `!!this.get('visible')` evaluates `false` for a new group where `visible` is undefined/null. The `visible` property is never initialized to `true` in the Group model defaults.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/models/group.js</file_path>\n<line_numbers>66</line_numbers>\n\n<comment>The `create` action reads `params[:name]` directly, but `asJSON()` in the JS Group model sends `visible: !!this.get('visible')` which will be the boolean `true` or `false`. When sent as form data, Rails converts booleans to strings, so `params[:visible] == \"true\"` works. However, `asJSON()` sends `visible: false` for a new group (default), and the `create` action sets `group.visible = params[:visible] == \"true\"`, so new groups will always be created with `visible: false` regardless of the checkbox state, because `!!this.get('visible')` evaluates `false` for a new group where `visible` is undefined/null. The `visible` property is never initialized to `true` in the Group model defaults.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/models/group.js</file_path>\n<line_numbers>66</line_numbers>\n\n<comment>The `create` action reads `params[:name]` directly, but `asJSON()` in the JS Group model sends `visible: !!this.get('visible')` which will be the boolean `true` or `false`. When sent as form data, Rails converts booleans to strings, so `params[:visible] == \"true\"` works. However, `asJSON()` sends `visible: false` for a new group (default), and the `create` action sets `group.visible = params[:visible] == \"true\"`, so new groups will always be created with `visible: false` regardless of the checkbox state, because `!!this.get('visible')` evaluates `false` for a new group where `visible` is undefined/null. The `visible` property is never initialized to `true` in the Group model defaults.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:54Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:29:54Z"
+          }
+        ]
       }
     ]
   },
@@ -37379,6 +38541,43 @@
             "created_at": "2026-03-18T20:27:36Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR3__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR3__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/models/blocked_email.rb",
+            "line": 14,
+            "body": "P3: BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation. In Rails, validation can run multiple times per user action (e.g., calling `valid?` before `save`, or re-rendering a form). Each call increments match_count, inflating the blocked-email statistics. Furthermore, `record.save` inside `should_block?` triggers BlockedEmail's own validations (including a uniqueness DB query), making validation unnecessarily expensive.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/models/blocked_email.rb</file_path>\n<line_numbers>14</line_numbers>\n\n<comment>BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation. In Rails, validation can run multiple times per user action (e.g., calling `valid?` before `save`, or re-rendering a form). Each call increments match_count, inflating the blocked-email statistics. Furthermore, `record.save` inside `should_block?` triggers BlockedEmail's own validations (including a uniqueness DB query), making validation unnecessarily expensive.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/models/blocked_email.rb</file_path>\n<line_numbers>14</line_numbers>\n\n<comment>BlockedEmail.should_block? performs a DB write (incrementing match_count and updating last_match_at) as a side effect of model validation. In Rails, validation can run multiple times per user action (e.g., calling `valid?` before `save`, or re-rendering a form). Each call increments match_count, inflating the blocked-email statistics. Furthermore, `record.save` inside `should_block?` triggers BlockedEmail's own validations (including a uniqueness DB query), making validation unnecessarily expensive.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:07Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/controllers/create_account_controller.js",
+            "line": 274,
+            "body": "P3: The error response from `users_controller.rb` now includes `errors: user.errors.to_hash` and `values: user.attributes.slice(...)`. The JS code checks `result.errors.email` to decide whether to add the email to `rejectedEmails`. However, `user.errors.to_hash` returns errors keyed by attribute symbol (in Rails), serialized as JSON string keys like `{\"email\": [\"is not allowed.\"]}`. The JS code adds the email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures. This means if a user tries to register with an already-taken email address, that email gets permanently marked as 'rejected' in the client \u2014 even though it's a valid email that belongs to someone else. The user would need to refresh the page to retry with that email after, say, using password reset.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/controllers/create_account_controller.js</file_path>\n<line_numbers>274</line_numbers>\n\n<comment>The error response from `users_controller.rb` now includes `errors: user.errors.to_hash` and `values: user.attributes.slice(...)`. The JS code checks `result.errors.email` to decide whether to add the email to `rejectedEmails`. However, `user.errors.to_hash` returns errors keyed by attribute symbol (in Rails), serialized as JSON string keys like `{\"email\": [\"is not allowed.\"]}`. The JS code adds the email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures. This means if a user tries to register with an already-taken email address, that email gets permanently marked as 'rejected' in the client \u2014 even though it's a valid email that belongs to someone else. The user would need to refresh the page to retry with that email after, say, using password reset.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/controllers/create_account_controller.js</file_path>\n<line_numbers>274</line_numbers>\n\n<comment>The error response from `users_controller.rb` now includes `errors: user.errors.to_hash` and `values: user.attributes.slice(...)`. The JS code checks `result.errors.email` to decide whether to add the email to `rejectedEmails`. However, `user.errors.to_hash` returns errors keyed by attribute symbol (in Rails), serialized as JSON string keys like `{\"email\": [\"is not allowed.\"]}`. The JS code adds the email to rejectedEmails for ANY email validation error (not just blocked-email errors), including uniqueness failures. This means if a user tries to register with an already-taken email address, that email gets permanently marked as 'rejected' in the client \u2014 even though it's a valid email that belongs to someone else. The user would need to refresh the page to retry with that email after, say, using password reset.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:07Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/controllers/create_account_controller.js",
+            "line": 17,
+            "body": "P4: rejectedEmails is initialized as `Em.A([])` directly in the class definition hash passed to `.extend()`. In Ember, reference-type values (arrays, objects) defined this way are shared across all instances of the class \u2014 they are prototype properties, not instance properties. If CreateAccountController is ever instantiated more than once (or in tests), all instances share the same rejectedEmails array, causing cross-instance state pollution.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/controllers/create_account_controller.js</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>rejectedEmails is initialized as `Em.A([])` directly in the class definition hash passed to `.extend()`. In Ember, reference-type values (arrays, objects) defined this way are shared across all instances of the class \u2014 they are prototype properties, not instance properties. If CreateAccountController is ever instantiated more than once (or in tests), all instances share the same rejectedEmails array, causing cross-instance state pollution.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/controllers/create_account_controller.js</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>rejectedEmails is initialized as `Em.A([])` directly in the class definition hash passed to `.extend()`. In Ember, reference-type values (arrays, objects) defined this way are shared across all instances of the class \u2014 they are prototype properties, not instance properties. If CreateAccountController is ever instantiated more than once (or in tests), all instances share the same rejectedEmails array, causing cross-instance state pollution.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:07Z"
+          },
+          {
+            "path": "app/models/blocked_email.rb",
+            "line": 16,
+            "body": "P4: BlockedEmail.should_block? calls `record.save` but doesn't check the return value. If save fails (e.g., a concurrent process deletes the record between the `where` and `save`, or a future validation is added to BlockedEmail), the method silently ignores the failure and still returns the blocking decision based on stale in-memory state. More critically, after a failed `save`, the in-memory `record.match_count` is already incremented but not persisted, so the return value `record.action_type == actions[:block]` operates on a dirty object.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/models/blocked_email.rb</file_path>\n<line_numbers>16</line_numbers>\n\n<comment>BlockedEmail.should_block? calls `record.save` but doesn't check the return value. If save fails (e.g., a concurrent process deletes the record between the `where` and `save`, or a future validation is added to BlockedEmail), the method silently ignores the failure and still returns the blocking decision based on stale in-memory state. More critically, after a failed `save`, the in-memory `record.match_count` is already incremented but not persisted, so the return value `record.action_type == actions[:block]` operates on a dirty object.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/models/blocked_email.rb</file_path>\n<line_numbers>16</line_numbers>\n\n<comment>BlockedEmail.should_block? calls `record.save` but doesn't check the return value. If save fails (e.g., a concurrent process deletes the record between the `where` and `save`, or a future validation is added to BlockedEmail), the method silently ignores the failure and still returns the blocking decision based on stale in-memory state. More critically, after a failed `save`, the in-memory `record.match_count` is already incremented but not persisted, so the return value `record.action_type == actions[:block]` operates on a dirty object.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:07Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:32:07Z"
+          }
+        ]
       }
     ]
   },
@@ -38356,6 +39555,43 @@
             "created_at": "2026-03-18T20:27:26Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR5__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR5__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/assets/stylesheets/common/foundation/mixins.scss",
+            "line": 126,
+            "body": "P3: The `order` mixin passes the raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values (default is 1), while the standard `order` property is 0-based (default is 0). The correct value for `-webkit-box-ordinal-group` should be `$val + 1`. Currently, if any element uses `@include order(1)`, it would get `-webkit-box-ordinal-group: 1` which is the same as the default \u2014 making it indistinguishable from unordered elements in old WebKit browsers, breaking the intended ordering.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/common/foundation/mixins.scss</file_path>\n<line_numbers>126</line_numbers>\n\n<comment>The `order` mixin passes the raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values (default is 1), while the standard `order` property is 0-based (default is 0). The correct value for `-webkit-box-ordinal-group` should be `$val + 1`. Currently, if any element uses `@include order(1)`, it would get `-webkit-box-ordinal-group: 1` which is the same as the default \u2014 making it indistinguishable from unordered elements in old WebKit browsers, breaking the intended ordering.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/common/foundation/mixins.scss</file_path>\n<line_numbers>126</line_numbers>\n\n<comment>The `order` mixin passes the raw `$val` to `-webkit-box-ordinal-group` and `-moz-box-ordinal-group`, but these legacy flexbox properties use 1-based values (default is 1), while the standard `order` property is 0-based (default is 0). The correct value for `-webkit-box-ordinal-group` should be `$val + 1`. Currently, if any element uses `@include order(1)`, it would get `-webkit-box-ordinal-group: 1` which is the same as the default \u2014 making it indistinguishable from unordered elements in old WebKit browsers, breaking the intended ordering.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:43Z"
+          },
+          {
+            "path": "app/assets/stylesheets/common/base/topic-post.scss",
+            "line": 280,
+            "body": "P3: In `.small-action`, the parent is now `display: flex` but child `.topic-avatar` still has `float: left` (line 270). While floats are technically ignored on flex items, this indicates an incomplete migration \u2014 the float should be removed for clarity and to avoid confusion. More importantly, the padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that previously created spacing from the avatar. With `1.5%` horizontal padding, the gap between avatar and description text will shrink significantly on narrow viewports and may overlap or appear too tight.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/common/base/topic-post.scss</file_path>\n<line_numbers>280</line_numbers>\n\n<comment>In `.small-action`, the parent is now `display: flex` but child `.topic-avatar` still has `float: left` (line 270). While floats are technically ignored on flex items, this indicates an incomplete migration \u2014 the float should be removed for clarity and to avoid confusion. More importantly, the padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that previously created spacing from the avatar. With `1.5%` horizontal padding, the gap between avatar and description text will shrink significantly on narrow viewports and may overlap or appear too tight.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/common/base/topic-post.scss</file_path>\n<line_numbers>280</line_numbers>\n\n<comment>In `.small-action`, the parent is now `display: flex` but child `.topic-avatar` still has `float: left` (line 270). While floats are technically ignored on flex items, this indicates an incomplete migration \u2014 the float should be removed for clarity and to avoid confusion. More importantly, the padding change on `.small-action-desc` from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed `4em` left-padding that previously created spacing from the avatar. With `1.5%` horizontal padding, the gap between avatar and description text will shrink significantly on narrow viewports and may overlap or appear too tight.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:43Z"
+          },
+          {
+            "path": "app/assets/stylesheets/common/foundation/mixins.scss",
+            "line": 121,
+            "body": "P4: `-ms-align-items` (line 121) is not a valid CSS property and will be silently ignored by IE10/11. The correct IE prefix for `align-items` is only `-ms-flex-align` (already on line 120). However, `-ms-flex-align` itself uses different value keywords: it expects `start`/`end` instead of `flex-start`/`flex-end`. While the current usages (`center`, `baseline`) happen to match, the mixin does not translate values, so future callers passing `flex-start` or `flex-end` would silently break in IE.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/common/foundation/mixins.scss</file_path>\n<line_numbers>121</line_numbers>\n\n<comment>`-ms-align-items` (line 121) is not a valid CSS property and will be silently ignored by IE10/11. The correct IE prefix for `align-items` is only `-ms-flex-align` (already on line 120). However, `-ms-flex-align` itself uses different value keywords: it expects `start`/`end` instead of `flex-start`/`flex-end`. While the current usages (`center`, `baseline`) happen to match, the mixin does not translate values, so future callers passing `flex-start` or `flex-end` would silently break in IE.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/common/foundation/mixins.scss</file_path>\n<line_numbers>121</line_numbers>\n\n<comment>`-ms-align-items` (line 121) is not a valid CSS property and will be silently ignored by IE10/11. The correct IE prefix for `align-items` is only `-ms-flex-align` (already on line 120). However, `-ms-flex-align` itself uses different value keywords: it expects `start`/`end` instead of `flex-start`/`flex-end`. While the current usages (`center`, `baseline`) happen to match, the mixin does not translate values, so future callers passing `flex-start` or `flex-end` would silently break in IE.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:43Z"
+          },
+          {
+            "path": "app/assets/stylesheets/common/base/topic.scss",
+            "line": 33,
+            "body": "P4: The `.badge-wrapper.bullet` rule in `topic.scss` had `margin-top: 5px` removed (lines 32-34 of the old file) to accommodate flexbox alignment. However, the same `.badge-wrapper.bullet` selector in `#topic-title .title-wrapper` (topic.scss line 12-14) retains `margin-top: 5px`. This means the bullet badge styling is now inconsistent between the header's `.extra-info-wrapper` context (no margin-top) and the `#topic-title` context (still has margin-top), which may not be intentional.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/stylesheets/common/base/topic.scss</file_path>\n<line_numbers>33</line_numbers>\n\n<comment>The `.badge-wrapper.bullet` rule in `topic.scss` had `margin-top: 5px` removed (lines 32-34 of the old file) to accommodate flexbox alignment. However, the same `.badge-wrapper.bullet` selector in `#topic-title .title-wrapper` (topic.scss line 12-14) retains `margin-top: 5px`. This means the bullet badge styling is now inconsistent between the header's `.extra-info-wrapper` context (no margin-top) and the `#topic-title` context (still has margin-top), which may not be intentional.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/stylesheets/common/base/topic.scss</file_path>\n<line_numbers>33</line_numbers>\n\n<comment>The `.badge-wrapper.bullet` rule in `topic.scss` had `margin-top: 5px` removed (lines 32-34 of the old file) to accommodate flexbox alignment. However, the same `.badge-wrapper.bullet` selector in `#topic-title .title-wrapper` (topic.scss line 12-14) retains `margin-top: 5px`. This means the bullet badge styling is now inconsistent between the header's `.extra-info-wrapper` context (no margin-top) and the `#topic-title` context (still has margin-top), which may not be intentional.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:30:43Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:30:43Z"
+          }
+        ]
       }
     ]
   },
@@ -39323,6 +40559,37 @@
             "created_at": "2026-03-18T20:27:20Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR6__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR6__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/serializers/user_serializer.rb",
+            "line": 153,
+            "body": "P2: Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized \u2014 even when `website` is nil/blank, defeating the intent of the guard. Every other conditional include method in this file uses `?` (e.g., `include_email?`, `include_stats?`, `include_card_image_badge_id?`).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/serializers/user_serializer.rb</file_path>\n<line_numbers>153</line_numbers>\n\n<comment>Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized \u2014 even when `website` is nil/blank, defeating the intent of the guard. Every other conditional include method in this file uses `?` (e.g., `include_email?`, `include_stats?`, `include_card_image_badge_id?`).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/serializers/user_serializer.rb</file_path>\n<line_numbers>153</line_numbers>\n\n<comment>Method `include_website_name` is missing the trailing `?`. ActiveModel::Serializer looks for `include_<attribute>?` (with question mark) to determine conditional inclusion. Without the `?`, this method is never called by the serializer, so `website_name` will always be serialized \u2014 even when `website` is nil/blank, defeating the intent of the guard. Every other conditional include method in this file uses `?` (e.g., `include_email?`, `include_stats?`, `include_card_image_badge_id?`).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:10Z"
+          },
+          {
+            "path": "app/serializers/user_serializer.rb",
+            "line": 43,
+            "body": "P3: `website_name` is not included in the `untrusted_attributes` list (lines 105-111), but `website` is. The `untrusted_attributes` mechanism hides attributes for TL0 users when viewed by anonymous users via `scope.restrict_user_fields?`. Since `website_name` bypasses this check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/serializers/user_serializer.rb</file_path>\n<line_numbers>43</line_numbers>\n\n<comment>`website_name` is not included in the `untrusted_attributes` list (lines 105-111), but `website` is. The `untrusted_attributes` mechanism hides attributes for TL0 users when viewed by anonymous users via `scope.restrict_user_fields?`. Since `website_name` bypasses this check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/serializers/user_serializer.rb</file_path>\n<line_numbers>43</line_numbers>\n\n<comment>`website_name` is not included in the `untrusted_attributes` list (lines 105-111), but `website` is. The `untrusted_attributes` mechanism hides attributes for TL0 users when viewed by anonymous users via `scope.restrict_user_fields?`. Since `website_name` bypasses this check and computes its value directly from `object.user_profile.website`, it will leak the user's website information for TL0 users even when `website` itself is correctly hidden.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:10Z"
+          },
+          {
+            "path": "spec/serializers/user_serializer_spec.rb",
+            "line": 78,
+            "body": "P4: The `websiteName` computed property was removed from the controller, and the template now reads `model.website_name` from the serializer. However, `json` in the test is computed via `serializer.as_json` with `let(:json)` using `let` (lazy, memoized). The `Discourse.stubs(:current_hostname)` call happens inside the `it` block, *after* `json` is already defined as a lazy let. If `json` were ever evaluated before the stub (e.g., by a shared example or `before` block), the stub would not be in effect. Currently this works by accident since nothing else triggers `json` before the stub, but the test structure is fragile \u2014 the stub should be in a `before` block before `json` is accessed.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>spec/serializers/user_serializer_spec.rb</file_path>\n<line_numbers>78</line_numbers>\n\n<comment>The `websiteName` computed property was removed from the controller, and the template now reads `model.website_name` from the serializer. However, `json` in the test is computed via `serializer.as_json` with `let(:json)` using `let` (lazy, memoized). The `Discourse.stubs(:current_hostname)` call happens inside the `it` block, *after* `json` is already defined as a lazy let. If `json` were ever evaluated before the stub (e.g., by a shared example or `before` block), the stub would not be in effect. Currently this works by accident since nothing else triggers `json` before the stub, but the test structure is fragile \u2014 the stub should be in a `before` block before `json` is accessed.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>spec/serializers/user_serializer_spec.rb</file_path>\n<line_numbers>78</line_numbers>\n\n<comment>The `websiteName` computed property was removed from the controller, and the template now reads `model.website_name` from the serializer. However, `json` in the test is computed via `serializer.as_json` with `let(:json)` using `let` (lazy, memoized). The `Discourse.stubs(:current_hostname)` call happens inside the `it` block, *after* `json` is already defined as a lazy let. If `json` were ever evaluated before the stub (e.g., by a shared example or `before` block), the stub would not be in effect. Currently this works by accident since nothing else triggers `json` before the stub, but the test structure is fragile \u2014 the stub should be in a `before` block before `json` is accessed.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:29:10Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 3 issues",
+            "created_at": "2026-04-12T02:29:10Z"
+          }
+        ]
       }
     ]
   },
@@ -41024,6 +42291,49 @@
             "created_at": "2026-03-18T20:27:32Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR4__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR4__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/views/embed/best.html.erb",
+            "line": 6,
+            "body": "P1: ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression (e.g., `end if condition`). This will raise a SyntaxError when the `best.html.erb` template is rendered. The controller spec doesn't catch this because RSpec controller specs don't render views by default.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/views/embed/best.html.erb</file_path>\n<line_numbers>6</line_numbers>\n\n<comment>ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression (e.g., `end if condition`). This will raise a SyntaxError when the `best.html.erb` template is rendered. The controller spec doesn't catch this because RSpec controller specs don't render views by default.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/views/embed/best.html.erb</file_path>\n<line_numbers>6</line_numbers>\n\n<comment>ERB syntax error: `end if` without a condition is invalid Ruby. Line 6 has `<%- end if %>` which should be `<%- end %>`. The `if` keyword after `end` requires a boolean expression (e.g., `end if condition`). This will raise a SyntaxError when the `best.html.erb` template is rendered. The controller spec doesn't catch this because RSpec controller specs don't render views by default.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:47Z"
+          },
+          {
+            "path": "app/views/layouts/embed.html.erb",
+            "line": 11,
+            "body": "P1: XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. The `<%= %>` ERB tag HTML-escapes the output, but HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/views/layouts/embed.html.erb</file_path>\n<line_numbers>11</line_numbers>\n\n<comment>XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. The `<%= %>` ERB tag HTML-escapes the output, but HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/views/layouts/embed.html.erb</file_path>\n<line_numbers>11</line_numbers>\n\n<comment>XSS via referer injection into JavaScript. `request.referer` is interpolated directly into a JavaScript string literal using single quotes. While the hostname is validated against `embeddable_host`, the path/query/fragment of the referer URL is not sanitized. A referer like `http://eviltrout.com/page?x='+alert(1)+'` would break out of the JS string and execute arbitrary JavaScript. The `<%= %>` ERB tag HTML-escapes the output, but HTML escaping does not prevent JS string breakout since single quotes are not HTML-escaped.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:47Z"
+          },
+          {
+            "path": "app/jobs/scheduled/poll_feed.rb",
+            "line": 35,
+            "body": "P2: In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`. SimpleRSS returns nil for missing fields.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/jobs/scheduled/poll_feed.rb</file_path>\n<line_numbers>35</line_numbers>\n\n<comment>In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`. SimpleRSS returns nil for missing fields.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/jobs/scheduled/poll_feed.rb</file_path>\n<line_numbers>35</line_numbers>\n\n<comment>In `poll_feed`, `i.content` can be nil for RSS items that only have a `description` field (not `content:encoded`). When `i.content` is nil, calling `.scrub` on nil raises `NoMethodError: undefined method 'scrub' for nil:NilClass`. SimpleRSS returns nil for missing fields.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:47Z"
+          },
+          {
+            "path": "app/assets/javascripts/embed.js",
+            "line": 17,
+            "body": "P3: The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a *substring* of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain (e.g., `http://evil.com` if discourseUrl is `http://evil.com.attacker.com/...`) to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/embed.js</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a *substring* of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain (e.g., `http://evil.com` if discourseUrl is `http://evil.com.attacker.com/...`) to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/embed.js</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>The postMessage origin validation uses `discourseUrl.indexOf(e.origin)` which checks if `e.origin` is a *substring* of `discourseUrl`. This allows an attacker at a domain that is a prefix of the legitimate domain (e.g., `http://evil.com` if discourseUrl is `http://evil.com.attacker.com/...`) to send messages that pass the check. The correct check should verify that `discourseUrl` starts with `e.origin + '/'` or compare parsed origins.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:47Z"
+          },
+          {
+            "path": "app/models/topic_embed.rb",
+            "line": 13,
+            "body": "P3: `contents << \"\\n<hr>...\"` on line 13 uses the `<<` operator which mutates the string in place. This modifies the caller's original string. In `import_remote`, `doc.content` gets mutated. If `import` is called twice with the same string (e.g., in tests via `let(:contents)`), the imported-from footer is appended multiple times. More critically, when `import` is called for an update path (content_sha1 check on line 34), the SHA1 was computed *after* mutation, so re-importing the same content will always differ from the stored SHA1 (since the stored one includes one footer but the new computation includes two).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/models/topic_embed.rb</file_path>\n<line_numbers>13</line_numbers>\n\n<comment>`contents << \"\\n<hr>...\"` on line 13 uses the `<<` operator which mutates the string in place. This modifies the caller's original string. In `import_remote`, `doc.content` gets mutated. If `import` is called twice with the same string (e.g., in tests via `let(:contents)`), the imported-from footer is appended multiple times. More critically, when `import` is called for an update path (content_sha1 check on line 34), the SHA1 was computed *after* mutation, so re-importing the same content will always differ from the stored SHA1 (since the stored one includes one footer but the new computation includes two).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/models/topic_embed.rb</file_path>\n<line_numbers>13</line_numbers>\n\n<comment>`contents << \"\\n<hr>...\"` on line 13 uses the `<<` operator which mutates the string in place. This modifies the caller's original string. In `import_remote`, `doc.content` gets mutated. If `import` is called twice with the same string (e.g., in tests via `let(:contents)`), the imported-from footer is appended multiple times. More critically, when `import` is called for an update path (content_sha1 check on line 34), the SHA1 was computed *after* mutation, so re-importing the same content will always differ from the stored SHA1 (since the stored one includes one footer but the new computation includes two).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:47Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:31:47Z"
+          }
+        ]
       }
     ]
   },
@@ -42191,6 +43501,43 @@
             "created_at": "2026-03-18T20:27:48Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR1__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR1__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/models/optimized_image.rb",
+            "line": 145,
+            "body": "P1: Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/models/optimized_image.rb</file_path>\n<line_numbers>145</line_numbers>\n\n<comment>Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/models/optimized_image.rb</file_path>\n<line_numbers>145</line_numbers>\n\n<comment>Ruby duplicate method definition: `downsize` is defined twice (lines 145 and 149). In Ruby, the second definition completely replaces the first, so only the 3-arg version `downsize(from, to, dimensions, opts={})` survives. The existing caller in `app/jobs/regular/resize_emoji.rb:14` calls `OptimizedImage.downsize(path, path, 100, 100, opts)` with 5 arguments, which will now raise `ArgumentError: wrong number of arguments (given 5, expected 3..4)` at runtime, breaking emoji resizing.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:28Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/lib/utilities.js",
+            "line": 182,
+            "body": "P2: The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/lib/utilities.js</file_path>\n<line_numbers>182</line_numbers>\n\n<comment>The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/lib/utilities.js</file_path>\n<line_numbers>182</line_numbers>\n\n<comment>The client-side file size validation is hardcoded to 10MB (`10 * 1024` KB) for ALL upload types, replacing the previous dynamic lookup `Discourse.SiteSettings['max_' + type + '_size_kb']`. This means the admin-configurable `max_image_size_kb` and `max_attachment_size_kb` site settings are completely ignored on the client side. Non-image attachments (which are not auto-downsized server-side) will also be accepted up to 10MB regardless of the configured limit, potentially allowing files that the server will reject.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:28Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/lib/utilities.js",
+            "line": 246,
+            "body": "P3: The 413 error handler also hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/lib/utilities.js</file_path>\n<line_numbers>246</line_numbers>\n\n<comment>The 413 error handler also hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/lib/utilities.js</file_path>\n<line_numbers>246</line_numbers>\n\n<comment>The 413 error handler also hardcodes `maxSizeKB` to 10MB instead of reading the site setting. This means when a web server returns a 413 (entity too large) for any upload type, the error message will always say '10 MB' even if the actual server limit is different (e.g., nginx's `client_max_body_size` could be set to 5MB). The user sees a misleading error message.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:28Z"
+          },
+          {
+            "path": "app/controllers/uploads_controller.rb",
+            "line": 55,
+            "body": "P3: The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/controllers/uploads_controller.rb</file_path>\n<line_numbers>55</line_numbers>\n\n<comment>The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/controllers/uploads_controller.rb</file_path>\n<line_numbers>55</line_numbers>\n\n<comment>The API URL download limit is hardcoded to `10.megabytes` instead of using the site setting `SiteSetting.max_image_size_kb.kilobytes`. While the intent is to allow larger images for auto-downsizing, this hardcoded value means: (1) if an admin sets `max_image_size_kb` higher than 10MB, API uploads are artificially limited; (2) non-image URL uploads also get the 10MB limit since the type check happens later. The original code correctly limited based on the configured setting.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:32:28Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:32:28Z"
+          }
+        ]
       }
     ]
   },
@@ -43300,6 +44647,49 @@
             "created_at": "2026-03-18T20:27:39Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "discourse__discourse-graphite__tusk__PR2__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR2__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "app/controllers/topics_controller.rb",
+            "line": 107,
+            "body": "P1: NullPointerError crash: `TopicUser.find_by` returns `nil` when no TopicUser record exists (e.g., user received a notification but never visited the topic). The very next line calls `tu.notification_level` on the nil result, which raises `NoMethodError: undefined method 'notification_level' for nil:NilClass`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/controllers/topics_controller.rb</file_path>\n<line_numbers>107</line_numbers>\n\n<comment>NullPointerError crash: `TopicUser.find_by` returns `nil` when no TopicUser record exists (e.g., user received a notification but never visited the topic). The very next line calls `tu.notification_level` on the nil result, which raises `NoMethodError: undefined method 'notification_level' for nil:NilClass`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/controllers/topics_controller.rb</file_path>\n<line_numbers>107</line_numbers>\n\n<comment>NullPointerError crash: `TopicUser.find_by` returns `nil` when no TopicUser record exists (e.g., user received a notification but never visited the topic). The very next line calls `tu.notification_level` on the nil result, which raises `NoMethodError: undefined method 'notification_level' for nil:NilClass`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:53Z"
+          },
+          {
+            "path": "config/locales/server.en.yml",
+            "line": 1855,
+            "body": "P2: The `unsubscribe_link` i18n string now requires an `%{unsubscribe_url}` interpolation variable, but existing email paths that use `add_unsubscribe_link: true` without passing `unsubscribe_url` (e.g., the `mailing_list_notify` path which calls `send_notification_email` without an `unsubscribe_url` key) will raise an `I18n::MissingInterpolationArgument` error or render a broken `[click here]()` link with an empty URL.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>config/locales/server.en.yml</file_path>\n<line_numbers>1855</line_numbers>\n\n<comment>The `unsubscribe_link` i18n string now requires an `%{unsubscribe_url}` interpolation variable, but existing email paths that use `add_unsubscribe_link: true` without passing `unsubscribe_url` (e.g., the `mailing_list_notify` path which calls `send_notification_email` without an `unsubscribe_url` key) will raise an `I18n::MissingInterpolationArgument` error or render a broken `[click here]()` link with an empty URL.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>config/locales/server.en.yml</file_path>\n<line_numbers>1855</line_numbers>\n\n<comment>The `unsubscribe_link` i18n string now requires an `%{unsubscribe_url}` interpolation variable, but existing email paths that use `add_unsubscribe_link: true` without passing `unsubscribe_url` (e.g., the `mailing_list_notify` path which calls `send_notification_email` without an `unsubscribe_url` key) will raise an `I18n::MissingInterpolationArgument` error or render a broken `[click here]()` link with an empty URL.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:53Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/components/dropdown-button.js.es6",
+            "line": 29,
+            "body": "P3: XSS vulnerability: the `title` value is injected directly into HTML via string concatenation (`buffer.push(\"<h4 class='title'>\" + title + \"</h4>\")`). If `title` contains user-controlled content (e.g., a topic title with `<script>` tags), it will be rendered as raw HTML. The old code had the same issue but always rendered; the new code adds a nil guard but still doesn't escape the value.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/components/dropdown-button.js.es6</file_path>\n<line_numbers>29</line_numbers>\n\n<comment>XSS vulnerability: the `title` value is injected directly into HTML via string concatenation (`buffer.push(\"<h4 class='title'>\" + title + \"</h4>\")`). If `title` contains user-controlled content (e.g., a topic title with `<script>` tags), it will be rendered as raw HTML. The old code had the same issue but always rendered; the new code adds a nil guard but still doesn't escape the value.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/components/dropdown-button.js.es6</file_path>\n<line_numbers>29</line_numbers>\n\n<comment>XSS vulnerability: the `title` value is injected directly into HTML via string concatenation (`buffer.push(\"<h4 class='title'>\" + title + \"</h4>\")`). If `title` contains user-controlled content (e.g., a topic title with `<script>` tags), it will be rendered as raw HTML. The old code had the same issue but always rendered; the new code adds a nil guard but still doesn't escape the value.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:53Z"
+          },
+          {
+            "path": "app/assets/javascripts/discourse/templates/topic/unsubscribe.hbs",
+            "line": 3,
+            "body": "P3: The unsubscribe controller uses `stopNotificiationsText` (note the typo: 'Notificiations' instead of 'Notifications') as the computed property name. The template references `{{{stopNotificiationsText}}}` with the same typo so it works, but the triple-stache `{{{ }}}` renders raw HTML. The computed property interpolates `model.fancyTitle` (which can contain user-supplied HTML entities) into the i18n string. If `fancyTitle` contains malicious HTML, it would be rendered unescaped.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>app/assets/javascripts/discourse/templates/topic/unsubscribe.hbs</file_path>\n<line_numbers>3</line_numbers>\n\n<comment>The unsubscribe controller uses `stopNotificiationsText` (note the typo: 'Notificiations' instead of 'Notifications') as the computed property name. The template references `{{{stopNotificiationsText}}}` with the same typo so it works, but the triple-stache `{{{ }}}` renders raw HTML. The computed property interpolates `model.fancyTitle` (which can contain user-supplied HTML entities) into the i18n string. If `fancyTitle` contains malicious HTML, it would be rendered unescaped.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>app/assets/javascripts/discourse/templates/topic/unsubscribe.hbs</file_path>\n<line_numbers>3</line_numbers>\n\n<comment>The unsubscribe controller uses `stopNotificiationsText` (note the typo: 'Notificiations' instead of 'Notifications') as the computed property name. The template references `{{{stopNotificiationsText}}}` with the same typo so it works, but the triple-stache `{{{ }}}` renders raw HTML. The computed property interpolates `model.fancyTitle` (which can contain user-supplied HTML entities) into the i18n string. If `fancyTitle` contains malicious HTML, it would be rendered unescaped.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:53Z"
