Share captured security evidence and harden evaluation boundaries by zrk222 · Pull Request #260 · zrk222/code-factory

zrk222 · 2026-10-28T10:43:52Z

Captured-source facts were repeatedly interpreted across scanner, receipt and secondary-review boundaries. This branch shares that evidence, keeps unknown and stale proofs explicit, and carries bounded retrieval/JEV context without granting native authority.

Security-only scans retain findings when external pytest collection is unsupported; the full oracle profile remains strict. Invalid selectors still raise. UNKNOWN collection cannot grant test-role authority. Persisted request-origin claims now reuse the shared provider mutation analysis, rejecting module and function-local overrides and escapes while retaining supported writer controls.

Final integration repairs are separate commits: 7269fa8c contains behavior, public documentation and boundary controls; 6a5411d6 aligns plugin-collection expectations with the explicit security-only contract. Earlier formatting repair is c014ed61. No new production module, dependency or cap increase.

Validation on the pushed candidate: all17 previously failing hosted test nodes pass locally; expanded storage family28/28; direct retrieval/preflight boundary5/5; other scoped cache, collection and CLI controls pass. CI-pinned Ruff0.14.10 format/lint/C901 pass. Exact hosted candidate-quality assertions pass: gradeA, composite99.1, static test-intent attribution99.9448%, maximum public complexity10. Strict architecture is HEALTHY437Python files/133core modules. Candidate source hashes and workflow observations are sealed locally; committed bytes match the tested candidate. These static/test metrics are not actionable-security precision.

Regular merge preserves history. Final hosted checks are required; no CI skip or admin bypass. Unrelated working-tree edits remain excluded. Scoped Sol review covered the integration repair; a final storage effect review is pending. This stabilization branch makes no accuracy, security-readiness or release claim.

Follow-up integration: hosted6a5411d6 exposed five cases missed by the first focused set. cc73953 removes the overbroad UNKNOWN source-capture veto; existing collection-specific fixture guards and BLOCKED/exit2 semantics remain intact. ee3e865 separately aligns source-availability and invalid-option error assertions. All27 affected controls, including the five failing cases and source-only CLI/receipt/reseal controls, pass locally. Strict architecture, pinned Ruff format/lint and hash-bound workflow audit pass. Final hosted commit remains required.

Final source-review correction:65ed44ee replaces the custom storage provider filter with the existing full provider-stability validator and supplies its hash/AST-validated Path-keyed source map. This closes independently reproduced cross-file request mutation/escape gaps.30storage controls pass, including3supported writer sinks; Sol bounded review confirms P1 resolved, with no remaining P0/P1 in this two-file repair. Format/lint, strict architecture and workflow audit pass. Public production function AST inventory is unchanged from the passing quality snapshot. Regular auto-merge restored; final hosted checks remain required.

Carry validated immutable scalar context through shared constructor analysis; prune only source-proven unreachable branches and retain unknown, rebound and key-escape paths. Include the binding helper in role/effect epochs so replacement or code mutation invalidates cached evidence. Validation: 24 focused guard controls and 2 repaired cache-epoch controls; scoped lint, format, C901, compile, independent ECC review, strict architecture and 7 workflow evidence checks pass. No population precision improvement is claimed.

Discharge duplicate unknown-property attribution only after exact source call binding and a complete callee effect summary. Keep projected field calls, unknown callees, mutated providers and key or owner escapes unresolved; preserve caller-owned heap and dispatch obligations. Validation: 18 focused controls including native CLI and receipt; independent ECC review, lint, format, C901, compile, strict architecture and workflow evidence gate pass. Full application precision remains unmeasured by this repair.

Install the existing bounded invocation memo around foreign AST projection and restore its prior ContextVar in finally. Preserve provider inventory semantics and current helper/source cache identity. Validation: two failing-before memo isolation/recovery controls now pass; same captured AST inventory retains all530 nodes with60 uncached walks reduced to0. Scoped lint, C901, format, compile, architecture and workflow evidence gate pass. No full-repository speed or precision claim.

Track separate source-owned allocations, fields, containers, finite loops, exact call bindings and conservative branch/exception states through the shared use-closure module. Keep opaque recipients, stale sources, unsafe callable-field dispatch and unclosed paths unresolved. Add explicit identity-sensitive provider transfer contracts without executing targets. Validation: focused driver/provider controls pass with scoped static and architecture checks. This is shared machinery only; application-family admission, independent reviewer follow-up and precision goals remain separate.