+          },
+          {
+            "path": "lib/email/message_builder.rb",
+            "line": 61,
+            "body": "P4: The `html_part` method only substitutes `%{respond_instructions}` and `%{unsubscribe_link}` placeholders when `add_unsubscribe_link` is true. But the notification.html.erb template always contains `%{respond_instructions}` (both inside the `context_posts.present?` block and at the bottom). When `add_unsubscribe_link` is false, these literal `%{respond_instructions}` and `%{unsubscribe_link}` strings will appear in the rendered email.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>lib/email/message_builder.rb</file_path>\n<line_numbers>61</line_numbers>\n\n<comment>The `html_part` method only substitutes `%{respond_instructions}` and `%{unsubscribe_link}` placeholders when `add_unsubscribe_link` is true. But the notification.html.erb template always contains `%{respond_instructions}` (both inside the `context_posts.present?` block and at the bottom). When `add_unsubscribe_link` is false, these literal `%{respond_instructions}` and `%{unsubscribe_link}` strings will appear in the rendered email.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>lib/email/message_builder.rb</file_path>\n<line_numbers>61</line_numbers>\n\n<comment>The `html_part` method only substitutes `%{respond_instructions}` and `%{unsubscribe_link}` placeholders when `add_unsubscribe_link` is true. But the notification.html.erb template always contains `%{respond_instructions}` (both inside the `context_posts.present?` block and at the bottom). When `add_unsubscribe_link` is false, these literal `%{respond_instructions}` and `%{unsubscribe_link}` strings will appear in the rendered email.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:31:53Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:31:53Z"
+          }
+        ]
       }
     ]
   },
@@ -44535,6 +45925,49 @@
             "created_at": "2026-03-18T20:27:15Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR22532__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22532__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/app-store/googlecalendar/lib/CalendarService.ts",
+            "line": 1024,
+            "body": "P2: Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive. The comment says 'Update SelectedCalendar.updatedAt' but nothing is actually updated, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/googlecalendar/lib/CalendarService.ts</file_path>\n<line_numbers>1024</line_numbers>\n\n<comment>Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive. The comment says 'Update SelectedCalendar.updatedAt' but nothing is actually updated, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/googlecalendar/lib/CalendarService.ts</file_path>\n<line_numbers>1024</line_numbers>\n\n<comment>Calling `updateManyByCredentialId(credentialId, {})` with an empty data object is a no-op. Prisma's `updateMany` does NOT auto-update `@updatedAt` fields \u2014 only `update()` and `upsert()` trigger the `@updatedAt` directive. The comment says 'Update SelectedCalendar.updatedAt' but nothing is actually updated, so the `updatedAt` timestamp on SelectedCalendar records will never change via this code path.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:13Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/calendars/deleteCache.handler.ts",
+            "line": 17,
+            "body": "P3: The `deleteCache` handler only checks `userId: user.id` on the credential, but does not account for delegation credentials. The `Credential` model supports delegation via `delegationCredentialId`, and the `CalendarCacheRepository` has special logic for delegated credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache. Conversely, the handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/calendars/deleteCache.handler.ts</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>The `deleteCache` handler only checks `userId: user.id` on the credential, but does not account for delegation credentials. The `Credential` model supports delegation via `delegationCredentialId`, and the `CalendarCacheRepository` has special logic for delegated credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache. Conversely, the handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/calendars/deleteCache.handler.ts</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>The `deleteCache` handler only checks `userId: user.id` on the credential, but does not account for delegation credentials. The `Credential` model supports delegation via `delegationCredentialId`, and the `CalendarCacheRepository` has special logic for delegated credentials. A user with a delegation credential may not be the credential's `userId` owner, meaning they could be blocked from deleting their own cache. Conversely, the handler doesn't verify that the credential is actually a calendar credential \u2014 any credential ID owned by the user would pass validation, allowing cache deletion attempts for non-calendar credentials.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:13Z"
+          },
+          {
+            "path": "packages/prisma/schema.prisma",
+            "line": 1719,
+            "body": "P3: The migration uses `DEFAULT NOW()` which evaluates to the timestamp at migration time for existing rows, but in PostgreSQL `NOW()` as a column default actually evaluates per-row at insert time, so new inserts are fine. However, the migration comment says 'Added the required column `updatedAt` to the `CalendarCache` table without a default value. This is not possible if the table is not empty' \u2014 yet the SQL does add a default. The real issue is that the Prisma schema uses `@default(now()) @updatedAt` together. The `@updatedAt` directive makes Prisma set the value on every update, but `@default(now())` combined with `@updatedAt` means Prisma will ignore the DB default on creates and use its own `now()`. For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/prisma/schema.prisma</file_path>\n<line_numbers>1719</line_numbers>\n\n<comment>The migration uses `DEFAULT NOW()` which evaluates to the timestamp at migration time for existing rows, but in PostgreSQL `NOW()` as a column default actually evaluates per-row at insert time, so new inserts are fine. However, the migration comment says 'Added the required column `updatedAt` to the `CalendarCache` table without a default value. This is not possible if the table is not empty' \u2014 yet the SQL does add a default. The real issue is that the Prisma schema uses `@default(now()) @updatedAt` together. The `@updatedAt` directive makes Prisma set the value on every update, but `@default(now())` combined with `@updatedAt` means Prisma will ignore the DB default on creates and use its own `now()`. For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/prisma/schema.prisma</file_path>\n<line_numbers>1719</line_numbers>\n\n<comment>The migration uses `DEFAULT NOW()` which evaluates to the timestamp at migration time for existing rows, but in PostgreSQL `NOW()` as a column default actually evaluates per-row at insert time, so new inserts are fine. However, the migration comment says 'Added the required column `updatedAt` to the `CalendarCache` table without a default value. This is not possible if the table is not empty' \u2014 yet the SQL does add a default. The real issue is that the Prisma schema uses `@default(now()) @updatedAt` together. The `@updatedAt` directive makes Prisma set the value on every update, but `@default(now())` combined with `@updatedAt` means Prisma will ignore the DB default on creates and use its own `now()`. For raw SQL inserts (which the codebase uses \u2014 see `upsertCachedAvailability` using `$queryRaw`), the `@updatedAt` directive won't fire, so `updatedAt` will remain at the DB default `NOW()` from insert time and never get updated by raw queries.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:13Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/calendars/connectedCalendars.handler.ts",
+            "line": 35,
+            "body": "P4: Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` means that if `updatedAt` is somehow a falsy value other than `undefined` (which it can't be for Date, but is a code smell), it would be coerced to null. More importantly, this pattern is inconsistent \u2014 the `_max.updatedAt` from Prisma's `groupBy` can be `null` when there are matching rows but the field is nullable. The `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. This is minor but `?? null` would be more semantically correct.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/calendars/connectedCalendars.handler.ts</file_path>\n<line_numbers>35</line_numbers>\n\n<comment>Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` means that if `updatedAt` is somehow a falsy value other than `undefined` (which it can't be for Date, but is a code smell), it would be coerced to null. More importantly, this pattern is inconsistent \u2014 the `_max.updatedAt` from Prisma's `groupBy` can be `null` when there are matching rows but the field is nullable. The `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. This is minor but `?? null` would be more semantically correct.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/calendars/connectedCalendars.handler.ts</file_path>\n<line_numbers>35</line_numbers>\n\n<comment>Using `cacheStatusMap.get(calendar.credentialId) || null` with `||` instead of `??` means that if `updatedAt` is somehow a falsy value other than `undefined` (which it can't be for Date, but is a code smell), it would be coerced to null. More importantly, this pattern is inconsistent \u2014 the `_max.updatedAt` from Prisma's `groupBy` can be `null` when there are matching rows but the field is nullable. The `Map.get()` returns `undefined` for missing keys vs `null` for present-but-null values, and `||` treats both the same. This is minor but `?? null` would be more semantically correct.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:13Z"
+          },
+          {
+            "path": "packages/features/apps/components/CredentialActionsDropdown.tsx",
+            "line": 44,
+            "body": "P4: The `CredentialActionsDropdown` component is always rendered now (replacing the old conditional that checked `!delegationCredentialId && !disableConnectionModification` before rendering anything). When `canDisconnect` is false and `hasCache` is false, the component returns null \u2014 but it still initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them. While not a crash bug, the `disconnectMutation` calls `trpc.viewer.credentials.delete.useMutation()` which sets up subscription handlers \u2014 this happens for every calendar credential even when the dropdown renders nothing.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/apps/components/CredentialActionsDropdown.tsx</file_path>\n<line_numbers>44</line_numbers>\n\n<comment>The `CredentialActionsDropdown` component is always rendered now (replacing the old conditional that checked `!delegationCredentialId && !disableConnectionModification` before rendering anything). When `canDisconnect` is false and `hasCache` is false, the component returns null \u2014 but it still initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them. While not a crash bug, the `disconnectMutation` calls `trpc.viewer.credentials.delete.useMutation()` which sets up subscription handlers \u2014 this happens for every calendar credential even when the dropdown renders nothing.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/apps/components/CredentialActionsDropdown.tsx</file_path>\n<line_numbers>44</line_numbers>\n\n<comment>The `CredentialActionsDropdown` component is always rendered now (replacing the old conditional that checked `!delegationCredentialId && !disableConnectionModification` before rendering anything). When `canDisconnect` is false and `hasCache` is false, the component returns null \u2014 but it still initializes three `useState` hooks, a tRPC mutation (`deleteCacheMutation`), and a disconnect mutation on every render even for non-Google-Calendar credentials that will never use them. While not a crash bug, the `disconnectMutation` calls `trpc.viewer.credentials.delete.useMutation()` which sets up subscription handlers \u2014 this happens for every calendar credential even when the dropdown renders nothing.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:07:13Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:07:13Z"
+          }
+        ]
       }
     ]
   },
@@ -45657,6 +47090,43 @@
             "created_at": "2026-03-18T20:27:29Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR8330__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8330__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/trpc/server/routers/viewer/slots.ts",
+            "line": 114,
+            "body": "P1: Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>114</line_numbers>\n\n<comment>Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>114</line_numbers>\n\n<comment>Object reference equality `===` is used to compare two different dayjs objects, which will always be `false` (dayjs objects are compared by reference, not value). This means the check for whether the override start equals the override end (a zero-length override representing an unavailable day) will never trigger, so unavailable-day overrides will be treated as available instead of unavailable.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:09:29Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/slots.ts",
+            "line": 142,
+            "body": "P2: Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>142</line_numbers>\n\n<comment>Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>142</line_numbers>\n\n<comment>Copy-paste bug: `end` is computed identically to `start` \u2014 both use `slotStartTime.hour() * 60 + slotStartTime.minute()`. The `end` variable should use `slotEndTime` instead. As written, the condition `end > workingHour.endTime` checks the slot's start against the working hour end, meaning a slot whose end time extends past working hours will not be rejected. This allows booking slots that partially fall outside working hours.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:09:29Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/slots.ts",
+            "line": 107,
+            "body": "P3: The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC. The code in `slots.ts` (getSlots) handles this differently using `inviteeUtcOffset - organizerUtcOffset`, suggesting an inconsistent approach between the two timezone adjustment implementations.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>107</line_numbers>\n\n<comment>The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC. The code in `slots.ts` (getSlots) handles this differently using `inviteeUtcOffset - organizerUtcOffset`, suggesting an inconsistent approach between the two timezone adjustment implementations.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>107</line_numbers>\n\n<comment>The `utcOffset` computation multiplies the organizer's UTC offset by -1 to convert organizer-local dates to UTC. However, when comparing `slotEndTime` (already in UTC) against `dayjs(date.start).add(utcOffset, 'minutes')`, the override boundaries are being shifted from organizer-local to UTC, but the slot times are already in UTC. The issue is that `dayjs(date.start)` parses the Date object as local time (not UTC), then adds the negated organizer offset. This double-conversion produces incorrect boundaries when the server's local timezone differs from UTC. The code in `slots.ts` (getSlots) handles this differently using `inviteeUtcOffset - organizerUtcOffset`, suggesting an inconsistent approach between the two timezone adjustment implementations.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:09:29Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/slots.ts",
+            "line": 103,
+            "body": "P3: The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range (returns true \u2192 function returns false), it prevents checking whether the slot is valid for a different user's override. The combination of mutation (`dateOverrideExist = true`) and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range (returns true \u2192 function returns false), it prevents checking whether the slot is valid for a different user's override. The combination of mutation (`dateOverrideExist = true`) and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/slots.ts</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>The `dateOverrideExist` flag is set inside `.find()` but `.find()` short-circuits on the first truthy return. When there are multiple date overrides for the same day (e.g., from different users in a round-robin), and the first override found marks the slot as outside that override's range (returns true \u2192 function returns false), it prevents checking whether the slot is valid for a different user's override. The combination of mutation (`dateOverrideExist = true`) and early-return via `.find()` means that if the first matching override rejects the slot, the slot is marked unavailable even if another override would allow it.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:09:29Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:09:29Z"
+          }
+        ]
       }
     ]
   },
@@ -46611,6 +48081,12 @@
             "created_at": "2026-03-18T20:27:22Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR14943__20260411",
+        "pr_url": null,
+        "review_comments": []
       }
     ]
   },
@@ -47498,6 +48974,49 @@
             "created_at": "2026-03-18T20:27:35Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR22345__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22345__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/lib/server/service/insightsBooking.ts",
+            "line": 171,
+            "body": "P3: Public methods `getAuthorizationConditions()` and `getFilterConditions()` return `Prisma.Sql` fragments containing OR operators without outer parentheses. For example, `buildOrgAuthorizationCondition` returns `(teamCondition) OR (userCondition)` and `buildFilterConditions` with only eventTypeId returns `(\"eventTypeId\" = X) OR (\"eventParentId\" = X)`. While `getBaseConditions()` wraps these in parens correctly, any caller that composes these public methods directly into a raw query with AND (e.g., `Prisma.sql\\`${authConditions} AND \"startTime\" > ${date}\\``) would hit SQL operator precedence bugs: AND binds tighter than OR, so one branch of the OR would bypass the AND condition. For auth conditions this would be a privilege escalation (org team bookings returned without date filter). The fix is to wrap the final return value of each builder in outer parentheses.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>171</line_numbers>\n\n<comment>Public methods `getAuthorizationConditions()` and `getFilterConditions()` return `Prisma.Sql` fragments containing OR operators without outer parentheses. For example, `buildOrgAuthorizationCondition` returns `(teamCondition) OR (userCondition)` and `buildFilterConditions` with only eventTypeId returns `(\"eventTypeId\" = X) OR (\"eventParentId\" = X)`. While `getBaseConditions()` wraps these in parens correctly, any caller that composes these public methods directly into a raw query with AND (e.g., `Prisma.sql\\`${authConditions} AND \"startTime\" > ${date}\\``) would hit SQL operator precedence bugs: AND binds tighter than OR, so one branch of the OR would bypass the AND condition. For auth conditions this would be a privilege escalation (org team bookings returned without date filter). The fix is to wrap the final return value of each builder in outer parentheses.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>171</line_numbers>\n\n<comment>Public methods `getAuthorizationConditions()` and `getFilterConditions()` return `Prisma.Sql` fragments containing OR operators without outer parentheses. For example, `buildOrgAuthorizationCondition` returns `(teamCondition) OR (userCondition)` and `buildFilterConditions` with only eventTypeId returns `(\"eventTypeId\" = X) OR (\"eventParentId\" = X)`. While `getBaseConditions()` wraps these in parens correctly, any caller that composes these public methods directly into a raw query with AND (e.g., `Prisma.sql\\`${authConditions} AND \"startTime\" > ${date}\\``) would hit SQL operator precedence bugs: AND binds tighter than OR, so one branch of the OR would bypass the AND condition. For auth conditions this would be a privilege escalation (org team bookings returned without date filter). The fix is to wrap the final return value of each builder in outer parentheses.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:20:04Z"
+          },
+          {
+            "path": "packages/lib/server/service/insightsBooking.ts",
+            "line": 57,
+            "body": "P3: The constructor parameter type was changed from `InsightsBookingServiceOptions` (a discriminated union where `teamId` is required only for scope='team') to `InsightsBookingServicePublicOptions` (a flat type with `teamId?: number` for all scopes). This means TypeScript no longer catches at compile time when a caller forgets to pass `teamId` for team scope. The call `new InsightsBookingService({ prisma, options: { scope: 'team', userId: 1, orgId: 1 } })` compiles without error but silently returns NOTHING_CONDITION (no data) because zod validation fails at runtime. The old type would have caused a compile error.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>57</line_numbers>\n\n<comment>The constructor parameter type was changed from `InsightsBookingServiceOptions` (a discriminated union where `teamId` is required only for scope='team') to `InsightsBookingServicePublicOptions` (a flat type with `teamId?: number` for all scopes). This means TypeScript no longer catches at compile time when a caller forgets to pass `teamId` for team scope. The call `new InsightsBookingService({ prisma, options: { scope: 'team', userId: 1, orgId: 1 } })` compiles without error but silently returns NOTHING_CONDITION (no data) because zod validation fails at runtime. The old type would have caused a compile error.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>57</line_numbers>\n\n<comment>The constructor parameter type was changed from `InsightsBookingServiceOptions` (a discriminated union where `teamId` is required only for scope='team') to `InsightsBookingServicePublicOptions` (a flat type with `teamId?: number` for all scopes). This means TypeScript no longer catches at compile time when a caller forgets to pass `teamId` for team scope. The call `new InsightsBookingService({ prisma, options: { scope: 'team', userId: 1, orgId: 1 } })` compiles without error but silently returns NOTHING_CONDITION (no data) because zod validation fails at runtime. The old type would have caused a compile error.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:20:04Z"
+          },
+          {
+            "path": "packages/lib/server/service/insightsBooking.ts",
+            "line": 104,
+            "body": "P4: Falsy guard `if (this.filters.eventTypeId)` skips the filter when eventTypeId is 0, and similarly `if (this.filters.memberUserId)` skips memberUserId 0. While database IDs are typically positive, the type is `number` which includes 0. A caller passing `{ eventTypeId: 0 }` would silently get unfiltered results instead of filtering by eventTypeId=0. This is pre-existing behavior carried into the raw SQL conversion, but it's worth fixing during this refactor.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>104</line_numbers>\n\n<comment>Falsy guard `if (this.filters.eventTypeId)` skips the filter when eventTypeId is 0, and similarly `if (this.filters.memberUserId)` skips memberUserId 0. While database IDs are typically positive, the type is `number` which includes 0. A caller passing `{ eventTypeId: 0 }` would silently get unfiltered results instead of filtering by eventTypeId=0. This is pre-existing behavior carried into the raw SQL conversion, but it's worth fixing during this refactor.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>104</line_numbers>\n\n<comment>Falsy guard `if (this.filters.eventTypeId)` skips the filter when eventTypeId is 0, and similarly `if (this.filters.memberUserId)` skips memberUserId 0. While database IDs are typically positive, the type is `number` which includes 0. A caller passing `{ eventTypeId: 0 }` would silently get unfiltered results instead of filtering by eventTypeId=0. This is pre-existing behavior carried into the raw SQL conversion, but it's worth fixing during this refactor.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:20:04Z"
+          },
+          {
+            "path": "packages/lib/server/service/insightsBooking.ts",
+            "line": 72,
+            "body": "P4: The `getBaseConditions` method treats `authConditions` as potentially falsy with `if (authConditions && filterConditions)` / `else if (authConditions)` / `else if (filterConditions)` / `else` branches. However, `getAuthorizationConditions()` always returns a `Prisma.Sql` object (even NOTHING_CONDITION is truthy), so `authConditions` is always truthy. The third and fourth branches (`else if (filterConditions)` and `else return NOTHING_CONDITION`) are unreachable dead code. This also means when auth fails (NOTHING_CONDITION), filter conditions are still computed and combined as `(1=0) AND (filter)` rather than short-circuiting.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>72</line_numbers>\n\n<comment>The `getBaseConditions` method treats `authConditions` as potentially falsy with `if (authConditions && filterConditions)` / `else if (authConditions)` / `else if (filterConditions)` / `else` branches. However, `getAuthorizationConditions()` always returns a `Prisma.Sql` object (even NOTHING_CONDITION is truthy), so `authConditions` is always truthy. The third and fourth branches (`else if (filterConditions)` and `else return NOTHING_CONDITION`) are unreachable dead code. This also means when auth fails (NOTHING_CONDITION), filter conditions are still computed and combined as `(1=0) AND (filter)` rather than short-circuiting.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/lib/server/service/insightsBooking.ts</file_path>\n<line_numbers>72</line_numbers>\n\n<comment>The `getBaseConditions` method treats `authConditions` as potentially falsy with `if (authConditions && filterConditions)` / `else if (authConditions)` / `else if (filterConditions)` / `else` branches. However, `getAuthorizationConditions()` always returns a `Prisma.Sql` object (even NOTHING_CONDITION is truthy), so `authConditions` is always truthy. The third and fourth branches (`else if (filterConditions)` and `else return NOTHING_CONDITION`) are unreachable dead code. This also means when auth fails (NOTHING_CONDITION), filter conditions are still computed and combined as `(1=0) AND (filter)` rather than short-circuiting.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:20:05Z"
+          },
+          {
+            "path": "packages/lib/server/service/__tests__/insightsBooking.integration-test.ts",
+            "line": 435,
+            "body": "P4: The entire 'Caching' test suite was removed (two tests verifying that `getAuthorizationConditions` and `getFilterConditions` return cached results on subsequent calls). The caching logic still exists in the service (lines 84-87 and 91-94) but is no longer tested. If someone accidentally breaks the caching (e.g., removes the `undefined` check), there would be no test to catch the regression, leading to repeated database queries for membership checks on every call.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/lib/server/service/__tests__/insightsBooking.integration-test.ts</file_path>\n<line_numbers>435</line_numbers>\n\n<comment>The entire 'Caching' test suite was removed (two tests verifying that `getAuthorizationConditions` and `getFilterConditions` return cached results on subsequent calls). The caching logic still exists in the service (lines 84-87 and 91-94) but is no longer tested. If someone accidentally breaks the caching (e.g., removes the `undefined` check), there would be no test to catch the regression, leading to repeated database queries for membership checks on every call.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/lib/server/service/__tests__/insightsBooking.integration-test.ts</file_path>\n<line_numbers>435</line_numbers>\n\n<comment>The entire 'Caching' test suite was removed (two tests verifying that `getAuthorizationConditions` and `getFilterConditions` return cached results on subsequent calls). The caching logic still exists in the service (lines 84-87 and 91-94) but is no longer tested. If someone accidentally breaks the caching (e.g., removes the `undefined` check), there would be no test to catch the regression, leading to repeated database queries for membership checks on every call.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:20:05Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:20:04Z"
+          }
+        ]
       }
     ]
   },
@@ -49202,6 +50721,49 @@
             "created_at": "2026-03-18T20:27:28Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR11059__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR11059__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/app-store/googlecalendar/lib/CalendarService.ts",
+            "line": 97,
+            "body": "P1: parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously `googleCredentialSchema.parse()` returned just the data. Now the DB will store `{success: true, data: {access_token, refresh_token, expiry_date}}` instead of `{access_token, refresh_token, expiry_date}`, corrupting the credential for all subsequent reads.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/googlecalendar/lib/CalendarService.ts</file_path>\n<line_numbers>97</line_numbers>\n\n<comment>parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously `googleCredentialSchema.parse()` returned just the data. Now the DB will store `{success: true, data: {access_token, refresh_token, expiry_date}}` instead of `{access_token, refresh_token, expiry_date}`, corrupting the credential for all subsequent reads.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/googlecalendar/lib/CalendarService.ts</file_path>\n<line_numbers>97</line_numbers>\n\n<comment>parseRefreshTokenResponse returns the full safeParse result object ({success, data}) but Google Calendar stores it directly as the credential key. Previously `googleCredentialSchema.parse()` returned just the data. Now the DB will store `{success: true, data: {access_token, refresh_token, expiry_date}}` instead of `{access_token, refresh_token, expiry_date}`, corrupting the credential for all subsequent reads.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:41Z"
+          },
+          {
+            "path": "packages/app-store/_utils/oauth/refreshOAuthTokens.ts",
+            "line": 14,
+            "body": "P1: When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses `res?.data` (undefined on fetch Response \u2192 TypeError on `token.access_token`), Hubspot accesses `.expiryDate`/`.expiresIn` (undefined on Response), Zoho-bigin and Zohocrm access `.data.error` (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/_utils/oauth/refreshOAuthTokens.ts</file_path>\n<line_numbers>14</line_numbers>\n\n<comment>When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses `res?.data` (undefined on fetch Response \u2192 TypeError on `token.access_token`), Hubspot accesses `.expiryDate`/`.expiresIn` (undefined on Response), Zoho-bigin and Zohocrm access `.data.error` (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/_utils/oauth/refreshOAuthTokens.ts</file_path>\n<line_numbers>14</line_numbers>\n\n<comment>When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response. But callers that previously received app-specific objects now crash: Google Calendar accesses `res?.data` (undefined on fetch Response \u2192 TypeError on `token.access_token`), Hubspot accesses `.expiryDate`/`.expiresIn` (undefined on Response), Zoho-bigin and Zohocrm access `.data.error` (axios response format, not fetch Response). The function returns fundamentally different types depending on the code path.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:41Z"
+          },
+          {
+            "path": "packages/app-store/_utils/oauth/parseRefreshTokenResponse.ts",
+            "line": 5,
+            "body": "P2: minimumTokenResponseSchema uses `[z.string().toString()]` as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field (like `expires_in` or `expiry_date`) exists in the response, making the 'minimum token validation' ineffective.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/_utils/oauth/parseRefreshTokenResponse.ts</file_path>\n<line_numbers>5</line_numbers>\n\n<comment>minimumTokenResponseSchema uses `[z.string().toString()]` as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field (like `expires_in` or `expiry_date`) exists in the response, making the 'minimum token validation' ineffective.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/_utils/oauth/parseRefreshTokenResponse.ts</file_path>\n<line_numbers>5</line_numbers>\n\n<comment>minimumTokenResponseSchema uses `[z.string().toString()]` as a computed property key. This evaluates z.string().toString() at schema-definition time to a literal string like 'ZodString', creating a schema that requires a property literally named 'ZodString' with a number value \u2014 not a dynamic key validator. The schema will never actually validate that an expiry field (like `expires_in` or `expiry_date`) exists in the response, making the 'minimum token validation' ineffective.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:41Z"
+          },
+          {
+            "path": "packages/app-store/salesforce/lib/CalendarService.ts",
+            "line": 103,
+            "body": "P2: _Note: Issue is at line 106, shown here on nearest changed line._\n\nSalesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using `credentialKey.access_token` \u2014 the OLD, pre-refresh token. The newly refreshed token from `accessTokenParsed.data` is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/salesforce/lib/CalendarService.ts</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>_Note: Issue is at line 106, shown here on nearest changed line._\n\nSalesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using `credentialKey.access_token` \u2014 the OLD, pre-refresh token. The newly refreshed token from `accessTokenParsed.data` is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/salesforce/lib/CalendarService.ts</file_path>\n<line_numbers>103</line_numbers>\n\n<comment>_Note: Issue is at line 106, shown here on nearest changed line._\n\nSalesforce getClient refreshes the access token and stores the new token in the DB, but then creates the jsforce.Connection using `credentialKey.access_token` \u2014 the OLD, pre-refresh token. The newly refreshed token from `accessTokenParsed.data` is saved to the DB but never used for the current connection, so the connection will attempt to use a stale/expired token.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:41Z"
+          },
+          {
+            "path": "apps/web/pages/api/webhook/app-credential.ts",
+            "line": 17,
+            "body": "P4: The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>apps/web/pages/api/webhook/app-credential.ts</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>apps/web/pages/api/webhook/app-credential.ts</file_path>\n<line_numbers>17</line_numbers>\n\n<comment>The webhook endpoint does not check the HTTP method, so it accepts GET, PUT, DELETE, and other methods in addition to POST. A GET request with query parameters won't have a JSON body, causing a zod parse error (unhandled 500). More importantly, non-POST methods bypass the semantic intent of the endpoint.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:41Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:08:40Z"
+          }
+        ]
       }
     ]
   },
@@ -50396,6 +51958,49 @@
             "created_at": "2026-03-18T20:37:30Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR7232__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR7232__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/features/ee/workflows/lib/reminders/emailReminderManager.ts",
+            "line": 222,
+            "body": "P2: When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record. The record becomes permanently orphaned\u2014it's never cleaned up by the CRON's cancel loop (since `cancelled` is not set to true) and won't be found by the unscheduled query either (since `scheduled` is true). Called from `handleNewBooking.ts:968` (reschedule), `workflows.tsx:210` (delete workflow), and `workflows.tsx:522` (delete step).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/ee/workflows/lib/reminders/emailReminderManager.ts</file_path>\n<line_numbers>222</line_numbers>\n\n<comment>When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record. The record becomes permanently orphaned\u2014it's never cleaned up by the CRON's cancel loop (since `cancelled` is not set to true) and won't be found by the unscheduled query either (since `scheduled` is true). Called from `handleNewBooking.ts:968` (reschedule), `workflows.tsx:210` (delete workflow), and `workflows.tsx:522` (delete step).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/ee/workflows/lib/reminders/emailReminderManager.ts</file_path>\n<line_numbers>222</line_numbers>\n\n<comment>When `immediateDelete` is true and `referenceId` is non-null, the function sends a SendGrid cancel request but then returns without deleting or marking the `workflowReminder` DB record. The record becomes permanently orphaned\u2014it's never cleaned up by the CRON's cancel loop (since `cancelled` is not set to true) and won't be found by the unscheduled query either (since `scheduled` is true). Called from `handleNewBooking.ts:968` (reschedule), `workflows.tsx:210` (delete workflow), and `workflows.tsx:522` (delete step).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:49Z"
+          },
+          {
+            "path": "packages/features/ee/workflows/api/scheduleEmailReminders.ts",
+            "line": 40,
+            "body": "P2: _Note: Issue is at line 34, shown here on nearest changed line._\n\nThe CRON job first deletes ALL email reminders with a past `scheduledDate` (line 34-41), then queries for `cancelled: true` reminders to cancel via SendGrid (line 44-51). If a cancelled reminder's `scheduledDate` has already passed (e.g., booking cancelled 5 minutes before the scheduled send, CRON runs after), the first `deleteMany` removes the DB record before the cancel loop processes it. The email is never cancelled in SendGrid and gets delivered despite the booking being cancelled.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/ee/workflows/api/scheduleEmailReminders.ts</file_path>\n<line_numbers>40</line_numbers>\n\n<comment>_Note: Issue is at line 34, shown here on nearest changed line._\n\nThe CRON job first deletes ALL email reminders with a past `scheduledDate` (line 34-41), then queries for `cancelled: true` reminders to cancel via SendGrid (line 44-51). If a cancelled reminder's `scheduledDate` has already passed (e.g., booking cancelled 5 minutes before the scheduled send, CRON runs after), the first `deleteMany` removes the DB record before the cancel loop processes it. The email is never cancelled in SendGrid and gets delivered despite the booking being cancelled.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/ee/workflows/api/scheduleEmailReminders.ts</file_path>\n<line_numbers>40</line_numbers>\n\n<comment>_Note: Issue is at line 34, shown here on nearest changed line._\n\nThe CRON job first deletes ALL email reminders with a past `scheduledDate` (line 34-41), then queries for `cancelled: true` reminders to cancel via SendGrid (line 44-51). If a cancelled reminder's `scheduledDate` has already passed (e.g., booking cancelled 5 minutes before the scheduled send, CRON runs after), the first `deleteMany` removes the DB record before the cancel loop processes it. The email is never cancelled in SendGrid and gets delivered despite the booking being cancelled.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:49Z"
+          },
+          {
+            "path": "packages/features/ee/workflows/api/scheduleEmailReminders.ts",
+            "line": 53,
+            "body": "P2: The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails (e.g., network error, invalid batch_id), the exception aborts the loop and all remaining reminders are never cancelled or deleted. Those emails will be sent despite their bookings being cancelled.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/ee/workflows/api/scheduleEmailReminders.ts</file_path>\n<line_numbers>53</line_numbers>\n\n<comment>The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails (e.g., network error, invalid batch_id), the exception aborts the loop and all remaining reminders are never cancelled or deleted. Those emails will be sent despite their bookings being cancelled.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/ee/workflows/api/scheduleEmailReminders.ts</file_path>\n<line_numbers>53</line_numbers>\n\n<comment>The entire cancellation for-loop (lines 56-73) is wrapped in a single try/catch. If any one SendGrid cancel request fails (e.g., network error, invalid batch_id), the exception aborts the loop and all remaining reminders are never cancelled or deleted. Those emails will be sent despite their bookings being cancelled.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:49Z"
+          },
+          {
+            "path": "packages/features/bookings/lib/handleNewBooking.ts",