Follow module credential reads through explicit TestCase invocations, captured signer objects and exact provider consumers. Independently reconstruct the SOURCE_BOUND_TEST_SCOPE_ONLY profile from current source/configuration bytes and model generations. Retain raw CRITICAL material and operational findings; reject opaque exports, descriptor/destructor/header effects and callable shadowing. Validation:27 focused tests across shared flow, contracts, native scope/receipt and provider memo checks; scoped lint, format, compile, C901, strict architecture and local workflow evidence pass. This bounded profile does not yet admit the five actual SimpleJWT cases. Independent agent review is pending the account usage reset; no precision, release or complete-goal claim.

Preserve request-to-sink premises when unrelated captured framework components are registered or configured. Keep parent namespace escapes, exact provider replacement and builtin mutation as vetoes through the existing selected-path conflict matcher. Validation:5 focused native CLI/receipt/reseal controls pass; Ruff lint/format/C901, changed-line SpecLine drift and workflow evidence pass. Architecture remains HEALTHY436Python files/133core modules with unchanged caps. No new module or runtime dependency. Frozen actual-family replay is registered separately; no held-out accuracy claim or independent review claim.

Separate complete source capture from pytest collection authority in the shared role pipeline. Keep collection-independent shell, captured-program and local-worker evidence available; withhold test and fixture projections and preserve raw severity plus UNKNOWN/INCOMPLETE status. Add a frozen-cohort startup gate that rechecks all arm inventories, configuration and execution profiles before launch, rejects stale or failed receipts, and records input failures. Targeted native CLI, receipt and independent reseal controls pass; architecture 437 Python files / 133 core modules, unchanged caps. This does not establish a new precision rate.

Reuse the existing bounded source capture for private configuration evidence across cold/cache-enabled scans, source-bound copies and independent resealing. Configuration content stays out of public JSON; mismatched bytes grant no authority. Validation: exact staged source exported independently; 3 native capture/copy/reseal/drift controls passed in 5.90s. Staged workflow audit PASS (6 checks); architecture HEALTHY 437 files/133 core modules. Pending execution-authority behavior is excluded from this commit.

Reuse the existing shared native admission decision across bounded and full-repository scans. Unresolved-only valid evidence returns a review-required exit zero; supported defects remain exit one and missing, stale, forged or incomplete evidence remains blocked with exit two. Original diagnostics and false clean/approval flags are retained. Validated together with the explicit policy-contract test update in the following commit: 50 unique relevant native, receipt, integrity, CLI and workspace-v2 controls pass; independent Sol ECC review clear; architecture HEALTHY 437 files/133 modules; workflow evidence PASS. This changes admission, not detection or measured accuracy.

Update only the approved complete-unresolved queue expectation from two to zero in the shared formatter and both CLI profiles. Keep supported-defect exit one, malformed/unknown/incomplete evidence exit two, raw HIGH diagnostics, NEEDS_CONTEXT details, receipt bindings and false approval flags asserted. Add native CLI-to-persisted-receipt-to-independent-reseal cases for advisory, supported and unknown-collection states, plus human-output visibility. Targeted results: 26 original integrity controls plus three native cases and 21 consumer controls pass. Only repaired cases were rerun; initial fixture and superseded-contract failures remain in local evidence.

Integrate the one-time Astra medium review of the owner proposal. Reconcile all cases and protected TPs before coding; require two distinct-repository witnesses of the same existing transfer loss; replace the shared path and delete superseded logic; require measured bulk FP gain with TP retention or roll back; only then freeze a fresh independently labelled cohort. No blanket exclusions, engine expansion or admission-only accuracy claim. Validation: report SHA verified; strictdoctor PASS; SpecLine plan gate115tasks PASS; hash-bound workflow audit4checks PASS. Proposal waterfall is170removals, not147; projected gains and sentinel source retention are not verified current results. No production code changed.

Parsed inputs and mutation facts share one cache. Admit immutable facts up to the existing 4096-entry total limit so 2048 parsed entries do not disable fact reuse. Preserve source/helper invalidation and predicate handling. Targeted cache and conservative-fallback checks: 5 passed; ruff/C901 and workflow evidence gate passed. Independent scoped ECC review found no actionable issues. No runtime speed or accuracy claim.

Parse bounded captured source bytes before collection. The failed open-wearables capture requires Python 3.14 and contains eight files rejected by the registered 3.11 runtime; corrected preflight blocks that profile in 4.94 seconds without auditing or executing the target. Thirteen targeted startup checks, ruff/C901, hash-bound workflow gate and independent scoped ECC review passed. This does not resolve the native Python crash or establish a precision gain.

Enforce the existing 80 MiB payload bound before copying source bindings or result rows. Keep the final owned-size and hash/runtime identity gates. The focused cache regression changes oversized deep-copy count from two to zero; ruff/C901 and the workflow evidence gate passed. Independent scoped ECC review found no actionable issues. No whole-scan RAM, CPU, speed or accuracy gain is claimed.

Replace whole-module callable enumeration with the traced request-provider dependency contract, retaining dynamic alias lookup identities and adding signature, scope-node and memo-budget bindings. Reuse one bounded generation token in mutation cache keys. No detector rule is deleted or relaxed. Final affected validation: 86 tests passed; ruff/C901, workflow evidence gate and independent ECC review passed. Offline key microbenchmark improved from 3.46s to 0.0083s for 20,000 hashes; this is not a whole-scan or precision claim.

Keep the full ordered source/helper/code/default identity fingerprint and one retained generation per source index; hot cache keys use its opaque token instead of repeatedly hashing nested dependency tuples. Seven targeted checks passed. Broader affected validation: 188 passed and one identical failure reproduced on frozen baseline 7c78274; no expectations changed. Ruff/C901, independent ECC review and scoped workflow evidence passed. Offline 10,000-key hashing measured 0.504s to 0.0024s; no whole-scan speed or accuracy claim.

Correct the stale expectation that mode=False writes inside if mode is None. Preserve five immutable-default binding cases and six explicit scalar guard cases, including keyword-only and positional-only None defaults. Production scanner behavior is unchanged. Final affected validation: 195 passed, zero failures; lint and workflow evidence passed. Independent review identified the need to retain the two parameter-kind guard cases, which are now explicit.

Fuse existing BM25 with bounded local dense cosine using reciprocal rank fusion. Reuse capture-owned embeddings across queries and pass similarity hints into secondary review without granting trace or actionability authority. Validation: 44 affected checks plus existing BM25 regression pass; lint/C901 pass. Cached MiniLM offline synthetic packet delivery passes. No actionable precision gain claimed. Strict architecture remains blocked by existing review_integrity.py 5044/5000 lines; independent ECC review unavailable due account usage limit.

Record real full-capture Blogin retrieval miss, assignment-only transport gap, and no-exhaustiveness/no-authority constraints. Preserve strict architecture prerequisite and rejected relocation evidence. Implementation remains incomplete.

Remove unused and single-use helper indirection, reuse oracle-body handling, share upload recursion, and consolidate repeated imports and ordered dependency names. No module relocation, rule addition or cap increase. Strict architecture HEALTHY: 437 Python files, 133 core modules; review_integrity.py 5000 lines. Upload/oracle/provider and callback targeted controls pass; lint/C901 and hash-bound workflow audit pass. No precision gain claimed.

Preserve explicit per-finding native retrieval status through CLI, persisted receipts and independent reseal. Carry call and assignment source evidence into secondary review with query, finding, corpus and hint identities; reject altered or dropped hint transport before dispatch. Similarity remains unvalidated context and cannot clear findings. Validation: 48 affected Python packet checks, 25 native tier/admission checks, 8 final tamper/receipt checks, 10 JS tests and 2 Python-to-JS source readback cases pass. Lint/C901, strict architecture HEALTHY and hash-bound workflow audit pass. Live JEV and actionable precision not measured; gateway key absent.

Reuse existing import, callee and argument binding machinery for secondary evidence handoff. Preserve mutated and opaque provider unknowns, source dependencies and existing packet limits. Reuse import inventory within each retriever; bound one-hop expansion without adding detector rules or dependencies. Validation: 50 affected checks, offline Python-to-JS source readback, lint/C901 and workflow evidence gate pass; architecture HEALTHY. Real-repository target-span replay remains unchanged, so no precision or provider-quality gain is claimed.