+            "line": 966,
+            "body": "P3: In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget. The surrounding try/catch (line 964) will never catch rejections from these promises\u2014they'll become unhandled promise rejections instead of being logged as intended.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/bookings/lib/handleNewBooking.ts</file_path>\n<line_numbers>966</line_numbers>\n\n<comment>In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget. The surrounding try/catch (line 964) will never catch rejections from these promises\u2014they'll become unhandled promise rejections instead of being logged as intended.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/bookings/lib/handleNewBooking.ts</file_path>\n<line_numbers>966</line_numbers>\n\n<comment>In the reschedule path, `forEach` calls `deleteScheduledEmailReminder` and `deleteScheduledSMSReminder` (both async) without awaiting the returned promises. Since `forEach` doesn't await async callbacks, all calls are fire-and-forget. The surrounding try/catch (line 964) will never catch rejections from these promises\u2014they'll become unhandled promise rejections instead of being logged as intended.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:49Z"
+          },
+          {
+            "path": "packages/features/ee/workflows/components/WorkflowStepContainer.tsx",
+            "line": 461,
+            "body": "P3: _Note: Issue is at line 463, shown here on nearest changed line._\n\nThe sender name/sender ID input section (lines 463-492) was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}`. After the refactor, this section is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant. This shows an unnecessary sender name field for email-type workflow steps.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/ee/workflows/components/WorkflowStepContainer.tsx</file_path>\n<line_numbers>461</line_numbers>\n\n<comment>_Note: Issue is at line 463, shown here on nearest changed line._\n\nThe sender name/sender ID input section (lines 463-492) was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}`. After the refactor, this section is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant. This shows an unnecessary sender name field for email-type workflow steps.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/ee/workflows/components/WorkflowStepContainer.tsx</file_path>\n<line_numbers>461</line_numbers>\n\n<comment>_Note: Issue is at line 463, shown here on nearest changed line._\n\nThe sender name/sender ID input section (lines 463-492) was previously rendered conditionally inside `{(isPhoneNumberNeeded || isSenderIdNeeded) && (...)}`. After the refactor, this section is now rendered unconditionally for all workflow step types, including email actions where sender ID/name inputs are irrelevant. This shows an unnecessary sender name field for email-type workflow steps.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:08:49Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:08:49Z"
+          }
+        ]
       }
     ]
   },
@@ -51824,6 +53429,49 @@
             "created_at": "2026-03-18T20:27:18Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR14740__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR14740__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts",
+            "line": 47,
+            "body": "P1: Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>47</line_numbers>\n\n<comment>Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>47</line_numbers>\n\n<comment>Authorization check uses `&&` (AND) instead of `||` (OR). The variable is named `isTeamAdminOrOwner` but requires the user to be BOTH a team admin AND a team owner. Since `isTeamOwner` only matches `role: \"OWNER\"`, any team admin who is not also an owner will be incorrectly denied access. This effectively makes the team admin check dead code \u2014 only owners pass.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:06:44Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts",
+            "line": 168,
+            "body": "P2: The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. In `sendAddGuestsEmails`, `newGuests.includes(attendee.email)` determines which email template to use: new guests get `AttendeeScheduledEmail` (booking confirmation), existing attendees get `AttendeeAddGuestsEmail` (notification). If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>168</line_numbers>\n\n<comment>The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. In `sendAddGuestsEmails`, `newGuests.includes(attendee.email)` determines which email template to use: new guests get `AttendeeScheduledEmail` (booking confirmation), existing attendees get `AttendeeAddGuestsEmail` (notification). If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>168</line_numbers>\n\n<comment>The handler passes the original `guests` array (raw user input) to `sendAddGuestsEmails` instead of the filtered `uniqueGuests`. In `sendAddGuestsEmails`, `newGuests.includes(attendee.email)` determines which email template to use: new guests get `AttendeeScheduledEmail` (booking confirmation), existing attendees get `AttendeeAddGuestsEmail` (notification). If a user submits emails that already exist as attendees, those existing attendees will incorrectly receive a full booking confirmation email instead of the \"new guests added\" notification.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:06:44Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts",
+            "line": 77,
+            "body": "P3: Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>77</line_numbers>\n\n<comment>Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>77</line_numbers>\n\n<comment>Blacklist comparison is case-sensitive on the guest side. `blacklistedGuestEmails` are lowercased via `.map(email => email.toLowerCase())`, but incoming `guest` emails are compared as-is. A user submitting `Blocked@Example.COM` would bypass the blacklist entry `blocked@example.com`.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:06:44Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts",
+            "line": 60,
+            "body": "P3: When `booking.userId` is null (it's `Int?` in the Prisma schema), `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable: if the user is an attendee (not the organizer) of a booking whose organizer account was deleted, the authorization check passes but this query crashes.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>60</line_numbers>\n\n<comment>When `booking.userId` is null (it's `Int?` in the Prisma schema), `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable: if the user is an attendee (not the organizer) of a booking whose organizer account was deleted, the authorization check passes but this query crashes.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/bookings/addGuests.handler.ts</file_path>\n<line_numbers>60</line_numbers>\n\n<comment>When `booking.userId` is null (it's `Int?` in the Prisma schema), `booking.userId || 0` resolves to `0`, and `prisma.user.findFirstOrThrow({ where: { id: 0 } })` will throw an unhandled Prisma NotFoundError instead of a proper TRPC error. This code path is reachable: if the user is an attendee (not the organizer) of a booking whose organizer account was deleted, the authorization check passes but this query crashes.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:06:44Z"
+          },
+          {
+            "path": "packages/emails/templates/organizer-add-guests-email.ts",
+            "line": 28,
+            "body": "P4: In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. While in practice attendees should exist, if the attendees array is ever empty (e.g., all attendees removed in a race condition), this would throw a runtime TypeError crashing the email send for the entire batch (since all emails are sent via `Promise.all`).\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/emails/templates/organizer-add-guests-email.ts</file_path>\n<line_numbers>28</line_numbers>\n\n<comment>In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. While in practice attendees should exist, if the attendees array is ever empty (e.g., all attendees removed in a race condition), this would throw a runtime TypeError crashing the email send for the entire batch (since all emails are sent via `Promise.all`).</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/emails/templates/organizer-add-guests-email.ts</file_path>\n<line_numbers>28</line_numbers>\n\n<comment>In `OrganizerAddGuestsEmail.getNodeMailerPayload()`, `this.calEvent.attendees[0].name` accesses index 0 without a guard. While in practice attendees should exist, if the attendees array is ever empty (e.g., all attendees removed in a race condition), this would throw a runtime TypeError crashing the email send for the entire batch (since all emails are sent via `Promise.all`).</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:06:44Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:06:44Z"
+          }
+        ]
       }
     ]
   },
@@ -53080,6 +54728,43 @@
             "created_at": "2026-03-18T20:27:45Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR10600__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10600__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/features/auth/lib/next-auth-options.ts",
+            "line": 131,
+            "body": "P1: Password bypass via backup code + dummy totpCode. The password check at line 121 is guarded by `!credentials.totpCode`. An attacker who has obtained a valid backup code can send a request with a dummy non-empty `totpCode` (e.g., \"x\") alongside the stolen `backupCode`. The non-empty `totpCode` causes the password check to be skipped (line 121 evaluates to false). Then at line 131, `credentials.backupCode` is truthy so the backup code path is entered \u2014 which succeeds without ever verifying the password. The TOTP path (line 157) is never reached because it's in the `else if`. This allows login with only email + backup code, no password required.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/auth/lib/next-auth-options.ts</file_path>\n<line_numbers>131</line_numbers>\n\n<comment>Password bypass via backup code + dummy totpCode. The password check at line 121 is guarded by `!credentials.totpCode`. An attacker who has obtained a valid backup code can send a request with a dummy non-empty `totpCode` (e.g., \"x\") alongside the stolen `backupCode`. The non-empty `totpCode` causes the password check to be skipped (line 121 evaluates to false). Then at line 131, `credentials.backupCode` is truthy so the backup code path is entered \u2014 which succeeds without ever verifying the password. The TOTP path (line 157) is never reached because it's in the `else if`. This allows login with only email + backup code, no password required.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/auth/lib/next-auth-options.ts</file_path>\n<line_numbers>131</line_numbers>\n\n<comment>Password bypass via backup code + dummy totpCode. The password check at line 121 is guarded by `!credentials.totpCode`. An attacker who has obtained a valid backup code can send a request with a dummy non-empty `totpCode` (e.g., \"x\") alongside the stolen `backupCode`. The non-empty `totpCode` causes the password check to be skipped (line 121 evaluates to false). Then at line 131, `credentials.backupCode` is truthy so the backup code path is entered \u2014 which succeeds without ever verifying the password. The TOTP path (line 157) is never reached because it's in the `else if`. This allows login with only email + backup code, no password required.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:15:58Z"
+          },
+          {
+            "path": "packages/features/auth/lib/next-auth-options.ts",
+            "line": 128,
+            "body": "P3: _Note: Issue is at line 113, shown here on nearest changed line._\n\nBackup code login is blocked for non-CAL identity provider users. The guard at line 113 throws `ThirdPartyIdentityProviderEnabled` when `!credentials.totpCode` is true. When a Google/SAML user with 2FA tries to use a backup code (which means no totpCode is provided), this check blocks the request before the backup code path at line 131 is ever reached. While the UI currently doesn't show the 'Lost access' option for external TOTP flows (ExternalTotpFooter has no such button), this means backup codes are completely non-functional for these users \u2014 defeating the purpose of backup codes as a recovery mechanism.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/auth/lib/next-auth-options.ts</file_path>\n<line_numbers>128</line_numbers>\n\n<comment>_Note: Issue is at line 113, shown here on nearest changed line._\n\nBackup code login is blocked for non-CAL identity provider users. The guard at line 113 throws `ThirdPartyIdentityProviderEnabled` when `!credentials.totpCode` is true. When a Google/SAML user with 2FA tries to use a backup code (which means no totpCode is provided), this check blocks the request before the backup code path at line 131 is ever reached. While the UI currently doesn't show the 'Lost access' option for external TOTP flows (ExternalTotpFooter has no such button), this means backup codes are completely non-functional for these users \u2014 defeating the purpose of backup codes as a recovery mechanism.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/auth/lib/next-auth-options.ts</file_path>\n<line_numbers>128</line_numbers>\n\n<comment>_Note: Issue is at line 113, shown here on nearest changed line._\n\nBackup code login is blocked for non-CAL identity provider users. The guard at line 113 throws `ThirdPartyIdentityProviderEnabled` when `!credentials.totpCode` is true. When a Google/SAML user with 2FA tries to use a backup code (which means no totpCode is provided), this check blocks the request before the backup code path at line 131 is ever reached. While the UI currently doesn't show the 'Lost access' option for external TOTP flows (ExternalTotpFooter has no such button), this means backup codes are completely non-functional for these users \u2014 defeating the purpose of backup codes as a recovery mechanism.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:15:58Z"
+          },
+          {
+            "path": "apps/web/pages/api/auth/two-factor/totp/disable.ts",
+            "line": 51,
+            "body": "P4: Unhandled exception in API route returns unstructured error. When `CALENDSO_ENCRYPTION_KEY` is missing in the backup code path, `throw new Error(ErrorCode.InternalServerError)` causes Next.js to return a generic 500 HTML error page instead of a structured JSON response. The client code (`DisableTwoFactorModal`) calls `response.json()` and checks `body.error`, which would fail on an HTML response, causing a silent client-side error. The existing TOTP path at lines 77/82/90 has the same pre-existing pattern, but the new backup code path introduces another instance.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>apps/web/pages/api/auth/two-factor/totp/disable.ts</file_path>\n<line_numbers>51</line_numbers>\n\n<comment>Unhandled exception in API route returns unstructured error. When `CALENDSO_ENCRYPTION_KEY` is missing in the backup code path, `throw new Error(ErrorCode.InternalServerError)` causes Next.js to return a generic 500 HTML error page instead of a structured JSON response. The client code (`DisableTwoFactorModal`) calls `response.json()` and checks `body.error`, which would fail on an HTML response, causing a silent client-side error. The existing TOTP path at lines 77/82/90 has the same pre-existing pattern, but the new backup code path introduces another instance.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>apps/web/pages/api/auth/two-factor/totp/disable.ts</file_path>\n<line_numbers>51</line_numbers>\n\n<comment>Unhandled exception in API route returns unstructured error. When `CALENDSO_ENCRYPTION_KEY` is missing in the backup code path, `throw new Error(ErrorCode.InternalServerError)` causes Next.js to return a generic 500 HTML error page instead of a structured JSON response. The client code (`DisableTwoFactorModal`) calls `response.json()` and checks `body.error`, which would fail on an HTML response, causing a silent client-side error. The existing TOTP path at lines 77/82/90 has the same pre-existing pattern, but the new backup code path introduces another instance.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:15:58Z"
+          },
+          {
+            "path": "apps/web/components/settings/EnableTwoFactorModal.tsx",
+            "line": 99,
+            "body": "P4: Object URL created for backup codes download is never revoked on component unmount or dialog close. `URL.createObjectURL(textBlob)` at line 99 creates a blob URL that persists until explicitly revoked. While `handleSetup` revokes the previous URL before creating a new one, the `resetState` function (called on dialog close/cancel) does not revoke it, and there is no cleanup effect. Each 2FA setup flow leaks one blob URL.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>apps/web/components/settings/EnableTwoFactorModal.tsx</file_path>\n<line_numbers>99</line_numbers>\n\n<comment>Object URL created for backup codes download is never revoked on component unmount or dialog close. `URL.createObjectURL(textBlob)` at line 99 creates a blob URL that persists until explicitly revoked. While `handleSetup` revokes the previous URL before creating a new one, the `resetState` function (called on dialog close/cancel) does not revoke it, and there is no cleanup effect. Each 2FA setup flow leaks one blob URL.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>apps/web/components/settings/EnableTwoFactorModal.tsx</file_path>\n<line_numbers>99</line_numbers>\n\n<comment>Object URL created for backup codes download is never revoked on component unmount or dialog close. `URL.createObjectURL(textBlob)` at line 99 creates a blob URL that persists until explicitly revoked. While `handleSetup` revokes the previous URL before creating a new one, the `resetState` function (called on dialog close/cancel) does not revoke it, and there is no cleanup effect. Each 2FA setup flow leaks one blob URL.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:15:58Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 4 issues",
+            "created_at": "2026-04-12T02:15:58Z"
+          }
+        ]
       }
     ]
   },
@@ -54543,6 +56228,49 @@
             "created_at": "2026-03-18T20:27:38Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR10967__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10967__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/app-store/googlecalendar/lib/CalendarService.ts",
+            "line": 256,
+            "body": "P1: In both `updateEvent` and `deleteEvent`, when `externalCalendarId` is falsy, the else-branch searches `event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId)` \u2014 but `externalCalendarId` is guaranteed falsy at that point, so the `.find()` will never match any element and always returns `undefined`. The old code correctly fell back to `event.destinationCalendar?.externalId`. This means calendar updates and deletes will use `undefined` as the calendar ID, likely falling back to the wrong calendar or failing.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/googlecalendar/lib/CalendarService.ts</file_path>\n<line_numbers>256</line_numbers>\n\n<comment>In both `updateEvent` and `deleteEvent`, when `externalCalendarId` is falsy, the else-branch searches `event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId)` \u2014 but `externalCalendarId` is guaranteed falsy at that point, so the `.find()` will never match any element and always returns `undefined`. The old code correctly fell back to `event.destinationCalendar?.externalId`. This means calendar updates and deletes will use `undefined` as the calendar ID, likely falling back to the wrong calendar or failing.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/googlecalendar/lib/CalendarService.ts</file_path>\n<line_numbers>256</line_numbers>\n\n<comment>In both `updateEvent` and `deleteEvent`, when `externalCalendarId` is falsy, the else-branch searches `event.destinationCalendar?.find((cal) => cal.externalId === externalCalendarId)` \u2014 but `externalCalendarId` is guaranteed falsy at that point, so the `.find()` will never match any element and always returns `undefined`. The old code correctly fell back to `event.destinationCalendar?.externalId`. This means calendar updates and deletes will use `undefined` as the calendar ID, likely falling back to the wrong calendar or failing.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:12:24Z"
+          },
+          {
+            "path": "packages/core/EventManager.ts",
+            "line": 119,
+            "body": "P1: When `evt.destinationCalendar` is `null` or an empty array, `const [mainHostDestinationCalendar] = evt.destinationCalendar ?? []` sets `mainHostDestinationCalendar` to `undefined`. The next line accesses `mainHostDestinationCalendar.integration` without optional chaining, which throws `TypeError: Cannot read property 'integration' of undefined`. Multiple callers (e.g., webhook handlers, payment handlers) set `destinationCalendar` to `[]` when no calendar is configured.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/core/EventManager.ts</file_path>\n<line_numbers>119</line_numbers>\n\n<comment>When `evt.destinationCalendar` is `null` or an empty array, `const [mainHostDestinationCalendar] = evt.destinationCalendar ?? []` sets `mainHostDestinationCalendar` to `undefined`. The next line accesses `mainHostDestinationCalendar.integration` without optional chaining, which throws `TypeError: Cannot read property 'integration' of undefined`. Multiple callers (e.g., webhook handlers, payment handlers) set `destinationCalendar` to `[]` when no calendar is configured.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/core/EventManager.ts</file_path>\n<line_numbers>119</line_numbers>\n\n<comment>When `evt.destinationCalendar` is `null` or an empty array, `const [mainHostDestinationCalendar] = evt.destinationCalendar ?? []` sets `mainHostDestinationCalendar` to `undefined`. The next line accesses `mainHostDestinationCalendar.integration` without optional chaining, which throws `TypeError: Cannot read property 'integration' of undefined`. Multiple callers (e.g., webhook handlers, payment handlers) set `destinationCalendar` to `[]` when no calendar is configured.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:12:24Z"
+          },
+          {
+            "path": "packages/core/EventManager.ts",
+            "line": 542,
+            "body": "P2: In `updateAllCalendarEvents`, if `reference.credentialId` is truthy but the credential is not found in `this.calendarCredentials` AND the DB lookup fails (null result or no `app.slug`), `credential` remains `undefined`. It is then unconditionally passed to `updateEvent(credential, ...)` on line 542, which expects a `CredentialWithAppName`. This will crash when `updateEvent` calls `getCalendar(credential)` on the undefined value.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/core/EventManager.ts</file_path>\n<line_numbers>542</line_numbers>\n\n<comment>In `updateAllCalendarEvents`, if `reference.credentialId` is truthy but the credential is not found in `this.calendarCredentials` AND the DB lookup fails (null result or no `app.slug`), `credential` remains `undefined`. It is then unconditionally passed to `updateEvent(credential, ...)` on line 542, which expects a `CredentialWithAppName`. This will crash when `updateEvent` calls `getCalendar(credential)` on the undefined value.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/core/EventManager.ts</file_path>\n<line_numbers>542</line_numbers>\n\n<comment>In `updateAllCalendarEvents`, if `reference.credentialId` is truthy but the credential is not found in `this.calendarCredentials` AND the DB lookup fails (null result or no `app.slug`), `credential` remains `undefined`. It is then unconditionally passed to `updateEvent(credential, ...)` on line 542, which expects a `CredentialWithAppName`. This will crash when `updateEvent` calls `getCalendar(credential)` on the undefined value.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:12:24Z"
+          },
+          {
+            "path": "packages/features/bookings/lib/handleNewBooking.ts",
+            "line": 1078,
+            "body": "P2: In `handleNewBooking`, when neither `eventType.destinationCalendar` nor `organizerUser.destinationCalendar` exist, `evt.destinationCalendar` is set to `null` (line 1067). The subsequent `evt.destinationCalendar?.push(...teamDestinationCalendars)` (line 1078) is then a no-op due to optional chaining. For collective team events where the organizer lacks a calendar but team members have destination calendars, those team member calendars are silently dropped \u2014 the event will have no destination calendars despite team members having them configured.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/bookings/lib/handleNewBooking.ts</file_path>\n<line_numbers>1078</line_numbers>\n\n<comment>In `handleNewBooking`, when neither `eventType.destinationCalendar` nor `organizerUser.destinationCalendar` exist, `evt.destinationCalendar` is set to `null` (line 1067). The subsequent `evt.destinationCalendar?.push(...teamDestinationCalendars)` (line 1078) is then a no-op due to optional chaining. For collective team events where the organizer lacks a calendar but team members have destination calendars, those team member calendars are silently dropped \u2014 the event will have no destination calendars despite team members having them configured.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/bookings/lib/handleNewBooking.ts</file_path>\n<line_numbers>1078</line_numbers>\n\n<comment>In `handleNewBooking`, when neither `eventType.destinationCalendar` nor `organizerUser.destinationCalendar` exist, `evt.destinationCalendar` is set to `null` (line 1067). The subsequent `evt.destinationCalendar?.push(...teamDestinationCalendars)` (line 1078) is then a no-op due to optional chaining. For collective team events where the organizer lacks a calendar but team members have destination calendars, those team member calendars are silently dropped \u2014 the event will have no destination calendars despite team members having them configured.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:12:24Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/organizations/create.handler.ts",
+            "line": 151,
+            "body": "P3: The slug/requestedSlug logic was inverted from the original. Old code: `...(!IS_TEAM_BILLING_ENABLED && { slug })` set `slug` when billing is DISABLED (org gets slug immediately, no payment needed). New code: `...(IS_TEAM_BILLING_ENABLED ? { slug } : {})` sets `slug` when billing is ENABLED. Combined with the `requestedSlug` line (which also only fires when billing is enabled), organizations without billing now get neither `slug` nor `requestedSlug`, while organizations with billing get both \u2014 the opposite of the original intent.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/organizations/create.handler.ts</file_path>\n<line_numbers>151</line_numbers>\n\n<comment>The slug/requestedSlug logic was inverted from the original. Old code: `...(!IS_TEAM_BILLING_ENABLED && { slug })` set `slug` when billing is DISABLED (org gets slug immediately, no payment needed). New code: `...(IS_TEAM_BILLING_ENABLED ? { slug } : {})` sets `slug` when billing is ENABLED. Combined with the `requestedSlug` line (which also only fires when billing is enabled), organizations without billing now get neither `slug` nor `requestedSlug`, while organizations with billing get both \u2014 the opposite of the original intent.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/organizations/create.handler.ts</file_path>\n<line_numbers>151</line_numbers>\n\n<comment>The slug/requestedSlug logic was inverted from the original. Old code: `...(!IS_TEAM_BILLING_ENABLED && { slug })` set `slug` when billing is DISABLED (org gets slug immediately, no payment needed). New code: `...(IS_TEAM_BILLING_ENABLED ? { slug } : {})` sets `slug` when billing is ENABLED. Combined with the `requestedSlug` line (which also only fires when billing is enabled), organizations without billing now get neither `slug` nor `requestedSlug`, while organizations with billing get both \u2014 the opposite of the original intent.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:12:24Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:12:24Z"
+          }
+        ]
       }
     ]
   },
@@ -55622,6 +57350,49 @@
             "created_at": "2026-03-18T20:27:49Z"
           }
         ]
+      },
+      {
+        "tool": "tusk",
+        "repo_name": "cal_dot_com__cal.com__tusk__PR8087__20260411",
+        "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8087__20260411/pull/1",
+        "review_comments": [
+          {
+            "path": "packages/app-store/vital/lib/reschedule.ts",
+            "line": 125,
+            "body": "P1: async callback inside forEach is fire-and-forget. The forEach does not await the async callback, so: (1) the try/catch block on line 124 will never catch errors from the async operations, (2) the deleteEvent/deleteMeeting calls are not awaited before the function proceeds to send reschedule emails on line 143, meaning calendar/video deletions may silently fail or race with subsequent operations.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/vital/lib/reschedule.ts</file_path>\n<line_numbers>125</line_numbers>\n\n<comment>async callback inside forEach is fire-and-forget. The forEach does not await the async callback, so: (1) the try/catch block on line 124 will never catch errors from the async operations, (2) the deleteEvent/deleteMeeting calls are not awaited before the function proceeds to send reschedule emails on line 143, meaning calendar/video deletions may silently fail or race with subsequent operations.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/vital/lib/reschedule.ts</file_path>\n<line_numbers>125</line_numbers>\n\n<comment>async callback inside forEach is fire-and-forget. The forEach does not await the async callback, so: (1) the try/catch block on line 124 will never catch errors from the async operations, (2) the deleteEvent/deleteMeeting calls are not awaited before the function proceeds to send reschedule emails on line 143, meaning calendar/video deletions may silently fail or race with subsequent operations.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:10:15Z"
+          },
+          {
+            "path": "packages/app-store/wipemycalother/lib/reschedule.ts",
+            "line": 125,
+            "body": "P1: Same async-forEach fire-and-forget bug as in vital/lib/reschedule.ts. The async callback's promises are discarded, so calendar deleteEvent calls are unawaited. Errors won't be caught and the function proceeds immediately to send reschedule emails.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/app-store/wipemycalother/lib/reschedule.ts</file_path>\n<line_numbers>125</line_numbers>\n\n<comment>Same async-forEach fire-and-forget bug as in vital/lib/reschedule.ts. The async callback's promises are discarded, so calendar deleteEvent calls are unawaited. Errors won't be caught and the function proceeds immediately to send reschedule emails.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/app-store/wipemycalother/lib/reschedule.ts</file_path>\n<line_numbers>125</line_numbers>\n\n<comment>Same async-forEach fire-and-forget bug as in vital/lib/reschedule.ts. The async callback's promises are discarded, so calendar deleteEvent calls are unawaited. Errors won't be caught and the function proceeds immediately to send reschedule emails.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:10:16Z"
+          },
+          {
+            "path": "packages/trpc/server/routers/viewer/bookings.tsx",
+            "line": 553,
+            "body": "P1: Same async-forEach fire-and-forget bug. The forEach on line 553 uses an async callback but forEach ignores the returned promises. The deleteEvent/deleteMeeting operations are unawaited, and the code immediately proceeds to send reschedule emails on line 570. Any errors from the async operations are silently swallowed.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/trpc/server/routers/viewer/bookings.tsx</file_path>\n<line_numbers>553</line_numbers>\n\n<comment>Same async-forEach fire-and-forget bug. The forEach on line 553 uses an async callback but forEach ignores the returned promises. The deleteEvent/deleteMeeting operations are unawaited, and the code immediately proceeds to send reschedule emails on line 570. Any errors from the async operations are silently swallowed.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/trpc/server/routers/viewer/bookings.tsx</file_path>\n<line_numbers>553</line_numbers>\n\n<comment>Same async-forEach fire-and-forget bug. The forEach on line 553 uses an async callback but forEach ignores the returned promises. The deleteEvent/deleteMeeting operations are unawaited, and the code immediately proceeds to send reschedule emails on line 570. Any errors from the async operations are silently swallowed.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:10:16Z"
+          },
+          {
+            "path": "packages/features/bookings/lib/handleCancelBooking.ts",
+            "line": 460,
+            "body": "P1: async-forEach fire-and-forget in handleCancelBooking. The forEach on line 460 uses an async callback, but forEach discards the promises. The `apiDeletes` array is populated inside the async callback, but since the callback is not awaited, `apiDeletes` will be empty when `Promise.all(apiDeletes)` is called later, causing calendar deletions for recurring events to be silently skipped.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/features/bookings/lib/handleCancelBooking.ts</file_path>\n<line_numbers>460</line_numbers>\n\n<comment>async-forEach fire-and-forget in handleCancelBooking. The forEach on line 460 uses an async callback, but forEach discards the promises. The `apiDeletes` array is populated inside the async callback, but since the callback is not awaited, `apiDeletes` will be empty when `Promise.all(apiDeletes)` is called later, causing calendar deletions for recurring events to be silently skipped.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/features/bookings/lib/handleCancelBooking.ts</file_path>\n<line_numbers>460</line_numbers>\n\n<comment>async-forEach fire-and-forget in handleCancelBooking. The forEach on line 460 uses an async callback, but forEach discards the promises. The `apiDeletes` array is populated inside the async callback, but since the callback is not awaited, `apiDeletes` will be empty when `Promise.all(apiDeletes)` is called later, causing calendar deletions for recurring events to be silently skipped.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:10:16Z"
+          },
+          {
+            "path": "packages/core/CalendarManager.ts",
+            "line": 28,
+            "body": "P2: getCalendarCredentials is a synchronous function that calls getCalendar (now async) inside flatMap on line 28. Since getCalendar now returns Promise<Calendar | null>, the `calendar` field in each returned object is a Promise, not a Calendar. This is silently stored and later awaited in getConnectedCalendars (line 48), which happens to work there, but the type contract is broken \u2014 the return type claims `calendar` is `Calendar | null` when it's actually `Promise<Calendar | null>`. Any caller that doesn't await `item.calendar` will get a truthy Promise object instead of null, bypassing null checks.\n\n<details>\n<summary>AI prompts</summary>\n\n**Check prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and let me know. Don't update any code just yet.\n\n<file_path>packages/core/CalendarManager.ts</file_path>\n<line_numbers>28</line_numbers>\n\n<comment>getCalendarCredentials is a synchronous function that calls getCalendar (now async) inside flatMap on line 28. Since getCalendar now returns Promise<Calendar | null>, the `calendar` field in each returned object is a Promise, not a Calendar. This is silently stored and later awaited in getConnectedCalendars (line 48), which happens to work there, but the type contract is broken \u2014 the return type claims `calendar` is `Calendar | null` when it's actually `Promise<Calendar | null>`. Any caller that doesn't await `item.calendar` will get a truthy Promise object instead of null, bypassing null checks.</comment>\n```\n\n**Check + fix prompt**\n```text\nCheck if this issue is valid. If so, find the root cause and fix it.\n\n<file_path>packages/core/CalendarManager.ts</file_path>\n<line_numbers>28</line_numbers>\n\n<comment>getCalendarCredentials is a synchronous function that calls getCalendar (now async) inside flatMap on line 28. Since getCalendar now returns Promise<Calendar | null>, the `calendar` field in each returned object is a Promise, not a Calendar. This is silently stored and later awaited in getConnectedCalendars (line 48), which happens to work there, but the type contract is broken \u2014 the return type claims `calendar` is `Calendar | null` when it's actually `Promise<Calendar | null>`. Any caller that doesn't await `item.calendar` will get a truthy Promise object instead of null, bypassing null checks.</comment>\n```\n\n</details>",
+            "created_at": "2026-04-12T02:10:16Z"
+          },
+          {
+            "path": null,
+            "line": null,
+            "body": "Found 5 issues",
+            "created_at": "2026-04-12T02:10:15Z"
+          }
+        ]
       }
     ]
   }
diff --git a/offline/results/openai_gpt-5.2/candidates.json b/offline/results/openai_gpt-5.2/candidates.json
index 67a86fc..e38233b 100644
--- a/offline/results/openai_gpt-5.2/candidates.json
+++ b/offline/results/openai_gpt-5.2/candidates.json
@@ -1103,6 +1103,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Lithuanian (lt) translation string is incorrectly in Italian ('Installa una delle seguenti applicazioni sul tuo cellulare:') in account messages file, so Lithuanian users see Italian text during TOTP setup",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Same incorrect Italian string appears in the Lithuanian (lt) login message file as well, affecting the TOTP setup step there too",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Chinese Simplified (zh_CN) account translation uses Traditional Chinese characters ('\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a') instead of Simplified, creating an incorrect/inconsistent zh_CN locale translation",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "VerifyMessageProperties.containsHtml() checks the English value before stripping anchor tags while translations have anchors sanitized, causing POLICY_SOME_HTML to be selected when English contains only <a> tags and unintentionally allowing other HTML tags (e.g., <br>, <p>, <strong>, <b>) in translations contrary to the intended policy",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/37634": {
@@ -1515,6 +1541,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Null-check bug in AccessTokenContext: Objects.requireNonNull is called on grantType with message \"Null rawTokenId not allowed\" instead of checking rawTokenId, so rawTokenId can be null and cause NPEs later (e.g., in encodeTokenId concatenation).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "isAccessTokenId matcher uses wrong substring indices: items[0].substring(3, 5) extracts the wrong characters for the 2-char grantType shortcut; it should extract positions 4-5 (substring(4, 6)) given the 6-char prefix layout.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "isAccessTokenId matcher has inverted grant-type comparison logic: it returns false when the extracted grant shortcut equals the expected value, instead of returning false when it does not match (missing negation).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/38446": {
@@ -1839,6 +1885,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getCredentials() creates a RecoveryAuthnCodesCredentialModel with a null ID because createFromValues() doesn't set the stored credential ID, so downstream removeStoredCredentialById(recoveryCodeCredentialModel.getId()) is called with null when codes are exhausted",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BackwardsCompatibilityUserStorage.isValid() for recovery codes accepts any matching code (anyMatch) instead of only the next sequential code, diverging from production behavior and allowing out-of-order validation",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BackwardsCompatibilityUserStorage.isValid() does not consume/remove a recovery code after successful validation, allowing the same code to be reused indefinitely and preventing tests from catching one-time-use behavior regressions",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "RecoveryAuthnCodesFormAuthenticator removes exhausted recovery-code credentials via removeStoredCredentialById() which targets local storage; for federated users the credential resides in federated storage so removal silently fails, leaving a stale exhausted credential that can cause a loop where allCodesUsed() remains true on subsequent logins",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BackwardsCompatibilityUserStorage.updateCredential() stores raw recovery codes JSON in credentialData (instead of standard metadata in credentialData and hashed codes in secretData), and isValid()/getCredentials() depend on this nonstandard format, making the implementation fragile and likely to break if storage is changed to the standard model format",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/36882": {
@@ -2032,7 +2110,27 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "Operator advanced-configuration documentation incorrectly claims the Operator will fail unless the `rolling-updates` feature is enabled, but the Operator code never checks `Profile.Feature.ROLLING_UPDATES` or runs `kc.sh update-compatibility` and instead decides rolling vs recreate purely via StatefulSet image/args/env comparisons",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "No test verifies the `FEATURE_DISABLED` exit code (4) for the feature-disabled scenario because `testFeatureNotEnabled` uses `@Launch` where `picocli.exit()` does not call `System.exit`, so only the error message is asserted and exit code 4 could change without failing tests",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Feature-disabled behavior is only tested for the `metadata` subcommand; the `check` subcommand has its own independent feature check path that is never covered by tests",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/keycloak/keycloak/pull/36880": {