Apply ECC review: late callee/source-budget truncation must mark context INCOMPLETE. Independently reject contradictory CAPTURED or missing checkpoint status at the receiving JEV boundary. Native findings and unresolved premises remain unchanged. Validation: 52 affected Python checks, 10 JavaScript evidence checks and 2 cross-language source readbacks pass; architecture HEALTHY, lint/C901 and hash-bound workflow gate pass. No new precision claim.

Replace unbounded historical contract generations with one current identity/token and strongly-owned binding tuple. Preserve stable-token reuse and invalidate changed or restored contracts without retaining prior models. Validation: 14 affected source-bound test-scope checks pass, including 1000-generation stress and native CLI receipt/recompute. Ruff/C901, strict architecture and workflow evidence gate pass. No precision improvement claimed.

Dispatch explicitly requested partial evidence reviews while preserving source checks, unknowns, budgets and null final actionability. Default admission remains unchanged. Validate 77 affected Node checks, independent scoped review, healthy architecture and workflow evidence. Live paired diagnostic reached all four provider requests; all remained insufficient evidence.

Version secondary rubric to 1.2.7. Require provider and authority evidence relevant to each claim, preserving incomplete outcomes, positive contradiction for false and all native gates. 77 affected Node checks pass with updated version expectation; architecture and workflow audit pass. Paired live diagnostics remain unresolved and establish no native precision gain.

Reuse source-generation observations without secondary scanner calls. Preserve exact source binding, unknown actionability and reported-not-verified authority boundary. 54 affected checks pass, with focused producer identity retest; architecture and workflow checks pass. No native precision or real-repository speed gain claimed. Independent review unavailable due usage limit.

Carry pytest.toml and .pytest.toml through the shared configuration capture, sealing and independent reconstruction inventory. Extend existing parser-to-verifier regressions to reject stale bytes and observe identity changes. No collection-ownership or release authority is inferred.

Update the explicit missing and oversized configuration inventories for the two added selectors. Keep this expectation footprint change separate from the capture behavior repair. Final scoped validation: 21 passed; lint, format, strict architecture and workflow audit pass.

Preserve known configured pytest settings and source hashes in static_pytest_collection before manifest sealing. Keep receipt attestation separate. Verify actual selector changes, cache parity and independent nested report ownership. This transport establishes no role demotion or measured accuracy claim.

Existing source-role tests expected exit 2 but unchanged HEAD returned 1 for UNKNOWN collection. Preserve all active findings, source hashes and INCOMPLETE state while returning BLOCKED. Existing expectations are unchanged. Combined scoped checks: 20 passed; trusted-core recomputation, lint, format, architecture and workflow audit pass.

In security-only mode, unsupported pytest collection semantics become UNKNOWN with empty patterns, INCOMPLETE state and blocked policy exit. Retain security findings and deny test-role credit. Full oracle mode remains strict; invalid parse, unreadable and oversized inputs still reject. Three new cases failed before and pass after; 21 affected checks, final reason controls, independent review and workflow audit pass. Original Logfire selector reproduced with an inert one-file control; no full-repository or accuracy claim.

Record fixed source exclusions, source/receipt preflight, explicit sampling scope and preserved startup corrections. No actionable accuracy or publication claim.

coderabbitai · 2026-10-28T10:44:03Z

Important

🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

Source mode: Auto
Repository: Unavailable
Branch: Unavailable
File: No configuration file from this repository applied
Commit: Unavailable
Layers: Organization UI
Review profile: CHILL
Plan: Advanced
Run ID: 1818b008-01e8-4e6f-848d-221f4938586e

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

github-actions · 2026-10-28T10:45:40Z

FactoryLine Proof Review

Commit: 65ed44ee05e176bf86a712c8b0d336a8cfb1a887
Diff-to-Proof Review SHA-256: 434d46291ea17dc16cf7bfb231a687cb39391192df17e898279cda34e2202f96
Proof Review SHA-256: dae8a56766bae91b3a15ec20ae71b56906f4db3cc46a264f1df3ffe06efe1acf