     "claude": [
@@ -2479,6 +2577,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Potential NullPointerException: getResourceTypeResource() may return null but its result is passed to PolicyStore.findByResource(server, resource) without a null check, violating the non-null contract and causing an NPE",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ClientPermissionsV2.getClientsWithPermission uses findByType(server, \"Clients\", ...) but per-client resources created via getOrCreateResource never set resource.setType(), so per-client permissions are not found and the method can return an empty set incorrectly",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getClientsWithPermission may include the all-clients type resource name \"Clients\" in the returned set; downstream AvailableRoleMappingResource.getRoleIdsWithPermissions calls realm.getClientById(\"Clients\") which returns null, leading to an NPE when calling .getRolesStream()",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ClientPermissionsV2.canMapRoles/canMapCompositeRoles/canMapClientScopeRoles do not check the MANAGE_CLIENTS admin role, preventing users with that admin role from mapping roles unless they also have fine-grained permissions",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Incorrect Javadoc: requireView() documentation claims it returns boolean, but the method is void and should be documented as throwing ForbiddenException when canView() is false",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Incorrect Javadoc: requireViewClientScopes() documentation claims it returns boolean, but the method is void and should be documented as throwing ForbiddenException when the check fails",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/37038": {
@@ -2858,6 +2994,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "GroupPermissionsV2.canManage() (no-arg) incorrectly checks both VIEW and MANAGE scopes, allowing users with only VIEW permission to pass manage checks (e.g., gating group creation via requireManage())",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getGroupIdsWithViewPermission() uses groupResource.getId() (authorization resource internal UUID) instead of groupResource.getName() (actual group model ID), breaking per-group permission evaluation because resourceStore.findByName() won\u2019t find resources by the internal UUID",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getGroupIdsWithViewPermission() adds authorization resource UUIDs to the granted set, but downstream queries compare these values to group model IDs (e.g., groupId column), so permission-based filtering never matches",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "GroupPermissionEvaluator.canManage() docstring claims VIEW or MANAGE is sufficient, which becomes misleading if canManage() is corrected to require MANAGE (or indicates an intentional but insecure design if left as-is)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Tests call ApiUtil.handleCreatedResponse(response) inside an outer try-with-resources even though handleCreatedResponse also closes the response, causing double-close/double-assert and potentially unsafe access to headers/streams after closure (pattern repeated at multiple line ranges)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/33832": {
@@ -3298,6 +3466,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Adding `order()` to the public `CryptoProvider` SPI interface without a default implementation breaks binary/source compatibility for third-party `ServiceLoader` implementations (they will fail to compile/load)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`ECDSAAlgorithmTest` reuses a single P-256 EC key pair for ES384 and ES512 tests, so signatures have ~32-byte r/s values instead of the required 48/66 bytes, meaning ES384/ES512-specific conversion bugs may not be caught",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`AuthzClientCryptoProvider.concatenatedRSToASN1DER` contains dead code that creates `ASN1Encoder` instances and writes integers but discards the results, causing unnecessary allocations each call",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`CryptoIntegration.detectProvider` builds the \"Ignored crypto providers\" log message with an unconditional trailing \", \" after the last provider name, producing malformed log output",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/keycloak/keycloak/pull/40940": {
@@ -3655,6 +3849,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "getSubGroupsStream(String, Integer, Integer) calls modelSupplier.get().getSubGroupsStream(...) without checking if modelSupplier.get() is null, risking an NPE under concurrent-deletion scenarios",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The getSubGroupsStream overload at line 262 calls modelSupplier.get() without a null check, risking an NPE under concurrent-deletion scenarios",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The getSubGroupsStream overload at line 268 calls modelSupplier.get() without a null check, risking an NPE under concurrent-deletion scenarios",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test in GroupTest asserts caughtExceptions is empty without joining the reader thread first, so exceptions from an in-flight groups() call may be recorded after the assertion and the test can miss failures",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1": {
@@ -4031,6 +4251,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "UsernameForm.java calls isConditionalPasskeysEnabled() with no arguments, but the parent method signature changed to isConditionalPasskeysEnabled(UserModel user), causing a compilation error (should pass context.getUser())",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Regression in UsernamePasswordForm.authenticate(): fillContextForm is no longer called when context.getUser() is null (initial login), so WebAuthn challenge/rpId/conditional UI attributes are not set on the initial login form, breaking passkey conditional UI setup",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Regression in UsernamePasswordForm.challenge(error path): fillContextForm is no longer called when context.getUser() is null, so after an invalid username/password on initial login the error page does not re-render passkey UI elements, preventing switching to passkey login",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Potential future NullPointerException risk in WebAuthnAuthenticator.fillContextForm(): user is passed to WebAuthnAuthenticatorsBean when shouldShowWebAuthnAuthenticators(context) is true; since shouldShowWebAuthnAuthenticators is protected/overridable, a subclass could return true without ensuring a non-null user, leading to null user usage",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/93824": {
@@ -4383,6 +4629,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Metrics tag key is inconsistent: uses \"shards\" instead of \"shard\", causing the metric to be missing from dashboards/alerts filtering on the \"shard\" tag",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When join() exceeds the deadline, a break exits the loop and skips calling .terminate() on remaining processes, leaving potentially hung processes alive (zombies)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Each sub-buffer calls get_memory_info() which queries Redis INFO on all cluster nodes; doing this once per sub-buffer causes redundant roundtrips and adds unnecessary latency on the submit() hot path",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5": {
@@ -4834,6 +5100,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "fetch_error_details incorrectly zips error_ids with events.values() even though nodestore.backend.get_multi() returns a dict keyed by node_ids, causing error IDs to be paired with the wrong event data when dict insertion order differs",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "nodestore/base.py items.update(cache_items) can append cached entries at the end of the dict, breaking insertion-order alignment with the original requested IDs and contributing to mis-association when relying on dict.values() ordering",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "gen_request_data compares error event timestamps in Unix epoch seconds to rrweb replay event timestamps in Unix epoch milliseconds, causing incorrect chronological ordering (errors always appear far earlier and are yielded before replay breadcrumbs)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Unit test masks the timestamp unit mismatch by using the same small float timestamp values for both error events and replay events, so it passes despite the bug",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1": {
@@ -5433,6 +5725,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil on a datetime value (from getattr(item, self.key) when order_by='-datetime'), causing a TypeError; it should convert datetime to a numeric timestamp first (as DateTimePaginator does).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When enable_advanced_features is True and cursor.offset is negative, BasePaginator/OptimizedCursorPaginator can slice a Django queryset with a negative start (queryset[start_offset:stop]), which raises AssertionError because Django querysets do not support negative indexing; the comment claiming Django handles negative slicing is incorrect.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OptimizedCursorPaginator.value_from_cursor returns the raw numeric cursor.value without converting it back to a datetime; when ordering/filtering by a DateTimeField (e.g., datetime__lte=<int>), this can produce incorrect results or a database type error (unlike DateTimePaginator which converts back to a timezone-aware datetime).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In organization_auditlogs.py, organization_context.member can be None for users without org membership; accessing organization_context.member.has_global_access will raise AttributeError, and this check runs on every request due to being evaluated before the use_optimized guard.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BasePaginator allows negative start_offset when cursor.is_prev is True (start_offset = offset without clamping), so a malformed/malicious cursor with negative offset can trigger Django AssertionError during queryset slicing; a refactor/rename obscured this safety behavior change.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/97529": {
@@ -5787,6 +6111,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Race condition due to reduced cacheMu lock scope in BuildIndex/getOrCreateIndex allowing concurrent callers to build the same index key simultaneously",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "For file-based indexes, concurrent builds can cause the second bleve.New(dir, mapper) to fail because the index directory already exists",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "For in-memory indexes, concurrent builds cause duplicate work and the first built index can be silently overwritten in the cache",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Data race/correctness bug: shared outer err variable is captured by multiple errgroup goroutines and concurrently written/read, so a goroutine may return another goroutine\u2019s error (or nil)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Data race: totalBatchesIndexed is incremented from multiple errgroup goroutines without synchronization when initWorkers > 1",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/80168": {
@@ -6105,7 +6461,45 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "MetricAlertDetectorHandler now subclasses StatefulDetectorHandler but does not implement required abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), so instantiating it will raise a TypeError when Detector.detector_handler constructs the handler",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "TestEvaluateGroupKeyValue.test_dedupe builds the expected occurrence using group_key=\"val1\" but calls evaluate_group_key_value with group_key=\"group_key\", so the expected fingerprint/group_key does not match the actual code path",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "TestEvaluateGroupKeyValue.test_dedupe is ineffective because IssueOccurrence.__eq__ compares only the hardcoded id, allowing the test to pass even when fingerprint/group_key are wrong",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_state_results_multi_group builds the expected occurrence for group_2 with value=6 even though the input packet has group_2=10 and the real code forwards value=10",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_state_results_multi_group can pass incorrectly because the mock occurrence builder does not vary output based on value and IssueOccurrence.__eq__ compares only by id",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_state_results constructs expected occurrences using detector.detector_handler, but detector.detector_handler returns a new handler instance on each access, so the expected handler instance differs from the one used in process_detectors, making the test fragile if handler instance state matters",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/getsentry/sentry/pull/80528": {
     "gemini": [
@@ -6331,6 +6725,14 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "get_monitor_environment_context returns monitor_environment.monitor.config (original config) instead of the locally modified config copy, so the schedule_type human-readable override is discarded and the context contains the raw schedule_type value",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/77754": {
@@ -6593,6 +6995,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Dataclass field `queued` uses `timezone.now()` as a default value, causing it to be evaluated at import time so instances share the same timestamp; should use `field(default_factory=timezone.now)`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`assignment_source` parameter was added to the abstract `sync_status_outbound` method even though status sync is unrelated to this PR\u2019s assignee sync changes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` does not accept `**kwargs`, so callers passing the new `assignment_source` argument will raise a TypeError",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test name `test_from_dict_empty_array` is misleading because it tests an empty dict `{}` rather than an empty array",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test name `test_from_dict_inalid_data` contains a typo ('inalid' instead of 'invalid'), making it harder to search and understand",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/95633": {
@@ -6960,6 +7394,56 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "close() stops/joins the commit thread before draining/shutting down the worker pool, so messages processed during worker drain have side effects but their offsets are never committed, causing duplicate processing after restart",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_thread_queue_parallel_error_handling has a docstring claiming errors shouldn't block commits for other messages, but the assertion checks that no commits happened (opposite of the stated intent)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "test_thread_queue_parallel_error_handling is timing-dependent/flaky because the commit loop runs every ~1s while the test only waits ~0.2s, so it can pass simply because commits haven't occurred yet",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OrderedQueueWorker.run() marks offsets complete in a finally block even when processing raises, so failed messages are treated as successfully processed for committing purposes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SimpleQueueProcessingStrategy docstring incorrectly claims natural backpressure, but worker queues are created unbounded (queue.Queue() without maxsize), allowing unbounded memory growth under load",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SimpleQueueProcessingStrategy docstring incorrectly claims offsets are only committed after successful processing, but offsets are completed/committed even on exceptions due to the finally block behavior",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OffsetTracker._get_partition_lock accesses/creates entries in partition_locks without synchronization, relying on CPython GIL behavior rather than guaranteed thread-safety",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "OffsetTracker.get_committable_offsets iterates all_offsets keys without holding a lock while other threads may mutate all_offsets (including via defaultdict auto-creation), relying on non-guaranteed CPython atomicity",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2": {
@@ -7501,6 +7985,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "OptimizedCursorPaginator.get_item_key applies math.floor()/math.ceil() directly to a datetime value (from order_by=\"-datetime\"), which will raise a TypeError; it should convert the datetime to a numeric value first (as DateTimePaginator does).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Optimized pagination path slices a Django queryset with a negative start index (queryset[negative_offset:stop]); Django ORM does not support negative indexing and raises AssertionError, so a crafted cursor with negative offset can trigger a 500 error.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "A flush-time safety guard that skipped segments exceeding max_segment_spans (with metrics/logging) was removed; this eliminates protection against edge cases where segments can exceed the cap despite the Lua insert-side limit.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "max_segment_spans is still initialized on the class but is now unused/dead code after the flush-side guard removal.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3": {
@@ -7942,6 +8452,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Cache key construction uses Python built-in hash(), which is randomized per process (PYTHONHASHSEED), causing cross-process cache key mismatches and cache misses in shared Redis/Memcached",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "invalidate_upsampling_cache also uses non-deterministic hash() in its cache key, so invalidation from one process may not clear entries written by another process",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Cache key for upsampling decision does not include dataset/request context, so a cached True/False can be reused across different query contexts and yield inconsistent behavior",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Logic change from any(project_id in allowlist) to all(project_id in allowlist) alters semantics and can prevent upsampling for mixed allowlisted/non-allowlisted project queries, leading to underestimated counts",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Removed validation in organization_events_stats.py (top_events > 0 branch) that disallowed grouping by 'timestamp' (ParseError check), allowing previously forbidden groupby and potentially causing unexpected behavior or downstream errors",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/103633": {
@@ -8309,6 +8851,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Cache key collision risk in userPermDenialCacheKey because it concatenates name and parent with '_' even though both fields may contain underscores, allowing different resources to produce identical denial cache keys and cause incorrect access denials",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test \"Should deny on explicit cache deny entry\" is invalid/misleading because it sets the permission cache entry to false while claiming it allows access, so the test would deny via the permission cache regardless of whether the denial cache is checked and therefore doesn't prove denial-cache priority",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "permDenialCache and permCache share the same underlying cache instance while storing different serialized Go types (bool vs map[string]bool) and use overlapping key prefixes, so an accidental key collision could cause json.Unmarshal type mismatches, spurious cache misses, and warning logs",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "permissionCacheUsage metric increments with label \"false\" when the permission cache is hit but the requested resource isn't in the cached permission set, conflating a cache miss with a cache hit containing stale/incomplete data and reducing monitoring clarity",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/getsentry/sentry/pull/67876": {
@@ -8645,53 +9213,85 @@
         "source": "extracted"
       },
       {
-        "text": "post_install schedules Celery tasks (migrate_repo/link_all_repos) fire-and-forget without explicit error handling or result propagation, so failures can go unnoticed and provide no user feedback",
+        "text": "post_install schedules Celery tasks (migrate_repo/link_all_repos) fire-and-forget without explicit error handling or result propagation, so failures can go unnoticed and provide no user feedback",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "pipeline_advancer hard-codes special-case redirect logic for provider_id == 'github', potentially diverging from prior FORWARD_INSTALL_FOR behavior and changing routing semantics in edge cases",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ],
+    "greptile-v4": [
+      {
+        "text": "Unhandled exception from get_user_info(): it calls resp.raise_for_status() and can raise requests.HTTPError for non-2xx GitHub API responses, which is not caught and will bubble up as a 500 instead of rendering the intended error page",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Potential KeyError when accessing integration.metadata[\"sender\"][\"login\"] if the Integration metadata lacks a \"sender\" key (e.g., legacy records or race/incomplete webhook data), causing a 500; should use .get() with a fallback",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "safe_urlopen() is called outside the try/except block even though it can raise on connection/timeout/DNS failures, leading to unhandled 500s; it should be moved inside the exception-handling block",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "User identity verification is skipped on first-time installations: when Integration.DoesNotExist is raised, the code calls pipeline.next_step() without validating github_authenticated_user against the installer, bypassing the intended security check",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ],
+    "macroscope": [
+      {
+        "text": "Test `test_installation_not_found` uses a mismatched OAuth `state` value, so it passes due to state validation failure instead of exercising the installation-not-found (404) path",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "pipeline_advancer hard-codes special-case redirect logic for provider_id == 'github', potentially diverging from prior FORWARD_INSTALL_FOR behavior and changing routing semantics in edge cases",
+        "text": "`github/integration.py` directly accesses `integration.metadata[\"sender\"][\"login\"]`, which can raise `KeyError` when the Integration metadata lacks a `sender` key because `build_integration` only sets it conditionally",
         "path": null,
         "line": null,
         "source": "extracted"
       }
     ],
-    "greptile-v4": [
+    "tusk": [
       {
-        "text": "Unhandled exception from get_user_info(): it calls resp.raise_for_status() and can raise requests.HTTPError for non-2xx GitHub API responses, which is not caught and will bubble up as a 500 instead of rendering the intended error page",
+        "text": "OAuth2 `state` parameter is set to `pipeline.signature`, a deterministic MD5 of pipeline view class names, making it static across users/sessions and allowing attackers to forge `state` and bypass CSRF protection; `state` should include a cryptographically random, session-bound nonce",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "Potential KeyError when accessing integration.metadata[\"sender\"][\"login\"] if the Integration metadata lacks a \"sender\" key (e.g., legacy records or race/incomplete webhook data), causing a 500; should use .get() with a fallback",
+        "text": "Accessing `integration.metadata[\"sender\"][\"login\"]` can raise `KeyError` when `sender` is missing (e.g., integrations created before sender metadata existed or via non-webhook paths), causing a 500 instead of a user-facing error",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "safe_urlopen() is called outside the try/except block even though it can raise on connection/timeout/DNS failures, leading to unhandled 500s; it should be moved inside the exception-handling block",
+        "text": "When `Integration.DoesNotExist` occurs, the code calls `pipeline.next_step()` without performing the authenticated sender/user validation, allowing first-time installations to bypass the intended security check",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "User identity verification is skipped on first-time installations: when Integration.DoesNotExist is raised, the code calls pipeline.next_step() without validating github_authenticated_user against the installer, bypassing the intended security check",
-        "path": null,
-        "line": null,
-        "source": "extracted"
-      }
-    ],
-    "macroscope": [
-      {
-        "text": "Test `test_installation_not_found` uses a mismatched OAuth `state` value, so it passes due to state validation failure instead of exercising the installation-not-found (404) path",
+        "text": "`redirect_uri` is included in the OAuth authorization request but omitted from the token exchange request, violating RFC 6749 requirements and potentially causing token exchange failures or unexpected provider behavior",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "`github/integration.py` directly accesses `integration.metadata[\"sender\"][\"login\"]`, which can raise `KeyError` when the Integration metadata lacks a `sender` key because `build_integration` only sets it conditionally",
+        "text": "`error()` assumes `org.organization` exists even though `self.active_organization` may be `None` or an unexpected type if `determine_active_organization()` wasn't called or returns a different structure, leading to potential crashes in error handling",
         "path": null,
         "line": null,
         "source": "extracted"
@@ -8950,6 +9550,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Test cleanup registers IDP removal using the literal string \"alias\" instead of the actual generated IDP alias (e.g., \"idp-alias-\" + i), so the IDPs created in the loop are not cleaned up and can leak state into other tests",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test cleanup at line 425 similarly uses an incorrect alias for removal (hardcoded/mismatched for \"idp-alias-20\"), leaving that IDP uncleared after the test",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In getForLogin cached path, results are returned from a HashSet<IdentityProviderModel>, which does not preserve delegate query ordering, causing inconsistent ordering behavior between cached and non-cached paths for callers that might rely on ordering",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In registerIDPLoginInvalidationOnUpdate, the \"original\" model is org-aware wrapped (isEnabled depends on org state) while the \"updated\" model is raw; comparing them via getLoginPredicate can differ solely due to wrapper behavior, causing unnecessary (over-)invalidation of the login cache",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/94942": {
@@ -9398,6 +10024,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "`enableSqlExpressions` always returns `false` regardless of `FlagSqlExpressions`, making the feature-flag check dead code and misleading about the ability to re-enable SQL expressions",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`sql_command.go:Execute` appends an (empty) result frame to `rsp.Values` even when `DB.QueryFramesInto` returns an error (\"not implemented\"), leading to results containing empty frames alongside an error",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/90939": {
@@ -9642,6 +10282,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "After releasing the read lock and acquiring the write lock, the code does not re-check whether `entryPointAssetsCache` was populated by another goroutine, leading to redundant manifest reads/parses (double-checked locking bug)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In non-dev mode under concurrent startup requests, the missing second cache check can cause a thundering herd of redundant file I/O when multiple goroutines re-read the manifest unnecessarily",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/80329": {
@@ -10089,6 +10743,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Cleanup service ticker interval was changed from 10 minutes to 1 minute, likely an accidental debug leftover, causing cleanup to run 10x more frequently and increasing production database load",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Cleanup ticker interval (1 minute) is shorter than the clean() method timeout (9 minutes), so a new cleanup cycle can start while the previous one is still running (overlapping executions)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Cleanup-path diagnostic logs incorrectly use r.log.Error() for informational messages, which will flood error logs and potentially trigger alerts on successful cleanup cycles",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SQLite parameter limit guard checks r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit instead of len(ids) > sqliteParameterLimit, causing unnecessary fallback to inline SQL when few IDs are returned and reducing safety vs parameterized queries",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test helper createTestAnnotations hardcodes annotation IDs starting at 1, which can conflict with SQLite auto-increment behavior across shared test DB usage and lead to ID collisions when other tests insert rows without explicit IDs",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/90045": {
@@ -10749,6 +11435,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Create error path records latency with `recordLegacyDuration` instead of `recordStorageDuration`, misattributing a failed storage operation and corrupting metrics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Update error path records latency with `recordLegacyDuration` instead of `recordStorageDuration`, misattributing a failed storage operation and corrupting metrics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "DeleteCollection goroutine records legacy write latency with `recordStorageDuration` instead of `recordLegacyDuration`, misattributing legacy latency to storage metrics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Delete success path passes `name` as the metric label to `recordStorageDuration` instead of `options.Kind`, creating high-cardinality Prometheus labels",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Delete method calls `klog.NewContext(ctx, d.Log)` with the base logger instead of the enriched `log` containing name/kind/method, losing contextual logging",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/106778": {
@@ -11287,6 +12005,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Missing React `key` prop on `GrafanaRuleListItem` when rendering the filter view list, which can cause React reconciliation problems on updates/reordering",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Delete menu item never renders in the new list view because `AlertRuleMenu` guards it with `canDelete && rulerRule && (...)`, but `rulerRule` is always undefined from the `GrafanaRuleListItem -> RuleActionsButtons.V2 -> AlertRuleMenu` path",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Pause menu item never renders in the new list view because it requires `rulerRuleType.grafana.rule(rulerRule)` but `rulerRule` is undefined in the new list view path",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Silence drawer never opens in the new list view because `RuleActionsButtons.V2` only renders the drawer when `rulerRuleType.grafana.alertingRule(rule)` is true, but `rule` (RulerRuleDTO) is undefined when only `promRule` is passed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`useGrafanaPromRuleAbility` is called with `promRule` typed as `Rule` instead of `GrafanaPromRuleDTO | SkipToken`; the boolean check `prometheusRuleType.grafana.rule()` does not narrow the type, risking a TypeScript error or incorrect permission checks when fields like `folderUid` are missing/optional",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/grafana/grafana/pull/107534": {
@@ -11553,7 +12303,39 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "Queries are interpolated twice in the shard-splitting path because both runShardSplitQuery() and runSplitQuery() call applyTemplateVariables(), causing non-idempotent interpolation side effects",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Ad-hoc filters are appended twice to the query expression because addAdHocFilters()/addLabelToQuery() unconditionally adds label matchers, so double interpolation duplicates matchers like env=\"prod\"",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The shard splitting test masks the double-interpolation bug because its applyTemplateVariables mock uses idempotent string replacement and asserts the double-call pattern (5 calls), so it doesn\u2019t catch duplicated ad-hoc filters",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The applyTemplateVariables test mock mutates the input query object in place instead of returning a new object like the real implementation, potentially hiding bugs that depend on request.targets remaining unmodified",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "runShardSplitQuery changed the order of filtering vs interpolation (filtering before interpolation), which can allow template-only expressions that interpolate to an empty string to slip through and be processed/sent as empty queries",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/grafana/grafana/pull/79265": {
     "gemini": [
@@ -12019,6 +12801,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "TagDevice was changed from asynchronous fire-and-forget to synchronous on the authentication request path, so slow/failing DB operations now block anonymous authentication and add latency to every anonymous request",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "TagDevice now inherits the HTTP request context (potentially short timeout) instead of using an independent background context with a 2-minute timeout, changing timeout/cancellation behavior for device tagging",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Device limit enforcement has a TOCTOU race: CreateOrUpdateDevice counts devices then updates/inserts, allowing concurrent requests to exceed the configured device limit",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "updateDevice returns ErrDeviceLimitReached when rowsAffected == 0, conflating 'device not found' with 'device exists but is outside the updated_at BETWEEN window (expired)', causing expired returning devices to be incorrectly rejected and producing a misleading error",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "FrontendSettingsDTO.AnonymousDeviceLimit always serializes as a number (0 when unset) due to missing omitempty, but the TypeScript type is number | undefined, so frontend checks for undefined to mean 'no limit configured' will never work",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9": {
@@ -12391,7 +13205,33 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "LRU translation cache key omits SiteSetting.default_locale even though it affects the fallback chain, so changing default_locale can serve stale cached translations resolved via the old fallback chain",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "No I18n.reload! (or equivalent cache invalidation) is triggered when SiteSetting.default_locale changes, leaving the translation cache populated with entries based on the previous default locale",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "translate only ensures config.locale is loaded; with I18n::Backend::Fallbacks enabled, fallback locales in the chain may not be loaded in non-controller contexts (Sidekiq/rake/console), causing fallbacks to silently fail",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ensure_loaded! initializes @loaded_locales ||= [] outside the LOAD_MUTEX, creating a race where concurrent threads can overwrite the array and lose loaded-locale state, leading to redundant loads or missed locale loading",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/grafana/grafana/pull/76186": {
     "gemini": [
@@ -12611,6 +13451,20 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "TestLogger.FromContext creates and returns a new disconnected TestLogger via NewTestLogger(), so LoggerMiddleware.logRequest logs to a throwaway instance and tests cannot capture/verify log output",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) dereference req.PluginContext without checking if req is nil, risking a nil pointer panic when used outside the Decorator or in tests",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10": {
@@ -13457,6 +14311,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Migration can crash on fresh installs when the 'embed_category' site setting is missing because the SQL query returns 0 rows and code accesses [0]['id'], causing NoMethodError and bypassing the category_id == 0 fallback",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Migration builds SQL INSERT statements by interpolating host values ('#{h}') from the old embeddable_hosts setting, allowing single quotes to break SQL and enabling SQL injection if the setting is corrupted",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Admin EmbeddableHostsController destroy/update actions can crash when EmbeddableHost.where(id: params[:id]).first returns nil; destroy calls host.destroy and update passes nil into save_host leading to NoMethodError",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "EmbeddableHost model before_validation callback calls self.host.sub! even when host is nil, causing NoMethodError before validations can reject missing host (e.g., when params[:embeddable_host][:host] is absent)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Fabricator definitions are in the wrong files: category_fabricator.rb contains Fabricator(:embeddable_host) while embeddable_host_fabricator.rb contains category fabricators, causing confusing mismatch between filenames and contents",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "JavaScript basePath uses type.replace('_','-') which only replaces the first underscore, so types with multiple underscores will be converted incorrectly and may fail to match ADMIN_MODELS",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7": {
@@ -13952,6 +14844,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In app/assets/stylesheets/desktop/topic-post.scss around line 291, the light-theme lightness for .embedded-posts link text was changed from 30% to 70%, making it washed out; the light/dark theme lightness values appear to be accidentally swapped",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/assets/stylesheets/mobile/modal.scss around line 102, dark-light-choose arguments for .custom-message-length are reversed compared to desktop (mobile uses 30% for light and 70% for dark), causing inverted character count contrast on light vs dark themes",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/assets/stylesheets/mobile/topic-post.scss around line 182, the light-theme lightness for h3 changed from 20% to 50%, making headings too grey on light themes; the dark-theme counterpart should be ~80% (complementary to 20%) rather than 50%",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/assets/stylesheets/desktop/user.scss around line 522 (and corresponding mobile user stylesheet), the light-theme lightness for .name changed from 30% to 50%, making names washed out and inconsistent with .username a in common/base/user.scss which keeps 30%",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8": {
@@ -14593,6 +15511,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Public GroupsController#members endpoint response format changed from a flat JSON array to an object with {members, meta}, breaking existing consumers expecting an array",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "GroupsController#members now always paginates with a default limit (50), so callers not providing pagination params may silently receive truncated member lists (previously non-automatic groups returned all members)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Off-by-one bug in totalPages calculation: using Math.floor(user_count / limit) + 1 produces an extra page when user_count is exactly divisible by limit; should use Math.ceil(user_count / limit)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Controller spec for removing members from automatic groups uses xhr :put even though the route is DELETE, so the test doesn't validate the real HTTP method/path and could miss routing/method bugs",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Admin groups update unconditionally sets group.visible based on params[:visible] == \"true\"; if visible is omitted in a partial update, visibility is unintentionally set to false",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "New group creation visibility can be incorrectly forced to false because the JS model asJSON() sends visible: !!this.get('visible') (false when unset) and the create action interprets params[:visible] == \"true\", so checkbox state/defaults may not be respected",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3": {
@@ -14993,6 +15949,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "BlockedEmail.should_block? performs DB writes (incrementing match_count and updating last_match_at) as a side effect during model validation, so repeated validation calls can inflate blocked-email statistics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BlockedEmail.should_block? calls record.save inside validation, triggering BlockedEmail validations (including a uniqueness query) and making validation unnecessarily expensive",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Client-side create account JS adds the email to rejectedEmails for any email validation error (including uniqueness/taken email), causing valid emails to be permanently marked rejected until page refresh",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "rejectedEmails is initialized as Em.A([]) on the Ember class prototype, so the array is shared across controller instances and can cause cross-instance state pollution",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "BlockedEmail.should_block? calls record.save without checking whether it succeeded, so failures can be silently ignored and the method may return based on dirty/stale in-memory state",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5": {
@@ -15340,7 +16328,45 @@
         "source": "extracted"
       }
     ],
-    "macroscope": []
+    "macroscope": [],
+    "tusk": [
+      {
+        "text": "The `order` mixin passes raw `$val` to `-webkit-box-ordinal-group`/`-moz-box-ordinal-group` even though those legacy properties are 1-based, so it should use `$val + 1` to match standard `order` behavior in old WebKit/Firefox",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `.small-action`, `.topic-avatar` still has `float: left` even though the parent is now `display: flex`, indicating an incomplete flex migration and causing confusing/obsolete styling",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In `.small-action-desc`, changing padding from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed left spacing for the avatar, potentially shrinking the gap on narrow viewports and causing text to appear too tight or overlap",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The mixin uses invalid `-ms-align-items`, which IE10/11 will ignore; only `-ms-flex-align` should be used for IE",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The `align-items` mixin does not translate `flex-start`/`flex-end` to IE10/11 `-ms-flex-align` values (`start`/`end`), so callers using `flex-start`/`flex-end` would break in IE",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`.badge-wrapper.bullet` has inconsistent `margin-top: 5px` between contexts (removed in one rule but retained under `#topic-title .title-wrapper`), leading to inconsistent bullet badge alignment/styling",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6": {
     "gemini": [
@@ -15728,6 +16754,26 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Serializer conditional include hook is misnamed: `include_website_name` lacks the required trailing `?`, so ActiveModel::Serializer never calls it and `website_name` is always serialized even when `website` is nil/blank",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`website_name` is missing from the `untrusted_attributes` list while `website` is included, allowing TL0 users' website info to leak via `website_name` when `scope.restrict_user_fields?` should hide it",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Test setup is fragile: `Discourse.stubs(:current_hostname)` is applied inside the `it` block after `let(:json)` is defined, so if `json` is evaluated before the stub (e.g., in a `before`/shared example) the stub won\u2019t take effect; the stub should be set before `json` is accessed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4": {
@@ -16621,6 +17667,44 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Invalid ERB/Ruby syntax in app/views/embed/best.html.erb line 6: `<%- end if %>` uses `end if` without a condition and should be `<%- end %>`, causing a SyntaxError when rendering the template",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "RSpec controller specs do not render views by default, so the ERB SyntaxError in best.html.erb is not caught by the existing controller spec coverage",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Potential XSS in app/views/layouts/embed.html.erb line 11: `request.referer` is interpolated into a single-quoted JavaScript string without JS-escaping/sanitizing path/query/fragment, allowing JS string breakout despite HTML escaping",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Nil handling bug in app/jobs/scheduled/poll_feed.rb line 35: `i.content` can be nil for RSS items lacking `content:encoded`, so calling `.scrub` can raise `NoMethodError`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Insecure postMessage origin validation in app/assets/javascripts/embed.js line 17: using `discourseUrl.indexOf(e.origin)` performs substring matching and can be bypassed by attacker-controlled domains that contain the expected origin as a substring/prefix",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Unintended string mutation in app/models/topic_embed.rb line 13: using `contents << ...` mutates the caller's string, causing the footer to be appended multiple times across repeated imports and leading to incorrect content SHA1 comparisons on re-import/update",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1": {
@@ -17322,6 +18406,50 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In app/models/optimized_image.rb, the method `downsize` is defined twice; the second definition overrides the first, leaving only the 3-arg/4-arg signature and removing the 5-arg variant",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The caller in app/jobs/regular/resize_emoji.rb:14 invokes `OptimizedImage.downsize` with 5 arguments, which will raise `ArgumentError (given 5, expected 3..4)` at runtime and break emoji resizing",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/assets/javascripts/discourse/lib/utilities.js, client-side upload size validation is hardcoded to 10MB for all upload types, ignoring the dynamic site settings (e.g., max_image_size_kb/max_attachment_size_kb)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Because client-side validation is hardcoded to 10MB, non-image attachments may be accepted client-side above the configured limits and then rejected by the server",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/assets/javascripts/discourse/lib/utilities.js, the 413 (entity too large) error handler hardcodes `maxSizeKB` to 10MB, causing the displayed error message to be incorrect when the real server limit differs",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/controllers/uploads_controller.rb, the API URL download limit is hardcoded to `10.megabytes` instead of using `SiteSetting.max_image_size_kb.kilobytes`, artificially limiting uploads when the admin setting is higher than 10MB",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In app/controllers/uploads_controller.rb, the hardcoded 10MB API URL download limit applies to non-image URL uploads as well because the type check occurs later",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2": {
@@ -17844,6 +18972,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Possible NoMethodError crash in topics_controller: TopicUser.find_by can return nil and code calls tu.notification_level without a nil check",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "I18n unsubscribe_link string now requires %{unsubscribe_url}; email paths using add_unsubscribe_link: true without providing unsubscribe_url can raise I18n::MissingInterpolationArgument or produce an empty/broken link",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS risk in dropdown-button component: title is concatenated into HTML without escaping, allowing user-controlled content to render as raw HTML",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "XSS risk in topic unsubscribe template/controller: triple-stache {{{stopNotificiationsText}}} renders raw HTML and interpolates model.fancyTitle into an i18n string, potentially allowing malicious HTML to be rendered unescaped",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Email html_part placeholder substitution bug: %{respond_instructions} and %{unsubscribe_link} are only replaced when add_unsubscribe_link is true, but the template always includes %{respond_instructions}, so literals can appear in emails when add_unsubscribe_link is false",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/22532": {
@@ -18402,65 +19562,103 @@
         "source": "extracted"
       },
       {
-        "text": "An empty wrapper div is still rendered when CredentialActionsDropdown returns null, leaving an invisible fixed-width placeholder in the UI layout",
+        "text": "An empty wrapper div is still rendered when CredentialActionsDropdown returns null, leaving an invisible fixed-width placeholder in the UI layout",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Date formatting uses a hard-coded \"en-US\" locale, ignoring the user's locale and rendering timestamps in an unexpected format for non-US users",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
+    ],
+    "mra-nano": [
+      {
+        "text": "SelectedCalendarRepository.updateManyByCredentialId is called with an empty Prisma update data object ({}), which can be a no-op or throw a runtime validation error",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "SelectedCalendarRepository.updateManyByCredentialId uses Prisma.SelectedCalendarUpdateInput, but updateMany likely requires Prisma.SelectedCalendarUpdateManyMutationInput, causing a type mismatch and potential functional bug",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "CalendarCacheRepository.getCacheStatusByCredentialIds can be called with an empty credentialIds array, producing a Prisma/SQL IN () clause that may crash or throw at runtime",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The connectedCalendarsHandler path does not guard against the edge case of zero connected credentials before calling getCacheStatusByCredentialIds",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "CalendarCacheRepositoryMock.getCacheStatusByCredentialIds has a signature with no parameters, which does not match the interface requiring credentialIds: number[], breaking TypeScript/interface conformance and potentially DI/tests",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "Date formatting uses a hard-coded \"en-US\" locale, ignoring the user's locale and rendering timestamps in an unexpected format for non-US users",
+        "text": "UI translation keys remove_app, yes_remove_app, and are_you_sure_you_want_to_remove_this_app are referenced but missing from en/common.json, risking missing/fallback text",
         "path": null,
         "line": null,
         "source": "extracted"
       }
     ],
-    "mra-nano": [
+    "macroscope": [
       {
-        "text": "SelectedCalendarRepository.updateManyByCredentialId is called with an empty Prisma update data object ({}), which can be a no-op or throw a runtime validation error",
+        "text": "fetchAvailabilityAndSetCache updates updatedAt for all calendars under a credential even though it only fetches availability for the selectedCalendars subset, causing unrelated (e.g., just-unwatched) calendars to have their timestamps incorrectly refreshed",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "SelectedCalendarRepository.updateManyByCredentialId uses Prisma.SelectedCalendarUpdateInput, but updateMany likely requires Prisma.SelectedCalendarUpdateManyMutationInput, causing a type mismatch and potential functional bug",
+        "text": "scripts/test-gcal-webhooks.sh uses macOS-specific `sed -i ''` syntax that fails on Linux because the empty string is treated as a filename, preventing $ENV_FILE from being updated",
         "path": null,
         "line": null,
         "source": "extracted"
-      },
+      }
+    ],
+    "tusk": [
       {
-        "text": "CalendarCacheRepository.getCacheStatusByCredentialIds can be called with an empty credentialIds array, producing a Prisma/SQL IN () clause that may crash or throw at runtime",
+        "text": "Calling `updateManyByCredentialId(credentialId, {})` with an empty update object is a no-op, so `SelectedCalendar.updatedAt` is never updated as intended (Prisma `updateMany` doesn\u2019t trigger `@updatedAt`).",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "The connectedCalendarsHandler path does not guard against the edge case of zero connected credentials before calling getCacheStatusByCredentialIds",
+        "text": "`deleteCache` handler authorizes credentials only by `credential.userId === user.id`, ignoring delegation via `delegationCredentialId`, which can prevent delegated users from deleting their own cache.",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "CalendarCacheRepositoryMock.getCacheStatusByCredentialIds has a signature with no parameters, which does not match the interface requiring credentialIds: number[], breaking TypeScript/interface conformance and potentially DI/tests",
+        "text": "`deleteCache` handler does not verify the credential is a calendar credential, so any user-owned credential ID could be used to attempt calendar cache deletion.",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "UI translation keys remove_app, yes_remove_app, and are_you_sure_you_want_to_remove_this_app are referenced but missing from en/common.json, risking missing/fallback text",
+        "text": "`CalendarCache.updatedAt` relies on Prisma `@updatedAt`, but raw SQL upserts/inserts (e.g., `$queryRaw` in `upsertCachedAvailability`) won\u2019t trigger it, so `updatedAt` may never update beyond the DB default insert timestamp.",
         "path": null,
         "line": null,
         "source": "extracted"
-      }
-    ],
-    "macroscope": [
+      },
       {
-        "text": "fetchAvailabilityAndSetCache updates updatedAt for all calendars under a credential even though it only fetches availability for the selectedCalendars subset, causing unrelated (e.g., just-unwatched) calendars to have their timestamps incorrectly refreshed",
+        "text": "Using `cacheStatusMap.get(id) || null` conflates `undefined` (missing key) with `null` (present-but-null) and should use `?? null` for correct semantics.",
         "path": null,
         "line": null,
         "source": "extracted"
       },
       {
-        "text": "scripts/test-gcal-webhooks.sh uses macOS-specific `sed -i ''` syntax that fails on Linux because the empty string is treated as a filename, preventing $ENV_FILE from being updated",
+        "text": "`CredentialActionsDropdown` is rendered unconditionally and initializes state and tRPC mutations even when it returns `null`, causing unnecessary hook/mutation setup overhead (including `trpc.viewer.credentials.delete.useMutation()` subscriptions) for credentials that won\u2019t show the dropdown.",
         "path": null,
         "line": null,
         "source": "extracted"
@@ -18971,6 +20169,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Uses object reference equality (===) to compare two different dayjs instances, so the override start==end (zero-length/unavailable-day) check never triggers and unavailable-day overrides are treated as available",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Copy-paste bug: `end` is computed from `slotStartTime` instead of `slotEndTime`, so slots whose end extends past working hours may not be rejected",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Timezone conversion bug: `dayjs(date.start)` is parsed in server-local time then adjusted by a negated organizer offset, causing incorrect override boundaries when server timezone differs from UTC and creating inconsistency with other offset logic",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Incorrect use of `.find()` with a mutated `dateOverrideExist` flag: short-circuiting on the first truthy return can reject a slot based on the first override and skip other overrides that might allow it",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/14943": {
@@ -19712,6 +20936,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Public methods getAuthorizationConditions() and getFilterConditions() return Prisma.Sql fragments containing OR expressions without wrapping the overall expression in outer parentheses, so callers composing them with AND can hit SQL operator-precedence bugs where one OR branch bypasses the AND filter (potential privilege escalation / incorrect filtering).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Constructor options type was changed from a discriminated union (requiring teamId when scope='team') to a flat type with optional teamId, removing compile-time enforcement and allowing callers to omit teamId for team scope, leading to runtime zod validation failure and silently returning NOTHING_CONDITION (no data).",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Filters use falsy guards (e.g., if (this.filters.eventTypeId) / if (this.filters.memberUserId)) which skip applying filters when the value is 0, causing unintended unfiltered results for valid numeric IDs equal to 0.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "getBaseConditions() contains dead/unreachable branches because authConditions is always a truthy Prisma.Sql object (including NOTHING_CONDITION), so logic that assumes authConditions can be falsy is incorrect and prevents short-circuiting when authorization fails.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Caching tests for getAuthorizationConditions() and getFilterConditions() were removed while caching logic remains, leaving caching behavior untested and allowing regressions that could cause repeated membership-check queries.",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/11059": {
@@ -20769,6 +22025,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "parseRefreshTokenResponse returns the full safeParse result ({success,data}) and Google Calendar stores it directly as the credential key, causing the DB to persist a wrapped object instead of the expected token fields and corrupting subsequent credential reads",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response instead of the previously expected parsed/app-specific token object, causing callers (Google Calendar/Hubspot/Zoho) to access missing properties and crash due to inconsistent return types across code paths",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "minimumTokenResponseSchema incorrectly uses [z.string().toString()] as a computed property key, creating a literal key (e.g., 'ZodString') and failing to validate presence of an expiry field like expires_in/expiry_date, making minimum token validation ineffective",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Salesforce getClient refreshes and saves a new access token but then instantiates jsforce.Connection with credentialKey.access_token (the old token), so the current connection uses a stale/expired token despite the DB being updated",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Webhook endpoint does not enforce POST-only requests; non-POST methods (e.g., GET without JSON body) can trigger zod parse errors/500s and violate the intended endpoint semantics",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/7232": {
@@ -21410,6 +22698,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "When `immediateDelete` is true and `referenceId` is non-null, the code cancels the SendGrid batch but returns without deleting or marking the `workflowReminder` DB record (e.g., not setting `cancelled=true`), leaving an orphaned reminder that won\u2019t be cleaned up by either the CRON cancel loop or the unscheduled query",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In the CRON job, `deleteMany` removes all reminders with past `scheduledDate` before processing `cancelled: true` reminders; if a reminder is cancelled shortly before send time and the CRON runs after `scheduledDate`, the DB record is deleted before SendGrid cancellation occurs, so the email is still delivered",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "The CRON cancellation loop is wrapped in a single try/catch; one SendGrid cancel failure aborts the entire loop, leaving remaining cancelled reminders unprocessed (not cancelled in SendGrid and not deleted), so emails may still be sent",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In the reschedule path, async deletions are invoked inside `forEach` without awaiting; errors won\u2019t be caught by the surrounding try/catch and can become unhandled promise rejections",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "After a refactor, the sender name/sender ID input section is rendered unconditionally instead of only when `isPhoneNumberNeeded || isSenderIdNeeded`, causing irrelevant sender fields to appear for email-type workflow steps",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/14740": {
@@ -22096,6 +23416,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Authorization logic for `isTeamAdminOrOwner` uses `&&` instead of `||`, requiring users to be both team admin and owner and incorrectly denying admins (making the admin check effectively dead code)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Handler passes raw `guests` input to `sendAddGuestsEmails` instead of filtered `uniqueGuests`, causing existing attendees to be treated as new guests and receive the wrong email template (booking confirmation instead of add-guests notification)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Blacklist email comparison lowercases `blacklistedGuestEmails` but compares incoming guest emails without lowercasing, allowing case-variant emails to bypass the blacklist",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "When `booking.userId` is null, `booking.userId || 0` leads to querying user id 0 with `findFirstOrThrow`, throwing an unhandled Prisma NotFoundError instead of a proper TRPC error (reachable if organizer account was deleted)",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "`OrganizerAddGuestsEmail.getNodeMailerPayload()` accesses `this.calEvent.attendees[0].name` without guarding for an empty attendees array, risking a runtime TypeError that can crash the whole email batch sent via `Promise.all`",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/10600": {
@@ -22686,6 +24038,32 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Password verification can be bypassed by supplying a non-empty dummy totpCode along with a valid backupCode, causing the password check to be skipped and the backup-code branch to succeed without verifying the password",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Backup code login is blocked for third-party identity provider (e.g., Google/SAML) users because a guard throws when totpCode is missing, preventing the backup-code path from being reached and making backup codes unusable for these users",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "API route throws an unhandled exception when CALENDSO_ENCRYPTION_KEY is missing, leading Next.js to return an HTML 500 page instead of structured JSON; client code expecting response.json() then fails/silently errors",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Blob/object URL created via URL.createObjectURL for backup code download is not revoked on dialog close/unmount (resetState lacks cleanup), causing a memory/resource leak per 2FA setup flow",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/10967": {
@@ -23350,6 +24728,38 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "In updateEvent/deleteEvent, when externalCalendarId is falsy the code does event.destinationCalendar?.find(cal => cal.externalId === externalCalendarId), which can never match and returns undefined; it should fall back to event.destinationCalendar?.externalId, otherwise updates/deletes use an undefined calendar ID",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In EventManager, destructuring const [mainHostDestinationCalendar] = evt.destinationCalendar ?? [] can yield undefined for null/empty destinationCalendar, but the next line accesses mainHostDestinationCalendar.integration without optional chaining, causing a TypeError",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In updateAllCalendarEvents, if reference.credentialId is set but the credential is missing from cache and the DB lookup fails, credential stays undefined and is still passed to updateEvent, which will crash when getCalendar(credential) is called",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In handleNewBooking, evt.destinationCalendar is set to null when neither eventType.destinationCalendar nor organizerUser.destinationCalendar exist, so evt.destinationCalendar?.push(...teamDestinationCalendars) becomes a no-op and team member destination calendars are silently dropped",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In organization creation, the slug/requestedSlug conditional logic is inverted: slug is now set only when billing is enabled (and requestedSlug also only when enabled), resulting in non-billing orgs getting neither slug nor requestedSlug and billing orgs getting both, opposite of intended behavior",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   },
   "https://github.com/calcom/cal.com/pull/8087": {
@@ -23742,6 +25152,80 @@
         "line": null,
         "source": "extracted"
       }
+    ],
+    "tusk": [
+      {
+        "text": "Using an async callback inside Array.forEach in packages/app-store/vital/lib/reschedule.ts causes fire-and-forget behavior because forEach doesn\u2019t await returned promises",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/app-store/vital/lib/reschedule.ts, errors thrown in async operations inside the forEach won\u2019t be caught by the surrounding try/catch",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/app-store/vital/lib/reschedule.ts, deleteEvent/deleteMeeting calls inside the async forEach are not awaited before sending reschedule emails, causing races or silent failures",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "Using an async callback inside Array.forEach in packages/app-store/wipemycalother/lib/reschedule.ts discards promises, leaving calendar deleteEvent calls unawaited",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/app-store/wipemycalother/lib/reschedule.ts, errors from async operations inside the forEach won\u2019t be caught and the function proceeds to send reschedule emails immediately",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/trpc/server/routers/viewer/bookings.tsx, the forEach on line 553 uses an async callback that forEach ignores, so deleteEvent/deleteMeeting operations are unawaited",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/trpc/server/routers/viewer/bookings.tsx, the code proceeds to send reschedule emails before deletions complete, and errors from deletions are silently swallowed",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/features/bookings/lib/handleCancelBooking.ts, an async callback inside forEach discards promises, so async work isn\u2019t awaited",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/features/bookings/lib/handleCancelBooking.ts, apiDeletes is populated inside the unawaited async forEach callback, so Promise.all(apiDeletes) runs with an empty array and skips calendar deletions for recurring events",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/core/CalendarManager.ts, getCalendarCredentials is synchronous but calls async getCalendar inside flatMap, resulting in returned objects where calendar is a Promise instead of Calendar|null",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/core/CalendarManager.ts, the function\u2019s type contract is broken because it claims calendar is Calendar|null but actually returns Promise<Calendar|null>",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      },
+      {
+        "text": "In packages/core/CalendarManager.ts, callers that don\u2019t await item.calendar may treat a truthy Promise as a valid calendar and bypass null checks",
+        "path": null,
+        "line": null,
+        "source": "extracted"
+      }
     ]
   }
 }
\ No newline at end of file
diff --git a/offline/results/openai_gpt-5.2/evaluations.json b/offline/results/openai_gpt-5.2/evaluations.json
index 7934278..161a029 100644
--- a/offline/results/openai_gpt-5.2/evaluations.json
+++ b/offline/results/openai_gpt-5.2/evaluations.json
@@ -1889,6 +1889,55 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37429__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37429__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The translation is in Italian instead of Lithuanian. This should be translated to Lithuanian to match the file's locale (messages_lt.properties).",
+          "severity": "Medium",
+          "matched_candidate": "Lithuanian (lt) translation string is incorrectly in Italian ('Installa una delle seguenti applicazioni sul tuo cellulare:') in account messages file, so Lithuanian users see Italian text during TOTP setup",
+          "confidence": 0.97,
+          "reasoning": "Both describe the same issue: a string in the Lithuanian locale file (messages_lt.properties / lt translation) is mistakenly written in Italian and should be translated to Lithuanian."
+        },
+        {
+          "golden_comment": "The totpStep1 value uses Traditional Chinese terms in the Simplified Chinese file (zh_CN), which is likely incorrect for this locale. Please verify the locale\u2011appropriate translation.",
+          "severity": "Medium",
+          "matched_candidate": "Chinese Simplified (zh_CN) account translation uses Traditional Chinese characters ('\u5728\u60a8\u7684\u624b\u6a5f\u4e0a\u5b89\u88dd\u4ee5\u4e0b\u61c9\u7528\u7a0b\u5f0f\u4e4b\u4e00\uff1a') instead of Simplified, creating an incorrect/inconsistent zh_CN locale translation",
+          "confidence": 0.93,
+          "reasoning": "Both point out that the zh_CN (Simplified Chinese) translation contains Traditional Chinese wording/characters, indicating an incorrect locale-appropriate translation that should be verified/corrected."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Same incorrect Italian string appears in the Lithuanian (lt) login message file as well, affecting the TOTP setup step there too"
+        },
+        {
+          "candidate": "VerifyMessageProperties.containsHtml() checks the English value before stripping anchor tags while translations have anchors sanitized, causing POLICY_SOME_HTML to be selected when English contains only <a> tags and unintentionally allowing other HTML tags (e.g., <br>, <p>, <strong>, <b>) in translations contrary to the intended policy"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The anchor sanitization logic has a potential issue where it consumes English matcher groups without proper validation. If the translated text has more anchor tags than the English text, this could lead to incorrect validation results.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The method name 'santizeAnchors' should be 'sanitizeAnchors' (missing 'i').",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 2,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37429__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37429__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/37634": {
@@ -3695,6 +3744,48 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37634__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37634__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Wrong parameter in null check (grantType vs. rawTokenId)",
+          "severity": "Critical",
+          "matched_candidate": "Null-check bug in AccessTokenContext: Objects.requireNonNull is called on grantType with message \"Null rawTokenId not allowed\" instead of checking rawTokenId, so rawTokenId can be null and cause NPEs later (e.g., in encodeTokenId concatenation).",
+          "confidence": 0.97,
+          "reasoning": "The candidate issue describes that Objects.requireNonNull checks grantType while the error message refers to rawTokenId, meaning rawTokenId is not actually null-checked and may cause later NPEs. This is the same underlying issue as the golden comment (wrong parameter in null check: grantType vs rawTokenId)."
+        },
+        {
+          "golden_comment": "In isAccessTokenId, the substring for the grant shortcut and the equality check look inverted: the grant shortcut occupies indices 4\u20135 (substring(4,6)), and a match should return true (combined with UUID check), not false.",
+          "severity": "High",
+          "matched_candidate": "isAccessTokenId matcher has inverted grant-type comparison logic: it returns false when the extracted grant shortcut equals the expected value, instead of returning false when it does not match (missing negation).",
+          "confidence": 0.86,
+          "reasoning": "Both comments identify the same core bug in isAccessTokenId: the grant shortcut comparison logic is inverted, causing the function to return false on a match instead of true (i.e., missing negation / reversed equality outcome). The golden also mentions substring indices, which the candidate doesn\u2019t, but the primary issue\u2014reversed equality check behavior\u2014is the same."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "Javadoc mentions \"usually like 3-letters shortcut\" but some implementations use 2-letter shortcuts (\"ac\", \"cc\", \"rt\", \"te\", \"pc\", \"ci\", \"ro\"). Consider updating documentation to reflect actual usage pattern.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": " Catching generic RuntimeException is too broad. The implementation throws IllegalArgumentException specifically - catch that instead for more precise testing.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 0,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37634__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37634__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/38446": {
@@ -5169,6 +5260,50 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR38446__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR38446__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "After creating the RecoveryAuthnCodesCredentialModel, consider setting its id from the stored credential (e.g., myUser.recoveryCodes.getId()); otherwise getId() will be null and downstream removal by id (e.g., removeStoredCredentialById in the authenticator flow) may not work.",
+          "severity": "Low",
+          "matched_candidate": "getCredentials() creates a RecoveryAuthnCodesCredentialModel with a null ID because createFromValues() doesn't set the stored credential ID, so downstream removeStoredCredentialById(recoveryCodeCredentialModel.getId()) is called with null when codes are exhausted",
+          "confidence": 0.96,
+          "reasoning": "Both describe that RecoveryAuthnCodesCredentialModel is created without setting the stored credential ID, leaving getId() null and causing downstream removal by id (removeStoredCredentialById) to fail or be called with null."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "BackwardsCompatibilityUserStorage.isValid() for recovery codes accepts any matching code (anyMatch) instead of only the next sequential code, diverging from production behavior and allowing out-of-order validation"
+        },
+        {
+          "candidate": "BackwardsCompatibilityUserStorage.isValid() does not consume/remove a recovery code after successful validation, allowing the same code to be reused indefinitely and preventing tests from catching one-time-use behavior regressions"
+        },
+        {
+          "candidate": "RecoveryAuthnCodesFormAuthenticator removes exhausted recovery-code credentials via removeStoredCredentialById() which targets local storage; for federated users the credential resides in federated storage so removal silently fails, leaving a stale exhausted credential that can cause a loop where allCodesUsed() remains true on subsequent logins"
+        },
+        {
+          "candidate": "BackwardsCompatibilityUserStorage.updateCredential() stores raw recovery codes JSON in credentialData (instead of standard metadata in credentialData and hashed codes in secretData), and isValid()/getCredentials() depend on this nonstandard format, making the implementation fragile and likely to break if storage is changed to the standard model format"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Unsafe raw List deserialization without type safety. Calling Optional.get() directly on the Optional returned by RecoveryAuthnCodesUtils.getCredential(user) without checking isPresent() can lead to a NoSuchElementException if the Optional is empty.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR38446__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR38446__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/36882": {
@@ -6317,6 +6452,39 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR36882__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR36882__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Operator advanced-configuration documentation incorrectly claims the Operator will fail unless the `rolling-updates` feature is enabled, but the Operator code never checks `Profile.Feature.ROLLING_UPDATES` or runs `kc.sh update-compatibility` and instead decides rolling vs recreate purely via StatefulSet image/args/env comparisons"
+        },
+        {
+          "candidate": "No test verifies the `FEATURE_DISABLED` exit code (4) for the feature-disabled scenario because `testFeatureNotEnabled` uses `@Launch` where `picocli.exit()` does not call `System.exit`, so only the error message is asserted and exit code 4 could change without failing tests"
+        },
+        {
+          "candidate": "Feature-disabled behavior is only tested for the `metadata` subcommand; the `check` subcommand has its own independent feature check path that is never covered by tests"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Incorrect method call for exit codes. The picocli.exit() method calls System.exit() directly, which is problematic:",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR36882__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36882__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/36880": {
@@ -7954,6 +8122,54 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR36880__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR36880__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In getClientsWithPermission(String scope), iterating resourceStore.findByType(server, AdminPermissionsSchema.CLIENTS_RESOURCE_TYPE) and returning resource.getName() will only ever consider the type-level 'Clients' resource (per-client resources have no type) and return its name, while AvailableRoleMappingResource#getRoleIdsWithPermissions expects actual client IDs to pass to realm.getClientById, which can lead to incorrect behavior or a null client and subsequent failures.",
+          "severity": "High",
+          "matched_candidate": "getClientsWithPermission may include the all-clients type resource name \"Clients\" in the returned set; downstream AvailableRoleMappingResource.getRoleIdsWithPermissions calls realm.getClientById(\"Clients\") which returns null, leading to an NPE when calling .getRolesStream()",
+          "confidence": 0.93,
+          "reasoning": "The candidate flags that getClientsWithPermission returns the type-level resource name \"Clients\" and that downstream code treats returned values as client IDs (realm.getClientById(\"Clients\") -> null) causing failures (e.g., NPE). This matches the golden issue about iterating type-level resources and returning resource names instead of actual client IDs, leading to null clients and incorrect behavior."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Potential NullPointerException: getResourceTypeResource() may return null but its result is passed to PolicyStore.findByResource(server, resource) without a null check, violating the non-null contract and causing an NPE"
+        },
+        {
+          "candidate": "ClientPermissionsV2.canMapRoles/canMapCompositeRoles/canMapClientScopeRoles do not check the MANAGE_CLIENTS admin role, preventing users with that admin role from mapping roles unless they also have fine-grained permissions"
+        },
+        {
+          "candidate": "Incorrect Javadoc: requireView() documentation claims it returns boolean, but the method is void and should be documented as throwing ForbiddenException when canView() is false"
+        },
+        {
+          "candidate": "Incorrect Javadoc: requireViewClientScopes() documentation claims it returns boolean, but the method is void and should be documented as throwing ForbiddenException when the check fails"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Inconsistent feature flag bug causing orphaned permissions. The AdminPermissions event listener, responsible for cleaning up permissions upon role, client, or group removal, is incorrectly guarded by the ADMIN_FINE_GRAINED_AUTHZ (V1) feature flag. This is inconsistent with other methods in the class that use ADMIN_FINE_GRAINED_AUTHZ_V2. Consequently, if ADMIN_FINE_GRAINED_AUTHZ_V2 is enabled but V1 is not, the permission cleanup logic will not execute, leading to orphaned permission data. Cleanup should occur regardless of which fine-grained authorization version is enabled.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "In hasPermission(ClientModel client, String scope), the resource lookup uses findByName(server, client.getId(), server.getId()), but AdminPermissionsSchema.getOrCreateResource creates per-client resources with the owner set to resourceServer.getClientId(), so this lookup will never find those resources and will always fall back to the 'all-clients' resource, effectively ignoring client-specific permissions.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 4,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR36880__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR36880__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/37038": {
@@ -9434,6 +9650,46 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR37038__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR37038__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Incorrect permission check in canManage() method",
+          "severity": "High",
+          "matched_candidate": "GroupPermissionsV2.canManage() (no-arg) incorrectly checks both VIEW and MANAGE scopes, allowing users with only VIEW permission to pass manage checks (e.g., gating group creation via requireManage())",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same bug: the canManage() permission check is incorrect, involving an improper scope check that can let users with only VIEW pass manage-gated actions."