Changed-scope walkthrough

Contracts — specs/shared-security-evidence.md
Docs — docs/evidence/owner-publication-supersession-20261027.json
Implementation — factoryline/audit_action_refs.py, factoryline/audit_rule_search.py, factoryline/context_efficiency.py, factoryline/credential_identity_audit.py, factoryline/credential_role_constructor_provider.py, factoryline/credential_role_identity_effects.py, factoryline/credential_role_method_use_getter_audit.py, factoryline/data/judge_framework.json, factoryline/finite_capture.py, factoryline/import_context.py, factoryline/jev_classifier.py, factoryline/primitive_observation_binding.py, factoryline/pytest_collection_audit.py, factoryline/review_audits.py, factoryline/review_integrity.py, factoryline/semantic_credential_sensor.py, scripts/evaluate_jev.mjs, scripts/evaluate_jev.test.mjs, scripts/evaluate_jev_evidence.test.mjs, scripts/jev_evidence.mjs, scripts/precision_cohort_preflight.py
Other — CHANGELOG.md, plans/release-record-reconciliation-20261027.md, plans/shared-security-evidence.md
Tests — tests/test_audit_rule_search.py, tests/test_cli_bootstrap.py, tests/test_context_efficiency.py, tests/test_credential_role_identity_fields.py, tests/test_deep_audit.py, tests/test_finite_capture_origins.py, tests/test_mock_fixture_provenance.py, tests/test_review_audits.py, tests/test_runtime_audit_integrity.py, tests/test_runtime_audit_observations.py

Fact-derived next action

bind_changed_path_to_proof — This changed path has no explicit Graph Ops proof-input edge.

Findings

blocking unmatched_changed_path — This changed path has no explicit Graph Ops proof-input edge.

Unproven claims

No explicit proof-input edge is declared for CHANGELOG.md.
No explicit proof-input edge is declared for docs/evidence/owner-publication-supersession-20261027.json.
No explicit proof-input edge is declared for factoryline/audit_action_refs.py.
No explicit proof-input edge is declared for factoryline/audit_rule_search.py.
No explicit proof-input edge is declared for factoryline/context_efficiency.py.
No explicit proof-input edge is declared for factoryline/credential_identity_audit.py.
No explicit proof-input edge is declared for factoryline/credential_role_constructor_provider.py.
No explicit proof-input edge is declared for factoryline/credential_role_identity_effects.py.
No explicit proof-input edge is declared for factoryline/credential_role_method_use_getter_audit.py.
No explicit proof-input edge is declared for factoryline/data/judge_framework.json.
No explicit proof-input edge is declared for factoryline/finite_capture.py.
No explicit proof-input edge is declared for factoryline/import_context.py.
No explicit proof-input edge is declared for factoryline/jev_classifier.py.
No explicit proof-input edge is declared for factoryline/primitive_observation_binding.py.
No explicit proof-input edge is declared for factoryline/pytest_collection_audit.py.
No explicit proof-input edge is declared for factoryline/review_audits.py.
No explicit proof-input edge is declared for factoryline/review_integrity.py.
No explicit proof-input edge is declared for factoryline/semantic_credential_sensor.py.
No explicit proof-input edge is declared for plans/release-record-reconciliation-20261027.md.
No explicit proof-input edge is declared for plans/shared-security-evidence.md.
No explicit proof-input edge is declared for scripts/evaluate_jev.mjs.
No explicit proof-input edge is declared for scripts/evaluate_jev.test.mjs.
No explicit proof-input edge is declared for scripts/evaluate_jev_evidence.test.mjs.
No explicit proof-input edge is declared for scripts/jev_evidence.mjs.
No explicit proof-input edge is declared for scripts/precision_cohort_preflight.py.
No explicit proof-input edge is declared for specs/shared-security-evidence.md.
No explicit proof-input edge is declared for tests/test_audit_rule_search.py.
No explicit proof-input edge is declared for tests/test_cli_bootstrap.py.
No explicit proof-input edge is declared for tests/test_context_efficiency.py.
No explicit proof-input edge is declared for tests/test_credential_role_identity_fields.py.
No explicit proof-input edge is declared for tests/test_deep_audit.py.
No explicit proof-input edge is declared for tests/test_finite_capture_origins.py.
No explicit proof-input edge is declared for tests/test_mock_fixture_provenance.py.
No explicit proof-input edge is declared for tests/test_review_audits.py.
No explicit proof-input edge is declared for tests/test_runtime_audit_integrity.py.
No explicit proof-input edge is declared for tests/test_runtime_audit_observations.py.
Pattern and guard-path audit state: not_configured; no runtime correctness or release approval is implied.