+        },
+        {
+          "golden_comment": "In getGroupIdsWithViewPermission, hasPermission is called with groupResource.getId() and the same groupResource.getId() is added to granted, but hasPermission resolves resources by name (treating the argument as a group id) and the GroupPermissionEvaluator contract says this method returns group IDs that are later used as UserModel.GROUPS and in getUsersCount group filters. This mismatch means per-group VIEW_MEMBERS/MANAGE_MEMBERS permissions may not yield the expected group IDs for filtering and counts, and evaluation may effectively only look at the type-level 'all-groups' resource; consider revisiting whether this should operate on the underlying group ids (resource names) instead so it aligns with the JPA queries and the interface contract.",
+          "severity": "High",
+          "matched_candidate": "getGroupIdsWithViewPermission() uses groupResource.getId() (authorization resource internal UUID) instead of groupResource.getName() (actual group model ID), breaking per-group permission evaluation because resourceStore.findByName() won\u2019t find resources by the internal UUID",
+          "confidence": 0.9,
+          "reasoning": "Both describe the same core mismatch: getGroupIdsWithViewPermission passes/returns groupResource.getId() (authorization resource UUID) where the permission evaluator and downstream filtering expect the underlying group identifier used as the resource name. This causes per-group permission checks to fail (resource lookup by name won\u2019t match) and yields incorrect group IDs for later queries/counts."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "GroupPermissionEvaluator.canManage() docstring claims VIEW or MANAGE is sufficient, which becomes misleading if canManage() is corrected to require MANAGE (or indicates an intentional but insecure design if left as-is)"
+        },
+        {
+          "candidate": "Tests call ApiUtil.handleCreatedResponse(response) inside an outer try-with-resources even though handleCreatedResponse also closes the response, causing double-close/double-assert and potentially unsafe access to headers/streams after closure (pattern repeated at multiple line ranges)"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR37038__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR37038__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/33832": {
@@ -10918,6 +11174,47 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR33832__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR33832__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Dead code exists where ASN1Encoder instances are created and written to, but their results are immediately discarded. The actual encoding is performed by new ASN1Encoder instances created in the subsequent return statement, rendering the earlier operations useless.",
+          "severity": "Low",
+          "matched_candidate": "`AuthzClientCryptoProvider.concatenatedRSToASN1DER` contains dead code that creates `ASN1Encoder` instances and writes integers but discards the results, causing unnecessary allocations each call",
+          "confidence": 0.93,
+          "reasoning": "The candidate flags dead code in `concatenatedRSToASN1DER` where ASN1Encoder instances are created, written to, and their outputs discarded, leading to unnecessary allocations. This matches the golden issue that earlier ASN1Encoder work is useless because the actual encoding is done by new instances in the return statement."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Adding `order()` to the public `CryptoProvider` SPI interface without a default implementation breaks binary/source compatibility for third-party `ServiceLoader` implementations (they will fail to compile/load)"
+        },
+        {
+          "candidate": "`ECDSAAlgorithmTest` reuses a single P-256 EC key pair for ES384 and ES512 tests, so signatures have ~32-byte r/s values instead of the required 48/66 bytes, meaning ES384/ES512-specific conversion bugs may not be caught"
+        },
+        {
+          "candidate": "`CryptoIntegration.detectProvider` builds the \"Ignored crypto providers\" log message with an unconditional trailing \", \" after the last provider name, producing malformed log output"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Returns wrong provider (default keystore instead of BouncyCastle)",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR33832__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR33832__20260411/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/40940": {
@@ -12287,6 +12584,47 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR40940__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR40940__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The reader thread isn\u2019t waited for; flipping deletedAll to true and asserting immediately can race and miss exceptions added just after the flag change, making this test flaky.",
+          "severity": "Medium",
+          "matched_candidate": "Test in GroupTest asserts caughtExceptions is empty without joining the reader thread first, so exceptions from an in-flight groups() call may be recorded after the assertion and the test can miss failures",
+          "confidence": 0.93,
+          "reasoning": "Both comments describe the same race/flakiness: the test asserts no exceptions without waiting/joining the reader thread, so exceptions can be added after the flag/condition change and after the assertion, causing missed failures."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "getSubGroupsStream(String, Integer, Integer) calls modelSupplier.get().getSubGroupsStream(...) without checking if modelSupplier.get() is null, risking an NPE under concurrent-deletion scenarios"
+        },
+        {
+          "candidate": "The getSubGroupsStream overload at line 262 calls modelSupplier.get() without a null check, risking an NPE under concurrent-deletion scenarios"
+        },
+        {
+          "candidate": "The getSubGroupsStream overload at line 268 calls modelSupplier.get() without a null check, risking an NPE under concurrent-deletion scenarios"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Returning null from getSubGroupsCount() violates the GroupModel contract (Javadoc says it never returns null) and may lead to NPEs in callers that expect a non-null count.",
+          "severity": "Critical"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR40940__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR40940__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/keycloak-greptile/pull/1": {
@@ -13702,6 +14040,46 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak-greptile__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak-greptile__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "ConditionalPasskeysEnabled() called without UserModel parameter",
+          "severity": "Medium",
+          "matched_candidate": "UsernameForm.java calls isConditionalPasskeysEnabled() with no arguments, but the parent method signature changed to isConditionalPasskeysEnabled(UserModel user), causing a compilation error (should pass context.getUser())",
+          "confidence": 0.94,
+          "reasoning": "Both describe the same underlying issue: the method ConditionalPasskeysEnabled/isConditionalPasskeysEnabled is being invoked without the required UserModel parameter after the signature changed, leading to a compilation error and needing to pass the user (e.g., context.getUser())."
+        },
+        {
+          "golden_comment": "With isConditionalPasskeysEnabled(UserModel user) requiring user != null, authenticate(...) will not call webauthnAuth.fillContextForm(context) on the initial login page where context.getUser() is still null, so conditional passkey UI will not be set up for first-time passkey login. Consider whether this should also be enabled when no user has been selected yet so ID-less passkey authentication on the initial login form continues to work.",
+          "severity": "Medium",
+          "matched_candidate": "Regression in UsernamePasswordForm.authenticate(): fillContextForm is no longer called when context.getUser() is null (initial login), so WebAuthn challenge/rpId/conditional UI attributes are not set on the initial login form, breaking passkey conditional UI setup",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same regression: on the initial login page context.getUser() is null, so authenticate() skips calling webauthnAuth.fillContextForm(context), preventing WebAuthn/conditional passkey UI attributes from being set up for first-time/ID-less passkey login."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Regression in UsernamePasswordForm.challenge(error path): fillContextForm is no longer called when context.getUser() is null, so after an invalid username/password on initial login the error page does not re-render passkey UI elements, preventing switching to passkey login"
+        },
+        {
+          "candidate": "Potential future NullPointerException risk in WebAuthnAuthenticator.fillContextForm(): user is passed to WebAuthnAuthenticatorsBean when shouldShowWebAuthnAuthenticators(context) is true; since shouldShowWebAuthnAuthenticators is protected/overridable, a subclass could return true without ensuring a non-null user, leading to null user usage"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak-greptile__tusk__PR1__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak-greptile__tusk__PR1__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/93824": {
@@ -15666,6 +16044,56 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR93824__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR93824__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Inconsistent metric tagging with 'shard' and 'shards'",
+          "severity": "Medium",
+          "matched_candidate": "Metrics tag key is inconsistent: uses \"shards\" instead of \"shard\", causing the metric to be missing from dashboards/alerts filtering on the \"shard\" tag",
+          "confidence": 0.96,
+          "reasoning": "The candidate issue explicitly notes inconsistent metric tag key usage between \"shards\" and \"shard\", which is exactly the inconsistency described in the golden comment, including the impact on dashboards/alerts."
+        },
+        {
+          "golden_comment": "Breaking out of the loop when the deadline has elapsed can skip terminating remaining flusher processes, potentially leaving them running after shutdown; consider ensuring termination is attempted even if the deadline is exceeded.",
+          "severity": "Medium",
+          "matched_candidate": "When join() exceeds the deadline, a break exits the loop and skips calling .terminate() on remaining processes, leaving potentially hung processes alive (zombies)",
+          "confidence": 0.93,
+          "reasoning": "Both describe that breaking out of the loop after the deadline is exceeded prevents terminating remaining flusher processes, potentially leaving them running/hung after shutdown."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Each sub-buffer calls get_memory_info() which queries Redis INFO on all cluster nodes; doing this once per sub-buffer causes redundant roundtrips and adds unnecessary latency on the submit() hot path"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Fixed sleep in tests can be flaky; wait on condition instead",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Because flusher processes are created via multiprocessing.get_context('spawn').Process, they are instances of multiprocessing.context.SpawnProcess, which on POSIX is not a subclass of multiprocessing.Process, so this isinstance check will always be false and hung processes won't be killed here.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "Sleep in test_consumer.py won\u2019t actually wait because time.sleep was monkeypatched above; consider restoring sleep or using a different sync to ensure the flusher has time to process.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 5,
+      "tp": 2,
+      "fp": 1,
+      "fn": 3,
+      "errors_count": 0,
+      "precision": 0.6666666666666666,
+      "recall": 0.4,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR93824__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR93824__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/5": {
@@ -17537,6 +17965,51 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR5__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR5__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Using zip(error_ids, events.values()) assumes the get_multi result preserves the input order; dict value order is not guaranteed to match error_ids, so event data can be paired with the wrong ID (missing nodes also shift alignment).",
+          "severity": "Low",
+          "matched_candidate": "fetch_error_details incorrectly zips error_ids with events.values() even though nodestore.backend.get_multi() returns a dict keyed by node_ids, causing error IDs to be paired with the wrong event data when dict insertion order differs",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same bug: zipping error_ids with events.values() assumes the get_multi dict\u2019s value iteration order aligns with the input IDs, which can mispair IDs to events (especially if order differs or items are missing)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "nodestore/base.py items.update(cache_items) can append cached entries at the end of the dict, breaking insertion-order alignment with the original requested IDs and contributing to mis-association when relying on dict.values() ordering"
+        },
+        {
+          "candidate": "gen_request_data compares error event timestamps in Unix epoch seconds to rrweb replay event timestamps in Unix epoch milliseconds, causing incorrect chronological ordering (errors always appear far earlier and are yielded before replay breadcrumbs)"
+        },
+        {
+          "candidate": "Unit test masks the timestamp unit mismatch by using the same small float timestamp values for both error events and replay events, so it passes despite the bug"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Breaking changes in error response format",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Detector validator uses wrong key when updating type",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 3,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR5__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR5__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/1": {
@@ -19464,6 +19937,55 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Django querysets do not support negative slicing",
+          "severity": "High",
+          "matched_candidate": "When enable_advanced_features is True and cursor.offset is negative, BasePaginator/OptimizedCursorPaginator can slice a Django queryset with a negative start (queryset[start_offset:stop]), which raises AssertionError because Django querysets do not support negative indexing; the comment claiming Django handles negative slicing is incorrect.",
+          "confidence": 0.93,
+          "reasoning": "Yes. The candidate explicitly describes slicing a Django queryset with a negative start offset and notes it raises an AssertionError because Django querysets do not support negative indexing/slicing, which is the same underlying issue as the golden comment."
+        },
+        {
+          "golden_comment": "When requests are authenticated with API keys or org auth tokens (which have user_id=None), organization_context.member is None. Line 71 attempts to access organization_context.member.has_global_access without checking if member is None, causing an AttributeError crash when optimized_pagination=true is used, even though the request passed all permission checks with valid org:write scope.",
+          "severity": "High",
+          "matched_candidate": "In organization_auditlogs.py, organization_context.member can be None for users without org membership; accessing organization_context.member.has_global_access will raise AttributeError, and this check runs on every request due to being evaluated before the use_optimized guard.",
+          "confidence": 0.9,
+          "reasoning": "Both describe the same core bug: organization_context.member may be None (e.g., API key/org token or non-member context) and the code accesses organization_context.member.has_global_access without a None check, leading to an AttributeError. The candidate adds extra detail about evaluation timing, but the underlying issue matches the golden comment."
+        },
+        {
+          "golden_comment": "get_item_key assumes a numeric key, but the paginator is used with order_by=-datetime in the audit logs endpoint; calling math.floor/ceil on a datetime will raise a TypeError.",
+          "severity": "High",
+          "matched_candidate": "OptimizedCursorPaginator.get_item_key calls math.floor/math.ceil on a datetime value (from getattr(item, self.key) when order_by='-datetime'), causing a TypeError; it should convert datetime to a numeric timestamp first (as DateTimePaginator does).",
+          "confidence": 0.97,
+          "reasoning": "Both describe the same bug: get_item_key assumes a numeric key and uses math.floor/ceil, but when the paginator is used with order_by='-datetime' it receives a datetime value, leading to a TypeError. The candidate also suggests converting datetime to a numeric timestamp, consistent with the expected issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "OptimizedCursorPaginator.value_from_cursor returns the raw numeric cursor.value without converting it back to a datetime; when ordering/filtering by a DateTimeField (e.g., datetime__lte=<int>), this can produce incorrect results or a database type error (unlike DateTimePaginator which converts back to a timezone-aware datetime)."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Importing non-existent OptimizedCursorPaginator",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 4,
+      "tp": 3,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.75,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR1__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR1__20260410/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/97529": {
@@ -20910,6 +21432,44 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR97529__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR97529__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "A race condition in BuildIndex allows multiple goroutines to concurrently build the same expensive index for the same key. This is caused by moving the b.cacheMu lock from protecting the entire function to only protecting the final cache assignment. ",
+          "severity": "High",
+          "matched_candidate": "Race condition due to reduced cacheMu lock scope in BuildIndex/getOrCreateIndex allowing concurrent callers to build the same index key simultaneously",
+          "confidence": 0.94,
+          "reasoning": "The candidate flags a race condition caused by narrowing the cacheMu lock scope in BuildIndex/getOrCreateIndex, leading to concurrent callers building the same index for the same key. This matches the golden issue about moving the lock to only protect final cache assignment and allowing duplicate expensive builds."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Data race/correctness bug: shared outer err variable is captured by multiple errgroup goroutines and concurrently written/read, so a goroutine may return another goroutine\u2019s error (or nil)"
+        },
+        {
+          "candidate": "Data race: totalBatchesIndexed is incremented from multiple errgroup goroutines without synchronization when initWorkers > 1"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Calling s.search.TotalDocs() here may race with concurrent index creation: TotalDocs iterates b.cache without synchronization, and the event watcher goroutine started just above could trigger BuildIndex writes concurrently, potentially causing a concurrent map read/write panic.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR97529__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR97529__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/80168": {
@@ -22257,6 +22817,53 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR80168__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR80168__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "MetricAlertDetectorHandler inherits from StatefulDetectorHandler but only contains pass, failing to implement its required abstract methods: counter_names (property), get_dedupe_value(), get_group_key_values(), and build_occurrence_and_event_data(). This will cause a TypeError at runtime when the class is instantiated.",
+          "severity": "High",
+          "matched_candidate": "MetricAlertDetectorHandler now subclasses StatefulDetectorHandler but does not implement required abstract methods (counter_names, get_dedupe_value, get_group_key_values, build_occurrence_and_event_data), so instantiating it will raise a TypeError when Detector.detector_handler constructs the handler",
+          "confidence": 0.98,
+          "reasoning": "The candidate issue describes the same problem: MetricAlertDetectorHandler subclasses StatefulDetectorHandler without implementing the required abstract methods (including counter_names and the listed methods), leading to a TypeError upon instantiation."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "TestEvaluateGroupKeyValue.test_dedupe builds the expected occurrence using group_key=\"val1\" but calls evaluate_group_key_value with group_key=\"group_key\", so the expected fingerprint/group_key does not match the actual code path"
+        },
+        {
+          "candidate": "TestEvaluateGroupKeyValue.test_dedupe is ineffective because IssueOccurrence.__eq__ compares only the hardcoded id, allowing the test to pass even when fingerprint/group_key are wrong"
+        },
+        {
+          "candidate": "test_state_results_multi_group builds the expected occurrence for group_2 with value=6 even though the input packet has group_2=10 and the real code forwards value=10"
+        },
+        {
+          "candidate": "test_state_results_multi_group can pass incorrectly because the mock occurrence builder does not vary output based on value and IssueOccurrence.__eq__ compares only by id"
+        },
+        {
+          "candidate": "test_state_results constructs expected occurrences using detector.detector_handler, but detector.detector_handler returns a new handler instance on each access, so the expected handler instance differs from the one used in process_detectors, making the test fragile if handler instance state matters"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Docstring says this returns a list of DetectorEvaluationResult, but the method now returns a dict keyed by DetectorGroupKey. Consider updating the docstring to match the new return type.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR80168__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80168__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/80528": {
@@ -23559,6 +24166,37 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR80528__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR80528__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The function modifies the config variable to include display values but then returns the original monitor.config instead of the modified version.",
+          "severity": "High",
+          "matched_candidate": "get_monitor_environment_context returns monitor_environment.monitor.config (original config) instead of the locally modified config copy, so the schedule_type human-readable override is discarded and the context contains the raw schedule_type value",
+          "confidence": 0.94,
+          "reasoning": "Both describe the same bug: the function creates/modifies a local config (adding display/human-readable values like schedule_type) but mistakenly returns the original monitor.config, so the modifications are lost."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "The code fetches MonitorCheckIn objects by ID when the required data already exists in previous_checkins. This creates an unnecessary database query.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 1,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 0,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 1.0,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR80528__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR80528__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/77754": {
@@ -25195,6 +25833,58 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR77754__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR77754__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Shared mutable default in dataclass timestamp",
+          "severity": "Medium",
+          "matched_candidate": "Dataclass field `queued` uses `timezone.now()` as a default value, causing it to be evaluated at import time so instances share the same timestamp; should use `field(default_factory=timezone.now)`",
+          "confidence": 0.95,
+          "reasoning": "The golden issue is about a shared mutable/default timestamp in a dataclass due to evaluating the default at definition/import time. The candidate explicitly flags using `timezone.now()` as a default (evaluated at import time) leading to shared timestamps across instances and recommends `default_factory`, which is the same underlying problem."
+        },
+        {
+          "golden_comment": "The method name has a typo: test_from_dict_inalid_data should be test_from_dict_invalid_data.",
+          "severity": "Low",
+          "matched_candidate": "Test name `test_from_dict_inalid_data` contains a typo ('inalid' instead of 'invalid'), making it harder to search and understand",
+          "confidence": 0.98,
+          "reasoning": "Both the golden comment and the candidate issue point out the same typo in the test method name `test_from_dict_inalid_data`, indicating it should be `test_from_dict_invalid_data`."
+        },
+        {
+          "golden_comment": "Method name says 'empty_array' but tests empty dict - consider renaming to 'test_from_dict_empty_dict' for clarity.",
+          "severity": "Low",
+          "matched_candidate": "Test name `test_from_dict_empty_array` is misleading because it tests an empty dict `{}` rather than an empty array",
+          "confidence": 0.93,
+          "reasoning": "Both point out that the test name mentions an empty array but the test actually uses an empty dict `{}`, making the name misleading and suggesting it should be renamed for clarity."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "`assignment_source` parameter was added to the abstract `sync_status_outbound` method even though status sync is unrelated to this PR\u2019s assignee sync changes"
+        },
+        {
+          "candidate": "`ExampleIntegration.sync_status_outbound(self, external_issue, is_resolved, project_id)` does not accept `**kwargs`, so callers passing the new `assignment_source` argument will raise a TypeError"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "to_dict() returns a datetime for queued; if this dict is passed in task kwargs (e.g., via apply_async), JSON serialization may fail depending on the serializer, which can cause enqueue errors.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 4,
+      "tp": 3,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.75,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR77754__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR77754__20260410/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/95633": {
@@ -26803,6 +27493,63 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR95633__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR95633__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The test test_thread_queue_parallel_error_handling has a docstring that doesn't match the test implementation.",
+          "severity": "Low",
+          "matched_candidate": "test_thread_queue_parallel_error_handling has a docstring claiming errors shouldn't block commits for other messages, but the assertion checks that no commits happened (opposite of the stated intent)",
+          "confidence": 0.94,
+          "reasoning": "The golden issue is that the docstring for test_thread_queue_parallel_error_handling does not match what the test actually does. The candidate explicitly describes this mismatch, noting the docstring intent vs the opposite assertion behavior. This is the same underlying problem."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "close() stops/joins the commit thread before draining/shutting down the worker pool, so messages processed during worker drain have side effects but their offsets are never committed, causing duplicate processing after restart"
+        },
+        {
+          "candidate": "test_thread_queue_parallel_error_handling is timing-dependent/flaky because the commit loop runs every ~1s while the test only waits ~0.2s, so it can pass simply because commits haven't occurred yet"
+        },
+        {
+          "candidate": "OrderedQueueWorker.run() marks offsets complete in a finally block even when processing raises, so failed messages are treated as successfully processed for committing purposes"
+        },
+        {
+          "candidate": "SimpleQueueProcessingStrategy docstring incorrectly claims natural backpressure, but worker queues are created unbounded (queue.Queue() without maxsize), allowing unbounded memory growth under load"
+        },
+        {
+          "candidate": "SimpleQueueProcessingStrategy docstring incorrectly claims offsets are only committed after successful processing, but offsets are completed/committed even on exceptions due to the finally block behavior"
+        },
+        {
+          "candidate": "OffsetTracker._get_partition_lock accesses/creates entries in partition_locks without synchronization, relying on CPython GIL behavior rather than guaranteed thread-safety"
+        },
+        {
+          "candidate": "OffsetTracker.get_committable_offsets iterates all_offsets keys without holding a lock while other threads may mutate all_offsets (including via defaultdict auto-creation), relying on non-guaranteed CPython atomicity"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The queue.shutdown() method with 'immediate=False' parameter may not exist in the standard Python queue module. This could cause AttributeError at runtime. Verify the correct API or implement a custom shutdown mechanism.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The magic number 50 for max_wait is used repeatedly throughout the tests. Consider extracting this as a named constant to improve maintainability.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 8,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 7,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.125,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR95633__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR95633__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/2": {
@@ -28669,6 +29416,53 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR2__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR2__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "OptimizedCursorPaginator negative-offset branch slices QuerySet with a negative start index",
+          "severity": "Critical",
+          "matched_candidate": "Optimized pagination path slices a Django queryset with a negative start index (queryset[negative_offset:stop]); Django ORM does not support negative indexing and raises AssertionError, so a crafted cursor with negative offset can trigger a 500 error.",
+          "confidence": 0.96,
+          "reasoning": "Both describe the same bug: in OptimizedCursorPaginator's negative-offset branch, the code slices a Django QuerySet with a negative start index, which Django disallows and can raise an error."
+        },
+        {
+          "golden_comment": "BasePaginator negative-offset branch slices QuerySet with a negative start index",
+          "severity": "High",
+          "matched_candidate": "Optimized pagination path slices a Django queryset with a negative start index (queryset[negative_offset:stop]); Django ORM does not support negative indexing and raises AssertionError, so a crafted cursor with negative offset can trigger a 500 error.",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same bug: in BasePaginator's negative-offset branch, the code slices a Django QuerySet using a negative start index, which Django ORM disallows and can raise an error."
+        },
+        {
+          "golden_comment": "OptimizedCursorPaginator.get_item_key uses floor/ceil on a datetime key (order_by='-datetime'), causing TypeError.",
+          "severity": "High",
+          "matched_candidate": "OptimizedCursorPaginator.get_item_key applies math.floor()/math.ceil() directly to a datetime value (from order_by=\"-datetime\"), which will raise a TypeError; it should convert the datetime to a numeric value first (as DateTimePaginator does).",
+          "confidence": 0.98,
+          "reasoning": "The candidate explicitly states that get_item_key uses math.floor/ceil on a datetime when ordering by \"-datetime\", leading to a TypeError, which is exactly the golden issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "A flush-time safety guard that skipped segments exceeding max_segment_spans (with metrics/logging) was removed; this eliminates protection against edge cases where segments can exceed the cap despite the Lua insert-side limit."
+        },
+        {
+          "candidate": "max_segment_spans is still initialized on the class but is now unused/dead code after the flush-side guard removal."
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.75,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR2__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR2__20260410/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/sentry-greptile/pull/3": {
@@ -30419,6 +31213,51 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry-greptile__macroscope__PR3__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry-greptile__macroscope__PR3__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Using Python\u2019s built-in hash() to build cache keys is non-deterministic across processes (hash randomization), so keys won\u2019t match across workers and invalidate_upsampling_cache may fail to delete them. Use a deterministic serialization of project_ids for the cache key.",
+          "severity": "Low",
+          "matched_candidate": "Cache key construction uses Python built-in hash(), which is randomized per process (PYTHONHASHSEED), causing cross-process cache key mismatches and cache misses in shared Redis/Memcached",
+          "confidence": 0.93,
+          "reasoning": "The candidate flags that using Python\u2019s built-in hash() for cache keys is randomized per process (PYTHONHASHSEED), leading to cross-process key mismatches and cache misses in shared caches. This is the same core issue as the golden comment about non-deterministic keys across workers causing invalidation to fail; it implies the need for deterministic key construction."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Cache key for upsampling decision does not include dataset/request context, so a cached True/False can be reused across different query contexts and yield inconsistent behavior"
+        },
+        {
+          "candidate": "Logic change from any(project_id in allowlist) to all(project_id in allowlist) alters semantics and can prevent upsampling for mixed allowlisted/non-allowlisted project queries, leading to underestimated counts"
+        },
+        {
+          "candidate": "Removed validation in organization_events_stats.py (top_events > 0 branch) that disallowed grouping by 'timestamp' (ParseError check), allowing previously forbidden groupby and potentially causing unexpected behavior or downstream errors"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "sample_rate = 0.0 is falsy and skipped",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The upsampling eligibility check passes the outer dataset instead of the actual dataset used by scoped_dataset. In paths where the query ultimately runs against discover (e.g., dashboard split) while the original dataset is metrics, upsampling may be skipped even when all projects are allowlisted.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 1,
+      "fp": 3,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry-greptile__tusk__PR3__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry-greptile__tusk__PR3__20260410/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/103633": {
@@ -31912,6 +32751,47 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR103633__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR103633__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The test comment says the cached permissions 'allow access', but the map stores false for dashboards:uid:dash1, so checkPermission will still treat this scope as not allowed.",
+          "severity": "Low",
+          "matched_candidate": "Test \"Should deny on explicit cache deny entry\" is invalid/misleading because it sets the permission cache entry to false while claiming it allows access, so the test would deny via the permission cache regardless of whether the denial cache is checked and therefore doesn't prove denial-cache priority",
+          "confidence": 0.9,
+          "reasoning": "Both point out the same inconsistency in the test: the comment/intent says cached permissions allow access, but the cache map sets the relevant scope to false, so checkPermission will treat it as denied. The candidate adds that this makes the test not prove denial-cache priority, which is a consequence of the same underlying issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Cache key collision risk in userPermDenialCacheKey because it concatenates name and parent with '_' even though both fields may contain underscores, allowing different resources to produce identical denial cache keys and cause incorrect access denials"
+        },
+        {
+          "candidate": "permDenialCache and permCache share the same underlying cache instance while storing different serialized Go types (bool vs map[string]bool) and use overlapping key prefixes, so an accidental key collision could cause json.Unmarshal type mismatches, spurious cache misses, and warning logs"
+        },
+        {
+          "candidate": "permissionCacheUsage metric increments with label \"false\" when the permission cache is hit but the requested resource isn't in the cached permission set, conflating a cache miss with a cache hit containing stale/incomplete data and reducing monitoring clarity"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The Check operation exhibits asymmetric cache trust logic: cached permission grants are trusted and returned immediately, but cached denials from the same permission cache are ignored, leading to a fresh database lookup. This allows stale cached grants to provide access to revoked resources, posing a security risk. ",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR103633__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR103633__20260411/pull/1"
     }
   },
   "https://github.com/getsentry/sentry/pull/67876": {
@@ -33542,6 +34422,54 @@
       "tool": "macroscope",
       "repo_name": "sentry__sentry__macroscope__PR67876__20260309",
       "pr_url": "https://github.com/code-review-benchmark/sentry__sentry__macroscope__PR67876__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "OAuth state uses pipeline.signature (static) instead of a per-request random value",
+          "severity": "Medium",
+          "matched_candidate": "OAuth2 `state` parameter is set to `pipeline.signature`, a deterministic MD5 of pipeline view class names, making it static across users/sessions and allowing attackers to forge `state` and bypass CSRF protection; `state` should include a cryptographically random, session-bound nonce",
+          "confidence": 0.97,
+          "reasoning": "The candidate explicitly notes that the OAuth2 `state` parameter is set to `pipeline.signature` and is deterministic/static, recommending a per-request cryptographically random, session-bound nonce. This matches the golden issue about using a static pipeline.signature instead of a per-request random value."
+        },
+        {
+          "golden_comment": "The code attempts to access integration.metadata[sender][login] without checking for the existence of the sender key. This causes a KeyError for integrations where the sender metadata was not set during creation",
+          "severity": "High",
+          "matched_candidate": "Accessing `integration.metadata[\"sender\"][\"login\"]` can raise `KeyError` when `sender` is missing (e.g., integrations created before sender metadata existed or via non-webhook paths), causing a 500 instead of a user-facing error",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same underlying issue: the code indexes into integration.metadata for sender/login without verifying the sender key exists, which can raise a KeyError for integrations lacking sender metadata (e.g., older or non-webhook-created integrations)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "When `Integration.DoesNotExist` occurs, the code calls `pipeline.next_step()` without performing the authenticated sender/user validation, allowing first-time installations to bypass the intended security check"
+        },
+        {
+          "candidate": "`redirect_uri` is included in the OAuth authorization request but omitted from the token exchange request, violating RFC 6749 requirements and potentially causing token exchange failures or unexpected provider behavior"
+        },
+        {
+          "candidate": "`error()` assumes `org.organization` exists even though `self.active_organization` may be `None` or an unexpected type if `determine_active_organization()` wasn't called or returns a different structure, leading to potential crashes in error handling"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Null reference if github_authenticated_user state is missing",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "sentry__sentry__tusk__PR67876__20260410",
+      "pr_url": "https://github.com/tusk-benchmark/sentry__sentry__tusk__PR67876__20260410/pull/1"
     }
   },
   "https://github.com/keycloak/keycloak/pull/32918": {
@@ -34856,6 +35784,44 @@
       "tool": "macroscope",
       "repo_name": "keycloak__keycloak__macroscope__PR32918__20260309",
       "pr_url": "https://github.com/code-review-benchmark/keycloak__keycloak__macroscope__PR32918__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Cleanup reference uses incorrect alias - should be 'idp-alias-' + i instead of 'alias'.",
+          "severity": "Medium",
+          "matched_candidate": "Test cleanup registers IDP removal using the literal string \"alias\" instead of the actual generated IDP alias (e.g., \"idp-alias-\" + i), so the IDPs created in the loop are not cleaned up and can leak state into other tests",
+          "confidence": 0.96,
+          "reasoning": "The candidate issue describes that cleanup uses the literal string \"alias\" instead of the generated IDP alias (\"idp-alias-\" + i), which is exactly the golden issue about incorrect alias usage in cleanup reference."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In getForLogin cached path, results are returned from a HashSet<IdentityProviderModel>, which does not preserve delegate query ordering, causing inconsistent ordering behavior between cached and non-cached paths for callers that might rely on ordering"
+        },
+        {
+          "candidate": "In registerIDPLoginInvalidationOnUpdate, the \"original\" model is org-aware wrapped (isEnabled depends on org state) while the \"updated\" model is raw; comparing them via getLoginPredicate can differ solely due to wrapper behavior, causing unnecessary (over-)invalidation of the login cache"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Recursive caching call using session instead of delegate",
+          "severity": "Critical"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "keycloak__keycloak__tusk__PR32918__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/keycloak__keycloak__tusk__PR32918__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/94942": {
@@ -36229,6 +37195,41 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR94942__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR94942__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The enableSqlExpressions function has flawed logic that always returns false, effectively disabling SQL expressions unconditionally:",
+          "severity": "Critical",
+          "matched_candidate": "`enableSqlExpressions` always returns `false` regardless of `FlagSqlExpressions`, making the feature-flag check dead code and misleading about the ability to re-enable SQL expressions",
+          "confidence": 0.96,
+          "reasoning": "The candidate states that `enableSqlExpressions` always returns false regardless of the flag, making the feature-flag check dead code. This matches the golden issue that the function\u2019s logic always returns false and thus disables SQL expressions unconditionally."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "`sql_command.go:Execute` appends an (empty) result frame to `rsp.Values` even when `DB.QueryFramesInto` returns an error (\"not implemented\"), leading to results containing empty frames alongside an error"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Several methods such as NewInMemoryDB().RunCommands and db.QueryFramesInto return 'not implemented'.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR94942__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR94942__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/90939": {
@@ -37482,6 +38483,37 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR90939__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR90939__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The GetWebAssets function implements an incomplete double-checked locking pattern for caching web assets. The function first checks if the cache is populated using a read lock (RLock), and if the cache is empty, it acquires a write lock to populate it. However, it fails to re-check whether the cache was populated by another goroutine while waiting to acquire the write lock.",
+          "severity": "Medium",
+          "matched_candidate": "After releasing the read lock and acquiring the write lock, the code does not re-check whether `entryPointAssetsCache` was populated by another goroutine, leading to redundant manifest reads/parses (double-checked locking bug)",
+          "confidence": 0.97,
+          "reasoning": "The candidate describes the same double-checked locking flaw: after dropping the read lock and acquiring the write lock, it fails to re-check whether the cache was already populated by another goroutine, causing redundant work."
+        }
+      ],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "In addition to the missing double-check, the function has a critical flaw in its error handling: it unconditionally assigns the fetch result to the cache (line 69: entryPointAssetsCache = result) regardless of whether the fetch succeeded or failed. When an error occurs during asset fetching, result is nil, and this nil value overwrites any previously valid cache entry.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 0,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR90939__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90939__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/80329": {
@@ -38825,6 +39857,45 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR80329__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR80329__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The code uses Error log level for what appears to be debugging information. This will pollute error logs in production. Consider using Debug or Info level instead.",
+          "severity": "Low",
+          "matched_candidate": "Cleanup-path diagnostic logs incorrectly use r.log.Error() for informational messages, which will flood error logs and potentially trigger alerts on successful cleanup cycles",
+          "confidence": 0.95,
+          "reasoning": "Both comments flag that the code logs diagnostic/informational messages at Error level, which is inappropriate and will pollute/flood production error logs (and may trigger alerts). They suggest using a lower log level instead."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Cleanup service ticker interval was changed from 10 minutes to 1 minute, likely an accidental debug leftover, causing cleanup to run 10x more frequently and increasing production database load"
+        },
+        {
+          "candidate": "Cleanup ticker interval (1 minute) is shorter than the clean() method timeout (9 minutes), so a new cleanup cycle can start while the previous one is still running (overlapping executions)"
+        },
+        {
+          "candidate": "SQLite parameter limit guard checks r.cfg.AnnotationCleanupJobBatchSize > sqliteParameterLimit instead of len(ids) > sqliteParameterLimit, causing unnecessary fallback to inline SQL when few IDs are returned and reducing safety vs parameterized queries"
+        },
+        {
+          "candidate": "Test helper createTestAnnotations hardcodes annotation IDs starting at 1, which can conflict with SQLite auto-increment behavior across shared test DB usage and lead to ID collisions when other tests insert rows without explicit IDs"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 1,
+      "tp": 1,
+      "fp": 4,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR80329__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR80329__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/90045": {
@@ -40633,6 +41704,50 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR90045__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR90045__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The context is being created with d.Log instead of the log variable that was initialized with additional context values (name, kind, method). This means those values won't be propagated to the logging context.",
+          "severity": "Medium",
+          "matched_candidate": "Delete method calls `klog.NewContext(ctx, d.Log)` with the base logger instead of the enriched `log` containing name/kind/method, losing contextual logging",
+          "confidence": 0.94,
+          "reasoning": "The candidate flags that klog.NewContext is called with d.Log (base logger) instead of the enriched `log` that includes name/kind/method, causing loss of contextual logging\u2014same underlying issue as the golden comment."
+        },
+        {
+          "golden_comment": "Bug: calling recordLegacyDuration when storage operation fails should be recordStorageDuration.",
+          "severity": "High",
+          "matched_candidate": "Update error path records latency with `recordLegacyDuration` instead of `recordStorageDuration`, misattributing a failed storage operation and corrupting metrics",
+          "confidence": 0.96,
+          "reasoning": "The candidate flags that in the error path for a failed storage operation, latency is recorded using `recordLegacyDuration` instead of `recordStorageDuration`, which is exactly the bug described in the golden comment."
+        },
+        {
+          "golden_comment": "Inconsistency: using name instead of options.Kind for metrics recording differs from other methods.",
+          "severity": "Medium",
+          "matched_candidate": "Delete success path passes `name` as the metric label to `recordStorageDuration` instead of `options.Kind`, creating high-cardinality Prometheus labels",
+          "confidence": 0.93,
+          "reasoning": "Both point out that the delete success path uses `name` as the metric label instead of `options.Kind`, which is inconsistent with other methods\u2019 metrics recording. The candidate adds the consequence (high-cardinality labels) but it\u2019s the same underlying issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "DeleteCollection goroutine records legacy write latency with `recordStorageDuration` instead of `recordLegacyDuration`, misattributing legacy latency to storage metrics"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 1,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR90045__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR90045__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/106778": {
@@ -42222,6 +43337,46 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR106778__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR106778__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The rendered GrafanaRuleListItem is missing the required key prop for React list items. This can cause rendering issues when the list order changes.",
+          "severity": "Medium",
+          "matched_candidate": "Missing React `key` prop on `GrafanaRuleListItem` when rendering the filter view list, which can cause React reconciliation problems on updates/reordering",
+          "confidence": 0.96,
+          "reasoning": "Both comments identify that GrafanaRuleListItem is rendered in a React list without a required `key` prop, leading to potential reconciliation/rendering issues when items update or reorder."
+        },
+        {
+          "golden_comment": "RuleActionsButtons is invoked with only promRule, but SilenceGrafanaRuleDrawer inside RuleActionsButtons still depends on a Grafana Ruler rule being present, so for Grafana rules coming from list views the 'Silence notifications' menu entry (now driven by Grafana Prom abilities) will toggle showSilenceDrawer without ever rendering the drawer. This means clicking 'Silence notifications' for these rules has no visible effect, even when abilities indicate silencing is allowed.",
+          "severity": "High",
+          "matched_candidate": "Silence drawer never opens in the new list view because `RuleActionsButtons.V2` only renders the drawer when `rulerRuleType.grafana.alertingRule(rule)` is true, but `rule` (RulerRuleDTO) is undefined when only `promRule` is passed",
+          "confidence": 0.9,
+          "reasoning": "Both describe the same underlying bug: in list views only `promRule` is provided to `RuleActionsButtons`, but the silence drawer rendering still depends on having a Grafana Ruler rule. As a result, clicking 'Silence notifications' toggles state but the drawer never renders/opens, producing no visible effect. The candidate frames it as `rule` being undefined and the drawer gated by `grafana.alertingRule(rule)`, which aligns with the golden issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Delete menu item never renders in the new list view because `AlertRuleMenu` guards it with `canDelete && rulerRule && (...)`, but `rulerRule` is always undefined from the `GrafanaRuleListItem -> RuleActionsButtons.V2 -> AlertRuleMenu` path"
+        },
+        {
+          "candidate": "`useGrafanaPromRuleAbility` is called with `promRule` typed as `Rule` instead of `GrafanaPromRuleDTO | SkipToken`; the boolean check `prometheusRuleType.grafana.rule()` does not narrow the type, risking a TypeScript error or incorrect permission checks when fields like `folderUid` are missing/optional"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR106778__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR106778__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/107534": {
@@ -43399,6 +44554,45 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR107534__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR107534__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Queries are interpolated twice in the shard-splitting path because both runShardSplitQuery() and runSplitQuery() call applyTemplateVariables(), causing non-idempotent interpolation side effects"
+        },
+        {
+          "candidate": "Ad-hoc filters are appended twice to the query expression because addAdHocFilters()/addLabelToQuery() unconditionally adds label matchers, so double interpolation duplicates matchers like env=\"prod\""
+        },
+        {
+          "candidate": "The shard splitting test masks the double-interpolation bug because its applyTemplateVariables mock uses idempotent string replacement and asserts the double-call pattern (5 calls), so it doesn\u2019t catch duplicated ad-hoc filters"
+        },
+        {
+          "candidate": "The applyTemplateVariables test mock mutates the input query object in place instead of returning a new object like the real implementation, potentially hiding bugs that depend on request.targets remaining unmodified"
+        },
+        {
+          "candidate": "runShardSplitQuery changed the order of filtering vs interpolation (filtering before interpolation), which can allow template-only expressions that interpolate to an empty string to slip through and be processed/sent as empty queries"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The applyTemplateVariables method is called with request.filters as the third parameter, but this parameter is not used in the corresponding test setup.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR107534__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR107534__20260411/pull/1"
     }
   },
   "https://github.com/grafana/grafana/pull/79265": {
@@ -45455,6 +46649,59 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR79265__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR79265__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Race condition: Multiple concurrent requests could pass the device count check simultaneously and create devices beyond the limit. Consider using a database transaction or lock.",
+          "severity": "High",
+          "matched_candidate": "Device limit enforcement has a TOCTOU race: CreateOrUpdateDevice counts devices then updates/inserts, allowing concurrent requests to exceed the configured device limit",
+          "confidence": 0.96,
+          "reasoning": "Both describe a race/TOCTOU condition where concurrent requests can pass the device count check and then insert/update, exceeding the device limit; both imply need for transactional/locking protection."
+        },
+        {
+          "golden_comment": "Anonymous authentication now fails entirely if anonDeviceService.TagDevice returns ErrDeviceLimitReached. Previously, device tagging was asynchronous and non-blocking. This change prevents anonymous users from authenticating when the device limit is reached.",
+          "severity": "Medium",
+          "matched_candidate": "TagDevice was changed from asynchronous fire-and-forget to synchronous on the authentication request path, so slow/failing DB operations now block anonymous authentication and add latency to every anonymous request",
+          "confidence": 0.78,
+          "reasoning": "Both comments note that TagDevice moved from async/non-blocking to synchronous on the auth path, which can block anonymous authentication when TagDevice is slow or fails. The golden specifically calls out ErrDeviceLimitReached causing auth to fail; the candidate generalizes to slow/failing DB ops blocking auth. Same underlying regression (tagging now blocks auth), though candidate doesn\u2019t mention the specific error case."
+        },
+        {
+          "golden_comment": "Returning ErrDeviceLimitReached when no rows were updated is misleading; the device might not exist.",
+          "severity": "Low",
+          "matched_candidate": "updateDevice returns ErrDeviceLimitReached when rowsAffected == 0, conflating 'device not found' with 'device exists but is outside the updated_at BETWEEN window (expired)', causing expired returning devices to be incorrectly rejected and producing a misleading error",
+          "confidence": 0.9,
+          "reasoning": "Both point out that returning ErrDeviceLimitReached when rowsAffected==0 is misleading because zero updated rows can mean the device doesn't exist (or otherwise didn't match the update criteria), not that the device limit was reached. The candidate adds an extra nuance (expired window) but the core conflation/misleading error is the same."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "FrontendSettingsDTO.AnonymousDeviceLimit always serializes as a number (0 when unset) due to missing omitempty, but the TypeScript type is number | undefined, so frontend checks for undefined to mean 'no limit configured' will never work"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "This call won\u2019t compile: dbSession.Exec(args...) is given a []interface{} where the first element is the query, but Exec\u2019s signature requires a first parameter of type string (not an interface{} splat).",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Time window calculation inconsistency: Using device.UpdatedAt.UTC().Add(-anonymousDeviceExpiration) as the lower bound but device.UpdatedAt as the current time may not match the intended logic. Consider using time.Now().UTC() consistently.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 3,
+      "fp": 1,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.6,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR79265__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR79265__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/9": {
@@ -46883,112 +48130,153 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR9__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR9__20260309/pull/1"
-    }
-  },
-  "https://github.com/grafana/grafana/pull/76186": {
-    "gemini": {
+    },
+    "tusk": {
       "skipped": false,
-      "true_positives": [],
-      "false_positives": [
-        {
-          "candidate": "SubscribeStream is not instrumented with a contextual logger despite having access to PluginContext, leading to missing contextual log fields (e.g., pluginId)"
-        },
-        {
-          "candidate": "RunStream is not instrumented with a contextual logger despite having access to PluginContext, leading to missing contextual log fields (e.g., pluginId)"
-        },
+      "true_positives": [
         {
-          "candidate": "A test function name is not aligned with the file rename and should be renamed for consistency"
+          "golden_comment": "Thread-safety issue with lazy @loaded_locales",
+          "severity": "Low",
+          "matched_candidate": "ensure_loaded! initializes @loaded_locales ||= [] outside the LOAD_MUTEX, creating a race where concurrent threads can overwrite the array and lose loaded-locale state, leading to redundant loads or missed locale loading",
+          "confidence": 0.95,
+          "reasoning": "The golden comment flags a thread-safety problem with lazy initialization of @loaded_locales. The candidate explicitly describes a race condition where @loaded_locales ||= [] is done outside the mutex, allowing concurrent threads to overwrite/lose state, which is exactly the same underlying thread-safety issue."
         }
       ],
-      "false_negatives": [
+      "false_positives": [
         {
-          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
-          "severity": "High"
+          "candidate": "LRU translation cache key omits SiteSetting.default_locale even though it affects the fallback chain, so changing default_locale can serve stale cached translations resolved via the old fallback chain"
         },
         {
-          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
-          "severity": "Low"
-        }
-      ],
-      "errors": [],
-      "total_candidates": 3,
-      "total_golden": 2,
-      "tp": 0,
-      "fp": 3,
-      "fn": 2,
-      "errors_count": 0,
-      "precision": 0.0,
-      "recall": 0.0,
-      "tool": "gemini",
-      "repo_name": "grafana__grafana__gemini__PR76186__20260128",
-      "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__gemini__PR76186__20260128/pull/1"
-    },
-    "claude": {
-      "skipped": false,
-      "true_positives": [],
-      "false_positives": [],
-      "false_negatives": [
-        {
-          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
-          "severity": "High"
+          "candidate": "No I18n.reload! (or equivalent cache invalidation) is triggered when SiteSetting.default_locale changes, leaving the translation cache populated with entries based on the previous default locale"
         },
         {
-          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
-          "severity": "Low"
+          "candidate": "translate only ensures config.locale is loaded; with I18n::Backend::Fallbacks enabled, fallback locales in the chain may not be loaded in non-controller contexts (Sidekiq/rake/console), causing fallbacks to silently fail"
         }
       ],
-      "errors": [],
-      "total_candidates": 0,
-      "total_golden": 2,
-      "tp": 0,
-      "fp": 0,
-      "fn": 2,
-      "errors_count": 0,
-      "precision": 0.0,
-      "recall": 0.0,
-      "tool": "claude",
-      "repo_name": "grafana__grafana__claude__PR76186__20260127",
-      "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__claude__PR76186__20260127/pull/1"
-    },
-    "graphite": {
-      "skipped": false,
-      "true_positives": [],
-      "false_positives": [],
       "false_negatives": [
         {
-          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
-          "severity": "High"
-        },
-        {
-          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "golden_comment": "Consider normalizing the input locale (e.g., to a symbol) when checking/loading here to avoid double-loading if the same locale is passed as a String vs Symbol (also applies to other locations in the PR).",
           "severity": "Low"
         }
       ],
       "errors": [],
-      "total_candidates": 0,
+      "total_candidates": 4,
       "total_golden": 2,
-      "tp": 0,
-      "fp": 0,
-      "fn": 2,
+      "tp": 1,
+      "fp": 3,
+      "fn": 1,
       "errors_count": 0,
-      "precision": 0.0,
-      "recall": 0.0,
-      "tool": "graphite",
-      "repo_name": "grafana__grafana__graphite__PR76186__20260123",
-      "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__graphite__PR76186__20260123/pull/1"
-    },
-    "qodo": {
+      "precision": 0.25,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR9__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR9__20260411/pull/1"
+    }
+  },
+  "https://github.com/grafana/grafana/pull/76186": {
+    "gemini": {
       "skipped": false,
       "true_positives": [],
       "false_positives": [
         {
-          "candidate": "TestLogger.FromContext returns a brand-new logger instance, potentially dropping previously accumulated logger state/context and masking production bugs; it should preserve existing context (e.g., return f or f.New(...))"
+          "candidate": "SubscribeStream is not instrumented with a contextual logger despite having access to PluginContext, leading to missing contextual log fields (e.g., pluginId)"
         },
         {
-          "candidate": "grafanaInfraLogWrapper.FromContext depends on a type assertion to *log.ConcreteLogger; if the underlying logger is a different implementation it falls back to d.New() and may unexpectedly discard contextual attributes"
+          "candidate": "RunStream is not instrumented with a contextual logger despite having access to PluginContext, leading to missing contextual log fields (e.g., pluginId)"
         },
         {
-          "candidate": "Contextual logging enrichment is moved to a separate middleware while LoggerMiddleware consumes logger.FromContext(ctx), creating a fragile dependency on middleware ordering/inclusion; missing or misordered middleware can silently degrade log fields"
+          "candidate": "A test function name is not aligned with the file rename and should be renamed for consistency"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 2,
+      "tp": 0,
+      "fp": 3,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "gemini",
+      "repo_name": "grafana__grafana__gemini__PR76186__20260128",
+      "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__gemini__PR76186__20260128/pull/1"
+    },
+    "claude": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 0,
+      "total_golden": 2,
+      "tp": 0,
+      "fp": 0,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "claude",
+      "repo_name": "grafana__grafana__claude__PR76186__20260127",
+      "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__claude__PR76186__20260127/pull/1"
+    },
+    "graphite": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 0,
+      "total_golden": 2,
+      "tp": 0,
+      "fp": 0,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "graphite",
+      "repo_name": "grafana__grafana__graphite__PR76186__20260123",
+      "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__graphite__PR76186__20260123/pull/1"
+    },
+    "qodo": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "TestLogger.FromContext returns a brand-new logger instance, potentially dropping previously accumulated logger state/context and masking production bugs; it should preserve existing context (e.g., return f or f.New(...))"
+        },
+        {
+          "candidate": "grafanaInfraLogWrapper.FromContext depends on a type assertion to *log.ConcreteLogger; if the underlying logger is a different implementation it falls back to d.New() and may unexpectedly discard contextual attributes"
+        },
+        {
+          "candidate": "Contextual logging enrichment is moved to a separate middleware while LoggerMiddleware consumes logger.FromContext(ctx), creating a fragile dependency on middleware ordering/inclusion; missing or misordered middleware can silently degrade log fields"
         }
       ],
       "false_negatives": [
@@ -48244,6 +49532,41 @@
       "tool": "macroscope",
       "repo_name": "grafana__grafana__macroscope__PR76186__20260309",
       "pr_url": "https://github.com/code-review-benchmark/grafana__grafana__macroscope__PR76186__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) panic when a nil request is received. This occurs because they directly access req.PluginContext (via the instrumentContext function) without first checking if req is nil. This is a regression, as previous middleware layers gracefully handled nil requests.",
+          "severity": "High",
+          "matched_candidate": "ContextualLoggerMiddleware methods (QueryData, CallResource, CheckHealth, CollectMetrics) dereference req.PluginContext without checking if req is nil, risking a nil pointer panic when used outside the Decorator or in tests",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same underlying bug: ContextualLoggerMiddleware methods access req.PluginContext without a nil check, causing a panic when req is nil (notably in tests or when used outside expected wrapping). The candidate matches the golden issue; it just omits the explicit 'regression' note but identifies the same failure mode and cause."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "TestLogger.FromContext creates and returns a new disconnected TestLogger via NewTestLogger(), so LoggerMiddleware.logRequest logs to a throwaway instance and tests cannot capture/verify log output"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The traceID is no longer logged for plugin requests. During a refactoring, the tracing import and the logic to extract and add traceID from the context to log parameters were removed from the LoggerMiddleware. The newly introduced ContextualLoggerMiddleware does not add this information, resulting in missing traceID in plugin request logs and impacting debugging and request tracing capabilities.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 2,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 1,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "grafana__grafana__tusk__PR76186__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/grafana__grafana__tusk__PR76186__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/10": {
@@ -50468,6 +51791,61 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR10__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR10__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "NoMethodError before_validation in EmbeddableHost",
+          "severity": "Critical",
+          "matched_candidate": "EmbeddableHost model before_validation callback calls self.host.sub! even when host is nil, causing NoMethodError before validations can reject missing host (e.g., when params[:embeddable_host][:host] is absent)",
+          "confidence": 0.93,
+          "reasoning": "The golden comment flags a NoMethodError occurring in a before_validation callback in EmbeddableHost. The candidate describes that before_validation calls `self.host.sub!` when `host` is nil, triggering a NoMethodError before validations handle missing host. This is the same underlying issue."
+        },
+        {
+          "golden_comment": "The update and destroy methods in Admin::EmbeddableHostsController do not validate the existence of the EmbeddableHost record retrieved by ID. If EmbeddableHost.where(id: params[:id]).first returns nil (i.e., the host does not exist), attempting to call methods on the nil object (e.g., save_host or destroy) will result in a NoMethodError.",
+          "severity": "Medium",
+          "matched_candidate": "Admin EmbeddableHostsController destroy/update actions can crash when EmbeddableHost.where(id: params[:id]).first returns nil; destroy calls host.destroy and update passes nil into save_host leading to NoMethodError",
+          "confidence": 0.96,
+          "reasoning": "The candidate issue describes the same problem: update/destroy fetch the record with `EmbeddableHost.where(...).first` without handling nil, leading to NoMethodError when calling `destroy` or passing nil to `save_host`."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Migration can crash on fresh installs when the 'embed_category' site setting is missing because the SQL query returns 0 rows and code accesses [0]['id'], causing NoMethodError and bypassing the category_id == 0 fallback"
+        },
+        {
+          "candidate": "Migration builds SQL INSERT statements by interpolating host values ('#{h}') from the old embeddable_hosts setting, allowing single quotes to break SQL and enabling SQL injection if the setting is corrupted"
+        },
+        {
+          "candidate": "Fabricator definitions are in the wrong files: category_fabricator.rb contains Fabricator(:embeddable_host) while embeddable_host_fabricator.rb contains category fabricators, causing confusing mismatch between filenames and contents"
+        },
+        {
+          "candidate": "JavaScript basePath uses type.replace('_','-') which only replaces the first underscore, so types with multiple underscores will be converted incorrectly and may fail to match ADMIN_MODELS"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "record_for_host compares lower(host) = ? but does not normalize the parameter\u2019s case, so mixed\u2011case referer hosts may fail to match even though comparison intends to be case\u2011insensitive.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Because this migration inserts embeddable_hosts rows with raw SQL, any existing embeddable_hosts values that include http:// or /https:// or path segments won\u2019t go through the EmbeddableHost model\u2019s normalization, so the new host lookup (which compares only the bare host) may fail for migrated data. Consider ensuring that migrated hosts are normalized to the same format as newly created EmbeddableHost records so existing embedding configurations keep working.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 4,
+      "tp": 2,
+      "fp": 4,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR10__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR10__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/7": {
@@ -52147,6 +53525,53 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR7__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR7__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In .topic-meta-data h5 a, the original code had color: scale-color($primary, $lightness: 30%) but was changed to dark-light-choose(scale-color($primary, $lightness: 70%), scale-color($secondary, $lightness: 30%)). The lightness for the light theme changed from 30% to 70%, which is a dramatic inversion",
+          "severity": "Low",
+          "matched_candidate": "In app/assets/stylesheets/desktop/topic-post.scss around line 291, the light-theme lightness for .embedded-posts link text was changed from 30% to 70%, making it washed out; the light/dark theme lightness values appear to be accidentally swapped",
+          "confidence": 0.78,
+          "reasoning": "Both describe the same underlying regression: a link color in topic-post.scss had its light-theme lightness changed from 30% to 70% due to switching to dark-light-choose, effectively inverting/misapplying the intended light/dark values and making the light theme too light/washed out. Candidate references a closely related selector (.embedded-posts link text) rather than explicitly .topic-meta-data h5 a, but the core issue (30% -> 70% light-theme lightness swap) matches."
+        },
+        {
+          "golden_comment": "This change for desktop/user.css changes $primary from 30% to 50% for the light theme; most other changes preserve the original $primary value and move the complement to $secondary for dark. Consider reviewing this (also applies to a similar .name change in the mobile variant).",
+          "severity": "Low",
+          "matched_candidate": "In app/assets/stylesheets/desktop/user.scss around line 522 (and corresponding mobile user stylesheet), the light-theme lightness for .name changed from 30% to 50%, making names washed out and inconsistent with .username a in common/base/user.scss which keeps 30%",
+          "confidence": 0.86,
+          "reasoning": "Both comments flag the same underlying regression: in the light theme, the .name styling\u2019s lightness/primary-related value was changed from 30% to 50% (desktop and similarly in mobile), deviating from the prior scheme where $primary stayed 30% and complements moved to $secondary for dark. The candidate focuses specifically on .name and its inconsistency, which is explicitly mentioned in the golden comment as a similar change to review."
+        },
+        {
+          "golden_comment": "In topic-post.css the original code used $lightness: 70% but the replacement uses $lightness: 30% for the light theme. This makes the text significantly darker than intended.",
+          "severity": "Low",
+          "matched_candidate": "In app/assets/stylesheets/desktop/topic-post.scss around line 291, the light-theme lightness for .embedded-posts link text was changed from 30% to 70%, making it washed out; the light/dark theme lightness values appear to be accidentally swapped",
+          "confidence": 0.72,
+          "reasoning": "Both comments concern the light theme lightness value being incorrect/swapped between 70% and 30% for embedded-post link text, leading to unintended appearance. The candidate describes the inverse direction (30%->70% vs 70%->30%) but flags the same underlying issue: the light theme lightness value is wrong/accidentally swapped."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In app/assets/stylesheets/mobile/modal.scss around line 102, dark-light-choose arguments for .custom-message-length are reversed compared to desktop (mobile uses 30% for light and 70% for dark), causing inverted character count contrast on light vs dark themes"
+        },
+        {
+          "candidate": "In app/assets/stylesheets/mobile/topic-post.scss around line 182, the light-theme lightness for h3 changed from 20% to 50%, making headings too grey on light themes; the dark-theme counterpart should be ~80% (complementary to 20%) rather than 50%"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 3,
+      "tp": 3,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.75,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR7__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR7__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/8": {
@@ -53994,6 +55419,57 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR8__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR8__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In the next action, capping the next offset at user_count can produce an empty page (e.g., total equal to limit results in offset == total, showing 2/2 with no members). This can cause confusing UX on the last page.",
+          "severity": "Medium",
+          "matched_candidate": "Off-by-one bug in totalPages calculation: using Math.floor(user_count / limit) + 1 produces an extra page when user_count is exactly divisible by limit; should use Math.ceil(user_count / limit)",
+          "confidence": 0.86,
+          "reasoning": "Both comments describe the same boundary-condition pagination bug when total/user_count is exactly divisible by limit. The golden frames it as capping next offset to user_count leading to an empty last page (offset == total). The candidate frames it as totalPages being computed with floor+1 instead of ceil, creating an extra page. These are two manifestations of the same underlying issue: an extra/empty final page due to incorrect handling of exact multiples of limit."
+        },
+        {
+          "golden_comment": "HTTP method mismatch in .remove_member - test uses PUT but remove_member action expects DELETE",
+          "severity": "Medium",
+          "matched_candidate": "Controller spec for removing members from automatic groups uses xhr :put even though the route is DELETE, so the test doesn't validate the real HTTP method/path and could miss routing/method bugs",
+          "confidence": 0.96,
+          "reasoning": "Both describe the same issue: the test uses PUT (xhr :put) while the remove_member route/action expects DELETE, meaning the spec isn't exercising the correct HTTP method and could miss routing/method problems."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Public GroupsController#members endpoint response format changed from a flat JSON array to an object with {members, meta}, breaking existing consumers expecting an array"
+        },
+        {
+          "candidate": "GroupsController#members now always paginates with a default limit (50), so callers not providing pagination params may silently receive truncated member lists (previously non-automatic groups returned all members)"
+        },
+        {
+          "candidate": "Admin groups update unconditionally sets group.visible based on params[:visible] == \"true\"; if visible is omitted in a partial update, visibility is unintentionally set to false"
+        },
+        {
+          "candidate": "New group creation visibility can be incorrectly forced to false because the JS model asJSON() sends visible: !!this.get('visible') (false when unset) and the create action interprets params[:visible] == \"true\", so checkbox state/defaults may not be respected"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": " The findMembers() call is now asynchronous and unhandled. The controller may not have member data immediately available, creating a race condition.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR8__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR8__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/3": {
@@ -55453,6 +56929,44 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR3__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR3__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "BlockedEmail.should_block_email? method has side effects during a read operation - it updates statistics even when just checking if an email should be blocked. This could cause race conditions in concurrent environments and makes the method name misleading.",
+          "severity": "Medium",
+          "matched_candidate": "BlockedEmail.should_block? performs DB writes (incrementing match_count and updating last_match_at) as a side effect during model validation, so repeated validation calls can inflate blocked-email statistics",
+          "confidence": 0.9,
+          "reasoning": "Both comments flag that the should_block* method performs database writes (updating match_count/last_match_at statistics) as a side effect when it is ostensibly just checking/validating whether an email should be blocked. This matches the golden concern about side effects during a read/check operation and misleading behavior, with the candidate adding the specific context of validation and inflated stats."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Client-side create account JS adds the email to rejectedEmails for any email validation error (including uniqueness/taken email), causing valid emails to be permanently marked rejected until page refresh"
+        },
+        {
+          "candidate": "rejectedEmails is initialized as Em.A([]) on the Ember class prototype, so the array is shared across controller instances and can cause cross-instance state pollution"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Regex pattern @(#{domains}) only matches domain suffixes, not full domains. evil.example.com would match whitelist entry example.com.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR3__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR3__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/5": {
@@ -56884,6 +58398,53 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR5__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR5__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "-ms-align-items never existed in any version of IE/Edge; the correct legacy property is -ms-flex-align.",
+          "severity": "Low",
+          "matched_candidate": "The mixin uses invalid `-ms-align-items`, which IE10/11 will ignore; only `-ms-flex-align` should be used for IE",
+          "confidence": 0.96,
+          "reasoning": "The candidate flags `-ms-align-items` as invalid/ignored in IE10/11 and states that `-ms-flex-align` is the correct property, which is the same underlying issue as the golden comment."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "The `order` mixin passes raw `$val` to `-webkit-box-ordinal-group`/`-moz-box-ordinal-group` even though those legacy properties are 1-based, so it should use `$val + 1` to match standard `order` behavior in old WebKit/Firefox"
+        },
+        {
+          "candidate": "In `.small-action`, `.topic-avatar` still has `float: left` even though the parent is now `display: flex`, indicating an incomplete flex migration and causing confusing/obsolete styling"
+        },
+        {
+          "candidate": "In `.small-action-desc`, changing padding from `0.5em 0 0.5em 4em` to `0 1.5%` removes the fixed left spacing for the avatar, potentially shrinking the gap on narrow viewports and causing text to appear too tight or overlap"
+        },
+        {
+          "candidate": "The `align-items` mixin does not translate `flex-start`/`flex-end` to IE10/11 `-ms-flex-align` values (`start`/`end`), so callers using `flex-start`/`flex-end` would break in IE"
+        },
+        {
+          "candidate": "`.badge-wrapper.bullet` has inconsistent `margin-top: 5px` between contexts (removed in one rule but retained under `#topic-title .title-wrapper`), leading to inconsistent bullet badge alignment/styling"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Mixing float: left with flexbox causes layout issues. Further this PR removes the float-based right alignment for .d-header .panel, which may cause the login panel in the non-Ember/noscript header (where .panel is nested inside .row and not a flex item) to stack under the title instead of remaining right-aligned.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR5__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR5__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/6": {
@@ -58167,6 +59728,39 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR6__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR6__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Serializer conditional include hook is misnamed: `include_website_name` lacks the required trailing `?`, so ActiveModel::Serializer never calls it and `website_name` is always serialized even when `website` is nil/blank"
+        },
+        {
+          "candidate": "`website_name` is missing from the `untrusted_attributes` list while `website` is included, allowing TL0 users' website info to leak via `website_name` when `scope.restrict_user_fields?` should hide it"
+        },
+        {
+          "candidate": "Test setup is fragile: `Discourse.stubs(:current_hostname)` is applied inside the `it` block after `let(:json)` is defined, so if `json` is evaluated before the stub (e.g., in a `before`/shared example) the stub won\u2019t take effect; the stub should be set before `json` is accessed"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The include_website_name method is missing the required ? suffix. Rails serializers expect include_ methods to end with ? for conditional attribute inclusion, a convention followed by other methods in this serializer. Without it, the website_name attribute may not be conditionally included as intended. Additionally, the '.' << website_host string concatenation should be replaced with '.' + website_host or '.#{website_host}' to avoid mutating string literals, which can lead to issues.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 3,
+      "total_golden": 1,
+      "tp": 0,
+      "fp": 3,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR6__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR6__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/4": {
@@ -60777,6 +62371,69 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR4__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR4__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The current origin validation using indexOf is insufficient and can be bypassed. An attacker could use a malicious domain like evil-discourseUrl.com to pass this check.",
+          "severity": "Medium",
+          "matched_candidate": "Insecure postMessage origin validation in app/assets/javascripts/embed.js line 17: using `discourseUrl.indexOf(e.origin)` performs substring matching and can be bypassed by attacker-controlled domains that contain the expected origin as a substring/prefix",
+          "confidence": 0.96,
+          "reasoning": "Both describe the same flaw: origin validation uses `indexOf`/substring matching, allowing bypass with attacker-controlled domains that include the trusted domain as a substring (e.g., evil-discourseUrl.com)."
+        },
+        {
+          "golden_comment": "The ERB block closes with end if, which is invalid Ruby/ERB and will raise at render; it should just be end to close the if block.",
+          "severity": "Medium",
+          "matched_candidate": "Invalid ERB/Ruby syntax in app/views/embed/best.html.erb line 6: `<%- end if %>` uses `end if` without a condition and should be `<%- end %>`, causing a SyntaxError when rendering the template",
+          "confidence": 0.98,
+          "reasoning": "Both describe the same ERB/Ruby syntax error: the template closes an if block with `end if`, which is invalid here and should be just `end`, otherwise it raises a SyntaxError at render time."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "RSpec controller specs do not render views by default, so the ERB SyntaxError in best.html.erb is not caught by the existing controller spec coverage"
+        },
+        {
+          "candidate": "Potential XSS in app/views/layouts/embed.html.erb line 11: `request.referer` is interpolated into a single-quoted JavaScript string without JS-escaping/sanitizing path/query/fragment, allowing JS string breakout despite HTML escaping"
+        },
+        {
+          "candidate": "Nil handling bug in app/jobs/scheduled/poll_feed.rb line 35: `i.content` can be nil for RSS items lacking `content:encoded`, so calling `.scrub` can raise `NoMethodError`"
+        },
+        {
+          "candidate": "Unintended string mutation in app/models/topic_embed.rb line 13: using `contents << ...` mutates the caller's string, causing the footer to be appended multiple times across repeated imports and leading to incorrect content SHA1 comparisons on re-import/update"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "SSRF vulnerability using open(url) without validation",
+          "severity": "Critical"
+        },
+        {
+          "golden_comment": "postMessage targetOrigin should be the origin (scheme+host+port), not the full referrer URL; using the full URL will cause the message to be dropped and prevent resizing.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "The code sets X-Frame-Options: ALLOWALL which completely disables clickjacking protection. The referer validation can be bypassed (referer headers are easily spoofed), and the fallback to empty string for nil referer masks validation failures.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "The TopicEmbed.import method is susceptible to a NoMethodError if the contents parameter is nil when attempting to append a string, and an XSS vulnerability due to unescaped url interpolation in the generated HTML.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 6,
+      "tp": 2,
+      "fp": 4,
+      "fn": 4,
+      "errors_count": 0,
+      "precision": 0.3333333333333333,
+      "recall": 0.3333333333333333,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR4__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR4__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/1": {
@@ -62547,6 +64204,57 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR1__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR1__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The downsize method is defined twice. The second definition, which expects a single dimensions string parameter, overrides the first, which expected separate max_width and max_height parameters. This makes the original method unreachable and breaks existing code that calls it with separate width and height arguments.",
+          "severity": "Medium",
+          "matched_candidate": "In app/models/optimized_image.rb, the method `downsize` is defined twice; the second definition overrides the first, leaving only the 3-arg/4-arg signature and removing the 5-arg variant",
+          "confidence": 0.86,
+          "reasoning": "Both comments flag that `downsize` is defined twice and that the later definition overrides the earlier one, making the original signature unreachable and potentially breaking callers. Although the candidate\u2019s description of the signatures differs (mentions 3/4-arg vs 5-arg rather than dimensions-string vs width/height), the core issue\u2014method redefinition overriding the prior API\u2014is the same."
+        },
+        {
+          "golden_comment": "Hardcoding maxSizeKB = 10 * 1024 ignores Discourse.SiteSettings['max_' + type + '_size_kb'], so the client-side limit can diverge from server-side and per-type settings (also applies to the 413 handler below).",
+          "severity": "Low",
+          "matched_candidate": "In app/assets/javascripts/discourse/lib/utilities.js, client-side upload size validation is hardcoded to 10MB for all upload types, ignoring the dynamic site settings (e.g., max_image_size_kb/max_attachment_size_kb)",
+          "confidence": 0.9,
+          "reasoning": "The candidate flags that client-side upload size validation is hardcoded to 10MB for all types and ignores per-type dynamic site settings like max_image_size_kb/max_attachment_size_kb. This matches the golden issue about hardcoding maxSizeKB = 10*1024 instead of using Discourse.SiteSettings['max_' + type + '_size_kb'], causing divergence from server-side/per-type limits (including the 413 handler implication)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Because client-side validation is hardcoded to 10MB, non-image attachments may be accepted client-side above the configured limits and then rejected by the server"
+        },
+        {
+          "candidate": "In app/assets/javascripts/discourse/lib/utilities.js, the 413 (entity too large) error handler hardcodes `maxSizeKB` to 10MB, causing the displayed error message to be incorrect when the real server limit differs"
+        },
+        {
+          "candidate": "In app/controllers/uploads_controller.rb, the API URL download limit is hardcoded to `10.megabytes` instead of using `SiteSetting.max_image_size_kb.kilobytes`, artificially limiting uploads when the admin setting is higher than 10MB"
+        },
+        {
+          "candidate": "In app/controllers/uploads_controller.rb, the hardcoded 10MB API URL download limit applies to non-image URL uploads as well because the type check occurs later"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Passing 80% as the dimensions can fail for animated GIFs when allow_animated_thumbnails is true, since the animated path uses gifsicle --resize-fit which expects WxH geometry, not a percentage; downsizing would then silently fail.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 7,
+      "total_golden": 3,
+      "tp": 2,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2857142857142857,
+      "recall": 0.6666666666666666,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR1__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR1__20260411/pull/1"
     }
   },
   "https://github.com/ai-code-review-evaluation/discourse-graphite/pull/2": {
@@ -64109,6 +65817,50 @@
       "tool": "macroscope",
       "repo_name": "discourse__discourse-graphite__macroscope__PR2__20260309",
       "pr_url": "https://github.com/code-review-benchmark/discourse__discourse-graphite__macroscope__PR2__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "logic: Potential nil pointer exception - if no TopicUser record exists, tu will be nil and calling methods on it will crash",
+          "severity": "High",
+          "matched_candidate": "Possible NoMethodError crash in topics_controller: TopicUser.find_by can return nil and code calls tu.notification_level without a nil check",
+          "confidence": 0.96,
+          "reasoning": "Both describe the same issue: TopicUser.find_by may return nil when no record exists, and subsequent calls on tu (e.g., notification_level) would raise a NoMethodError/nil pointer crash without a nil check."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "I18n unsubscribe_link string now requires %{unsubscribe_url}; email paths using add_unsubscribe_link: true without providing unsubscribe_url can raise I18n::MissingInterpolationArgument or produce an empty/broken link"
+        },
+        {
+          "candidate": "XSS risk in dropdown-button component: title is concatenated into HTML without escaping, allowing user-controlled content to render as raw HTML"
+        },
+        {
+          "candidate": "XSS risk in topic unsubscribe template/controller: triple-stache {{{stopNotificiationsText}}} renders raw HTML and interpolates model.fancyTitle into an i18n string, potentially allowing malicious HTML to be rendered unescaped"
+        },
+        {
+          "candidate": "Email html_part placeholder substitution bug: %{respond_instructions} and %{unsubscribe_link} are only replaced when add_unsubscribe_link is true, but the template always includes %{respond_instructions}, so literals can appear in emails when add_unsubscribe_link is false"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Typo in property name: 'stopNotificiationsText' should be 'stopNotificationsText' (missing 'n' in 'Notifications')",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "discourse__discourse-graphite__tusk__PR2__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/discourse__discourse-graphite__tusk__PR2__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/22532": {
@@ -65711,6 +67463,53 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR22532__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR22532__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The updateManyByCredentialId call uses an empty data object, which prevents Prisma's @updatedAt decorator from updating the updatedAt timestamp. This results in inaccurate cache status tracking, as the timestamp isn't updated when the cache is refreshed. To fix this, explicitly set the updatedAt field.",
+          "severity": "Medium",
+          "matched_candidate": "Calling `updateManyByCredentialId(credentialId, {})` with an empty update object is a no-op, so `SelectedCalendar.updatedAt` is never updated as intended (Prisma `updateMany` doesn\u2019t trigger `@updatedAt`).",
+          "confidence": 0.93,
+          "reasoning": "The candidate flags that calling updateManyByCredentialId with an empty update object is a no-op and therefore Prisma\u2019s @updatedAt/updatedAt field is not updated. This is the same core issue as the golden comment (timestamp not refreshed due to empty data object), even though it doesn\u2019t explicitly mention the cache tracking impact or the fix."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "`deleteCache` handler authorizes credentials only by `credential.userId === user.id`, ignoring delegation via `delegationCredentialId`, which can prevent delegated users from deleting their own cache."
+        },
+        {
+          "candidate": "`deleteCache` handler does not verify the credential is a calendar credential, so any user-owned credential ID could be used to attempt calendar cache deletion."
+        },
+        {
+          "candidate": "`CalendarCache.updatedAt` relies on Prisma `@updatedAt`, but raw SQL upserts/inserts (e.g., `$queryRaw` in `upsertCachedAvailability`) won\u2019t trigger it, so `updatedAt` may never update beyond the DB default insert timestamp."
+        },
+        {
+          "candidate": "Using `cacheStatusMap.get(id) || null` conflates `undefined` (missing key) with `null` (present-but-null) and should use `?? null` for correct semantics."
+        },
+        {
+          "candidate": "`CredentialActionsDropdown` is rendered unconditionally and initializes state and tRPC mutations even when it returns `null`, causing unnecessary hook/mutation setup overhead (including `trpc.viewer.credentials.delete.useMutation()` subscriptions) for credentials that won\u2019t show the dropdown."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "logic: macOS-specific sed syntax with empty string after -i flag will fail on Linux systems",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 6,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 5,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.16666666666666666,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR22532__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22532__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/8330": {
@@ -67224,6 +69023,46 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR8330__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR8330__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Incorrect end time calculation using slotStartTime instead of slotEndTime",
+          "severity": "Medium",
+          "matched_candidate": "Copy-paste bug: `end` is computed from `slotStartTime` instead of `slotEndTime`, so slots whose end extends past working hours may not be rejected",
+          "confidence": 0.97,
+          "reasoning": "The candidate explicitly states that `end` is computed from `slotStartTime` instead of `slotEndTime`, which is exactly the incorrect end time calculation described in the golden comment."
+        },
+        {
+          "golden_comment": "Using === for dayjs object comparison will always return false as it compares object references, not values. Use .isSame() method instead: dayjs(date.start).add(utcOffset, 'minutes').isSame(dayjs(date.end).add(utcOffset, minutes))",
+          "severity": "Medium",
+          "matched_candidate": "Uses object reference equality (===) to compare two different dayjs instances, so the override start==end (zero-length/unavailable-day) check never triggers and unavailable-day overrides are treated as available",
+          "confidence": 0.94,
+          "reasoning": "Both describe the same bug: comparing two separately created dayjs instances with === checks reference equality, so it won\u2019t detect equal dates/times. The golden suggests using .isSame() for value comparison; the candidate explains the same consequence (start==end check never triggers)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Timezone conversion bug: `dayjs(date.start)` is parsed in server-local time then adjusted by a negated organizer offset, causing incorrect override boundaries when server timezone differs from UTC and creating inconsistency with other offset logic"
+        },
+        {
+          "candidate": "Incorrect use of `.find()` with a mutated `dateOverrideExist` flag: short-circuiting on the first truthy return can reject a slot based on the first override and skip other overrides that might allow it"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 2,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.5,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR8330__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8330__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/14943": {
@@ -68583,6 +70422,33 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR14943__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR14943__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [],
+      "false_negatives": [
+        {
+          "golden_comment": "Using retryCount: reminder.retryCount + 1 reads a possibly stale value and can lose increments under concurrency; consider an atomic increment via Prisma (increment: 1) to avoid race conditions (also applies to the similar update in the catch block).",
+          "severity": "High"
+        },
+        {
+          "golden_comment": "The deletion logic in scheduleSMSReminders.ts incorrectly deletes non-SMS workflow reminders (e.g., Email, WhatsApp) that have retryCount > 1. This occurs because the retryCount condition within the OR clause for deletion lacks a method: WorkflowMethods.SMS filter, causing it to apply to all reminder types instead of only SMS reminders, which is the intended scope of this function.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 0,
+      "total_golden": 2,
+      "tp": 0,
+      "fp": 0,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR14943__20260411",
+      "pr_url": null
     }
   },
   "https://github.com/calcom/cal.com/pull/22345": {
@@ -69972,6 +71838,50 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR22345__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR22345__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "In getBaseConditions(), the else if (filterConditions) and final else branches are unreachable. This is because getAuthorizationConditions() always returns a non-null Prisma.Sql object, making authConditions always truthy, which means only the first two if/else if conditions are ever evaluated.",
+          "severity": "Low",
+          "matched_candidate": "getBaseConditions() contains dead/unreachable branches because authConditions is always a truthy Prisma.Sql object (including NOTHING_CONDITION), so logic that assumes authConditions can be falsy is incorrect and prevents short-circuiting when authorization fails.",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same root problem: authConditions is always a non-null/truthy Prisma.Sql (even for a 'nothing' condition), so branches in getBaseConditions() that rely on authConditions being falsy become dead/unreachable and the control flow is incorrect."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Public methods getAuthorizationConditions() and getFilterConditions() return Prisma.Sql fragments containing OR expressions without wrapping the overall expression in outer parentheses, so callers composing them with AND can hit SQL operator-precedence bugs where one OR branch bypasses the AND filter (potential privilege escalation / incorrect filtering)."
+        },
+        {
+          "candidate": "Constructor options type was changed from a discriminated union (requiring teamId when scope='team') to a flat type with optional teamId, removing compile-time enforcement and allowing callers to omit teamId for team scope, leading to runtime zod validation failure and silently returning NOTHING_CONDITION (no data)."
+        },
+        {
+          "candidate": "Filters use falsy guards (e.g., if (this.filters.eventTypeId) / if (this.filters.memberUserId)) which skip applying filters when the value is 0, causing unintended unfiltered results for valid numeric IDs equal to 0."
+        },
+        {
+          "candidate": "Caching tests for getAuthorizationConditions() and getFilterConditions() were removed while caching logic remains, leaving caching behavior untested and allowing regressions that could cause repeated membership-check queries."
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Fetching userIdsFromOrg only when teamsFromOrg.length > 0 can exclude org-level members for orgs without child teams; consider deriving from teamIds (which includes orgId) or removing the guard so org-only orgs still include member user bookings.",
+          "severity": "Medium"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 4,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.2,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR22345__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR22345__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/11059": {
@@ -72507,6 +74417,65 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR11059__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR11059__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Invalid Zod schema syntax. Computed property keys like [z.string().toString()] are not valid in Zod object schemas and will cause runtime errors. ",
+          "severity": "High",
+          "matched_candidate": "minimumTokenResponseSchema incorrectly uses [z.string().toString()] as a computed property key, creating a literal key (e.g., 'ZodString') and failing to validate presence of an expiry field like expires_in/expiry_date, making minimum token validation ineffective",
+          "confidence": 0.9,
+          "reasoning": "The candidate flags the same core problem: using a computed property key derived from z.string().toString() in a Zod object schema, which results in an unintended literal key and breaks schema validation/runtime behavior. It adds extra context about expiry validation, but the underlying invalid Zod schema syntax matches the golden issue."
+        },
+        {
+          "golden_comment": "parseRefreshTokenResponse returns a Zod safeParse result ({ success, data, error }), not the credential key object. Persisting that as key stores the wrapper instead of the token payload; we should store the parsed data or use schema parse.",
+          "severity": "High",
+          "matched_candidate": "parseRefreshTokenResponse returns the full safeParse result ({success,data}) and Google Calendar stores it directly as the credential key, causing the DB to persist a wrapped object instead of the expected token fields and corrupting subsequent credential reads",
+          "confidence": 0.94,
+          "reasoning": "The candidate notes that parseRefreshTokenResponse returns a safeParse wrapper ({success,data}) and that this wrapper is being stored directly as the credential key, leading to persisted wrapped objects instead of token fields\u2014same core issue as the golden comment (should store parsed data or use parse)."
+        },
+        {
+          "golden_comment": "When APP_CREDENTIAL_SHARING_ENABLED and CALCOM_CREDENTIAL_SYNC_ENDPOINT are set, the refreshFunction helper returns the fetch Response, but several callers (for example GoogleCalendarService.refreshAccessToken expecting res.data, and HubspotCalendarService.refreshAccessToken expecting a HubspotToken) assume it returns the integration-specific token object. That mismatch will cause runtime errors in the sync-enabled path unless the return type or those call sites are adjusted.",
+          "severity": "High",
+          "matched_candidate": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response instead of the previously expected parsed/app-specific token object, causing callers (Google Calendar/Hubspot/Zoho) to access missing properties and crash due to inconsistent return types across code paths",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same core bug: in the credential-sharing/sync-enabled path, the refresh helper returns a raw fetch Response rather than an integration-specific token object, while multiple callers (e.g., GoogleCalendarService/HubspotCalendarService) expect parsed token data (res.data / HubspotToken), leading to runtime errors due to inconsistent return types."
+        },
+        {
+          "golden_comment": "When the sync endpoint path is used, res is a fetch Response and has no .data; res?.data will be undefined and token.access_token will throw at runtime. This relies on a consistent return shape from refreshOAuthTokens, which isn\u2019t guaranteed currently.",
+          "severity": "High",
+          "matched_candidate": "When APP_CREDENTIAL_SHARING_ENABLED is true, refreshOAuthTokens returns a raw fetch Response instead of the previously expected parsed/app-specific token object, causing callers (Google Calendar/Hubspot/Zoho) to access missing properties and crash due to inconsistent return types across code paths",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same core problem: in the sync/credential-sharing-enabled path, refreshOAuthTokens returns a raw fetch Response rather than a parsed token object, so callers expecting properties like res.data/token.access_token will see undefined and crash due to inconsistent return shapes across code paths."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Salesforce getClient refreshes and saves a new access token but then instantiates jsforce.Connection with credentialKey.access_token (the old token), so the current connection uses a stale/expired token despite the DB being updated"
+        },
+        {
+          "candidate": "Webhook endpoint does not enforce POST-only requests; non-POST methods (e.g., GET without JSON body) can trigger zod parse errors/500s and violate the intended endpoint semantics"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The parseRefreshTokenResponse function incorrectly sets refresh_token to the hardcoded string 'refresh_token' when it's missing from the OAuth refresh token response. This invalidates the token, breaking subsequent token refreshes and causing authentication failures.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 4,
+      "fp": 2,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.8,
+      "recall": 0.8,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR11059__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR11059__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/7232": {
@@ -74146,6 +76115,49 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR7232__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR7232__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Asynchronous functions deleteScheduledEmailReminder and deleteScheduledSMSReminder are called without await inside forEach loops. This occurs during booking rescheduling/cancellation, and workflow/workflow step deletion/updates. Consequently, scheduled workflow reminders may not be reliably cancelled, potentially leaving them active.",
+          "severity": "Medium",
+          "matched_candidate": "In the reschedule path, async deletions are invoked inside `forEach` without awaiting; errors won\u2019t be caught by the surrounding try/catch and can become unhandled promise rejections",
+          "confidence": 0.86,
+          "reasoning": "Both point out the same core bug: async deletion functions are called inside a forEach without awaiting, so the deletions may not complete reliably and errors may escape try/catch (unhandled rejections). The candidate mentions the reschedule path specifically, which is part of the golden scope, though it doesn\u2019t enumerate all other contexts; still the underlying issue matches."
+        },
+        {
+          "golden_comment": "When immediateDelete is true, the deleteScheduledEmailReminder function cancels the SendGrid email but fails to delete the corresponding WorkflowReminder record from the database. This creates orphaned database entries and is inconsistent with the immediateDelete: false path, which marks the record as cancelled. The SendGrid DELETE API call is also omitted in this path.",
+          "severity": "High",
+          "matched_candidate": "When `immediateDelete` is true and `referenceId` is non-null, the code cancels the SendGrid batch but returns without deleting or marking the `workflowReminder` DB record (e.g., not setting `cancelled=true`), leaving an orphaned reminder that won\u2019t be cleaned up by either the CRON cancel loop or the unscheduled query",
+          "confidence": 0.86,
+          "reasoning": "The candidate flags that when immediateDelete is true the function returns after canceling the SendGrid batch without deleting or marking the WorkflowReminder record as cancelled, leaving an orphaned DB entry. This matches the golden issue about inconsistent handling vs the immediateDelete:false path and orphaned records (even though the candidate doesn\u2019t explicitly mention the missing SendGrid DELETE call, it captures the core DB-orphan/inconsistency bug)."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In the CRON job, `deleteMany` removes all reminders with past `scheduledDate` before processing `cancelled: true` reminders; if a reminder is cancelled shortly before send time and the CRON runs after `scheduledDate`, the DB record is deleted before SendGrid cancellation occurs, so the email is still delivered"
+        },
+        {
+          "candidate": "The CRON cancellation loop is wrapped in a single try/catch; one SendGrid cancel failure aborts the entire loop, leaving remaining cancelled reminders unprocessed (not cancelled in SendGrid and not deleted), so emails may still be sent"
+        },
+        {
+          "candidate": "After a refactor, the sender name/sender ID input section is rendered unconditionally instead of only when `isPhoneNumberNeeded || isSenderIdNeeded`, causing irrelevant sender fields to appear for email-type workflow steps"
+        }
+      ],
+      "false_negatives": [],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 2,
+      "tp": 2,
+      "fp": 3,
+      "fn": 0,
+      "errors_count": 0,
+      "precision": 0.4,
+      "recall": 1.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR7232__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR7232__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/14740": {
@@ -76328,6 +78340,62 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR14740__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR14740__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Case sensitivity bypass in email blacklist",
+          "severity": "High",
+          "matched_candidate": "Blacklist email comparison lowercases `blacklistedGuestEmails` but compares incoming guest emails without lowercasing, allowing case-variant emails to bypass the blacklist",
+          "confidence": 0.96,
+          "reasoning": "Both describe the same issue: the email blacklist check is case-sensitive/inconsistent because only the blacklist entries are lowercased, allowing case-variant incoming emails to bypass the blacklist."
+        },
+        {
+          "golden_comment": "The logic for checking team admin/owner permissions is incorrect. This condition uses AND (&&) which requires both isTeamAdmin AND isTeamOwner to be true, but it should use OR (||) since a user needs to be either an admin OR an owner to have permission.",
+          "severity": "Critical",
+          "matched_candidate": "Authorization logic for `isTeamAdminOrOwner` uses `&&` instead of `||`, requiring users to be both team admin and owner and incorrectly denying admins (making the admin check effectively dead code)",
+          "confidence": 0.98,
+          "reasoning": "The candidate flags the same permission-check bug: using && instead of || for admin/owner authorization, which wrongly requires both roles and denies valid admins/owners."
+        },
+        {
+          "golden_comment": "This calls the email sender with the original guests, so existing attendees included in the input will be treated as new when sending notifications, leading to incorrect emails.",
+          "severity": "Medium",
+          "matched_candidate": "Handler passes raw `guests` input to `sendAddGuestsEmails` instead of filtered `uniqueGuests`, causing existing attendees to be treated as new guests and receive the wrong email template (booking confirmation instead of add-guests notification)",
+          "confidence": 0.93,
+          "reasoning": "Both describe the same bug: the email sender is invoked with the unfiltered/original guests list, so existing attendees included in the input are treated as newly added and receive incorrect notifications/templates."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "When `booking.userId` is null, `booking.userId || 0` leads to querying user id 0 with `findFirstOrThrow`, throwing an unhandled Prisma NotFoundError instead of a proper TRPC error (reachable if organizer account was deleted)"
+        },
+        {
+          "candidate": "`OrganizerAddGuestsEmail.getNodeMailerPayload()` accesses `this.calEvent.attendees[0].name` without guarding for an empty attendees array, risking a runtime TypeError that can crash the whole email batch sent via `Promise.all`"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "uniqueGuests filters out existing attendees and blacklisted emails but does not deduplicate duplicates within the input; createMany can insert duplicate attendee rows if the client submits repeated emails.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Starting with an array containing an empty string may cause validation issues. Consider starting with an empty array [] and handling the empty state in the MultiEmail component instead.",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 3,
+      "fp": 2,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.6,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR14740__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR14740__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/10600": {
@@ -78356,6 +80424,54 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR10600__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR10600__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [],
+      "false_positives": [
+        {
+          "candidate": "Password verification can be bypassed by supplying a non-empty dummy totpCode along with a valid backupCode, causing the password check to be skipped and the backup-code branch to succeed without verifying the password"
+        },
+        {
+          "candidate": "Backup code login is blocked for third-party identity provider (e.g., Google/SAML) users because a guard throws when totpCode is missing, preventing the backup-code path from being reached and making backup codes unusable for these users"
+        },
+        {
+          "candidate": "API route throws an unhandled exception when CALENDSO_ENCRYPTION_KEY is missing, leading Next.js to return an HTML 500 page instead of structured JSON; client code expecting response.json() then fails/silently errors"
+        },
+        {
+          "candidate": "Blob/object URL created via URL.createObjectURL for backup code download is not revoked on dialog close/unmount (resetState lacks cleanup), causing a memory/resource leak per 2FA setup flow"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The exported function TwoFactor handles backup codes and is in BackupCode.tsx. Inconsistent naming.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Error message mentions 'backup code login' but this is a disable endpoint, not login",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "Backup code validation is case-sensitive due to the use of indexOf(). This causes validation to fail if a user enters uppercase hex characters, as backup codes should be case-insensitive for a better user experience.",
+          "severity": "Medium"
+        },
+        {
+          "golden_comment": "Because backupCodes are decrypted and mutated in memory before being written back, two concurrent login requests using the same backupCode could both pass this check and update, so a single backup code may effectively be accepted more than once if used concurrently, weakening the intended one-time-use semantics.",
+          "severity": "High"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 4,
+      "total_golden": 4,
+      "tp": 0,
+      "fp": 4,
+      "fn": 4,
+      "errors_count": 0,
+      "precision": 0.0,
+      "recall": 0.0,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR10600__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10600__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/10967": {
@@ -80617,6 +82733,62 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR10967__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR10967__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "Potential null reference if mainHostDestinationCalendar is undefined if evt.destinationCalendar is null or an empty array ",
+          "severity": "High",
+          "matched_candidate": "In EventManager, destructuring const [mainHostDestinationCalendar] = evt.destinationCalendar ?? [] can yield undefined for null/empty destinationCalendar, but the next line accesses mainHostDestinationCalendar.integration without optional chaining, causing a TypeError",
+          "confidence": 0.94,
+          "reasoning": "Yes. Both describe the same underlying problem: destructuring from evt.destinationCalendar when it is null/empty can produce an undefined mainHostDestinationCalendar, and subsequent property access (e.g., .integration) without guarding/optional chaining can throw a TypeError/null reference."
+        },
+        {
+          "golden_comment": "Logic error: when externalCalendarId is provided, you're searching for a calendar where externalId === externalCalendarId, but this will always fail since you're looking for a calendar that matches itself. Should likely find by credentialId or use different logic.",
+          "severity": "High",
+          "matched_candidate": "In updateEvent/deleteEvent, when externalCalendarId is falsy the code does event.destinationCalendar?.find(cal => cal.externalId === externalCalendarId), which can never match and returns undefined; it should fall back to event.destinationCalendar?.externalId, otherwise updates/deletes use an undefined calendar ID",
+          "confidence": 0.74,
+          "reasoning": "Golden issue: when externalCalendarId is provided, the lookup logic compares externalId to externalCalendarId in a way that will never succeed (wrong selection criteria; should use different identifier/logic). Candidate issue: says when externalCalendarId is falsy, the code still tries to find by externalId === externalCalendarId, which can never match, leading to undefined calendar ID. Both point to the same core bug: the find condition using externalCalendarId is incorrect/guaranteed to fail, causing undefined calendar selection. Candidate differs on the condition (falsy vs provided) but identifies the same flawed lookup."
+        },
+        {
+          "golden_comment": "Logic inversion in organization creation: The slug property is now conditionally set when IS_TEAM_BILLING_ENABLED is true, instead of when it's false as originally intended. This change, combined with requestedSlug still being set when IS_TEAM_BILLING_ENABLED is true, results in both properties being set when billing is enabled, and neither when disabled",
+          "severity": "Medium",
+          "matched_candidate": "In organization creation, the slug/requestedSlug conditional logic is inverted: slug is now set only when billing is enabled (and requestedSlug also only when enabled), resulting in non-billing orgs getting neither slug nor requestedSlug and billing orgs getting both, opposite of intended behavior",
+          "confidence": 0.94,
+          "reasoning": "The candidate describes the same inverted conditional logic around setting slug vs requestedSlug based on IS_TEAM_BILLING_ENABLED, leading to both being set when enabled and neither when disabled, which matches the golden issue."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "In updateAllCalendarEvents, if reference.credentialId is set but the credential is missing from cache and the DB lookup fails, credential stays undefined and is still passed to updateEvent, which will crash when getCalendar(credential) is called"
+        },
+        {
+          "candidate": "In handleNewBooking, evt.destinationCalendar is set to null when neither eventType.destinationCalendar nor organizerUser.destinationCalendar exist, so evt.destinationCalendar?.push(...teamDestinationCalendars) becomes a no-op and team member destination calendars are silently dropped"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "The optional chaining on mainHostDestinationCalendar?.integration is redundant since you already check mainHostDestinationCalendar in the ternary condition.",
+          "severity": "Low"
+        },
+        {
+          "golden_comment": "The Calendar interface now requires createEvent(event, credentialId), but some implementations (e.g., Lark/Office365) still declare createEvent(event) only\u2014this breaks the interface contract (also applies to other locations in the PR).",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 5,
+      "total_golden": 5,
+      "tp": 3,
+      "fp": 2,
+      "fn": 2,
+      "errors_count": 0,
+      "precision": 0.6,
+      "recall": 0.6,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR10967__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR10967__20260411/pull/1"
     }
   },
   "https://github.com/calcom/cal.com/pull/8087": {
@@ -82095,6 +84267,65 @@
       "tool": "macroscope",
       "repo_name": "cal_dot_com__cal.com__macroscope__PR8087__20260309",
       "pr_url": "https://github.com/code-review-benchmark/cal_dot_com__cal.com__macroscope__PR8087__20260309/pull/1"
+    },
+    "tusk": {
+      "skipped": false,
+      "true_positives": [
+        {
+          "golden_comment": "The code uses forEach with async callbacks, which causes asynchronous operations (e.g., calendar/video event deletions, payment refunds) to run concurrently without being awaited. This 'fire-and-forget' behavior leads to unhandled promise rejections, race conditions, and incomplete cleanup, as surrounding try-catch blocks cannot properly handle errors from these unawaited promises. Replace forEach with for...of loops or Promise.all() with map() to ensure proper sequential execution and error handling.",
+          "severity": "Critical",
+          "matched_candidate": "Using an async callback inside Array.forEach in packages/app-store/vital/lib/reschedule.ts causes fire-and-forget behavior because forEach doesn\u2019t await returned promises",
+          "confidence": 0.9,
+          "reasoning": "The candidate flags the same core issue: using an async callback inside Array.forEach leads to promises not being awaited (fire-and-forget). This matches the golden concern about unawaited concurrent async operations and resulting error-handling/race-condition risks, even though it doesn\u2019t mention specific consequences or fixes."
+        }
+      ],
+      "false_positives": [
+        {
+          "candidate": "Using an async callback inside Array.forEach in packages/app-store/wipemycalother/lib/reschedule.ts discards promises, leaving calendar deleteEvent calls unawaited"
+        },
+        {
+          "candidate": "In packages/app-store/wipemycalother/lib/reschedule.ts, errors from async operations inside the forEach won\u2019t be caught and the function proceeds to send reschedule emails immediately"
+        },
+        {
+          "candidate": "In packages/trpc/server/routers/viewer/bookings.tsx, the forEach on line 553 uses an async callback that forEach ignores, so deleteEvent/deleteMeeting operations are unawaited"
+        },
+        {
+          "candidate": "In packages/trpc/server/routers/viewer/bookings.tsx, the code proceeds to send reschedule emails before deletions complete, and errors from deletions are silently swallowed"
+        },
+        {
+          "candidate": "In packages/features/bookings/lib/handleCancelBooking.ts, an async callback inside forEach discards promises, so async work isn\u2019t awaited"
+        },
+        {
+          "candidate": "In packages/features/bookings/lib/handleCancelBooking.ts, apiDeletes is populated inside the unawaited async forEach callback, so Promise.all(apiDeletes) runs with an empty array and skips calendar deletions for recurring events"
+        },
+        {
+          "candidate": "In packages/core/CalendarManager.ts, getCalendarCredentials is synchronous but calls async getCalendar inside flatMap, resulting in returned objects where calendar is a Promise instead of Calendar|null"
+        },
+        {
+          "candidate": "In packages/core/CalendarManager.ts, the function\u2019s type contract is broken because it claims calendar is Calendar|null but actually returns Promise<Calendar|null>"
+        },
+        {
+          "candidate": "In packages/core/CalendarManager.ts, callers that don\u2019t await item.calendar may treat a truthy Promise as a valid calendar and bypass null checks"
+        }
+      ],
+      "false_negatives": [
+        {
+          "golden_comment": "Consider adding try-catch around the await to handle import failures gracefully",
+          "severity": "Low"
+        }
+      ],
+      "errors": [],
+      "total_candidates": 12,
+      "total_golden": 2,
+      "tp": 1,
+      "fp": 9,
+      "fn": 1,
+      "errors_count": 0,
+      "precision": 0.08333333333333333,
+      "recall": 0.5,
+      "tool": "tusk",
+      "repo_name": "cal_dot_com__cal.com__tusk__PR8087__20260411",
+      "pr_url": "https://github.com/tusk-benchmark/cal_dot_com__cal.com__tusk__PR8087__20260411/pull/1"
     }
   }
 }
\ No newline at end of file