Authority boundary

Advisory only. This payload has no execution, approval, publication, deployment, signing, messaging, credential, connector, source write, test execution, repair authority. It does not use CodeRabbit credentials, interpret AI comments as proof, approve, merge, or modify source.

Existing Diff-to-Proof map

flowchart LR
  REVIEW["Diff-to-Proof Review"]
  C1["Changed: CHANGELOG.md"]
  C1 --> REVIEW
  C2["Changed: docs/evidence/owner-publication-supersession-20261027.json"]
  C2 --> REVIEW
  C3["Changed: factoryline/audit_action_refs.py"]
  C3 --> REVIEW
  C4["Changed: factoryline/audit_rule_search.py"]
  C4 --> REVIEW
  C5["Changed: factoryline/context_efficiency.py"]
  C5 --> REVIEW
  C6["Changed: factoryline/credential_identity_audit.py"]
  C6 --> REVIEW
  C7["Changed: factoryline/credential_role_constructor_provider.py"]
  C7 --> REVIEW
  C8["Changed: factoryline/credential_role_identity_effects.py"]
  C8 --> REVIEW
  C9["Changed: factoryline/credential_role_method_use_getter_audit.py"]
  C9 --> REVIEW
  C10["Changed: factoryline/data/judge_framework.json"]
  C10 --> REVIEW
  C11["Changed: factoryline/finite_capture.py"]
  C11 --> REVIEW
  C12["Changed: factoryline/import_context.py"]
  C12 --> REVIEW
  C13["Changed: factoryline/jev_classifier.py"]
  C13 --> REVIEW
  C14["Changed: factoryline/primitive_observation_binding.py"]
  C14 --> REVIEW
  C15["Changed: factoryline/pytest_collection_audit.py"]
  C15 --> REVIEW
  C16["Changed: factoryline/review_audits.py"]
  C16 --> REVIEW
  C17["Changed: factoryline/review_integrity.py"]
  C17 --> REVIEW
  C18["Changed: factoryline/semantic_credential_sensor.py"]
  C18 --> REVIEW
  C19["Changed: plans/release-record-reconciliation-20261027.md"]
  C19 --> REVIEW
  C20["Changed: plans/shared-security-evidence.md"]
  C20 --> REVIEW
  C21["Changed: scripts/evaluate_jev.mjs"]
  C21 --> REVIEW
  C22["Changed: scripts/evaluate_jev.test.mjs"]
  C22 --> REVIEW
  C23["Changed: scripts/evaluate_jev_evidence.test.mjs"]
  C23 --> REVIEW
  C24["Changed: scripts/jev_evidence.mjs"]
  C24 --> REVIEW
  C25["Changed: scripts/precision_cohort_preflight.py"]
  C25 --> REVIEW
  C26["Changed: specs/shared-security-evidence.md"]
  C26 --> REVIEW
  C27["Changed: tests/test_audit_rule_search.py"]
  C27 --> REVIEW
  C28["Changed: tests/test_cli_bootstrap.py"]
  C28 --> REVIEW
  C29["Changed: tests/test_context_efficiency.py"]
  C29 --> REVIEW
  C30["Changed: tests/test_credential_role_identity_fields.py"]
  C30 --> REVIEW
  C31["Changed: tests/test_deep_audit.py"]
  C31 --> REVIEW
  C32["Changed: tests/test_finite_capture_origins.py"]
  C32 --> REVIEW
  C33["Changed: tests/test_mock_fixture_provenance.py"]
  C33 --> REVIEW
  C34["Changed: tests/test_review_audits.py"]
  C34 --> REVIEW
  C35["Changed: tests/test_runtime_audit_integrity.py"]
  C35 --> REVIEW
  C36["Changed: tests/test_runtime_audit_observations.py"]
  C36 --> REVIEW
  U1["Unmatched: CHANGELOG.md"]
  REVIEW --> U1
  U2["Unmatched: docs/evidence/owner-publication-supersession-20261027.json"]
  REVIEW --> U2
  U3["Unmatched: factoryline/audit_action_refs.py"]
  REVIEW --> U3
  U4["Unmatched: factoryline/audit_rule_search.py"]
  REVIEW --> U4
  U5["Unmatched: factoryline/context_efficiency.py"]
  REVIEW --> U5
  U6["Unmatched: factoryline/credential_identity_audit.py"]
  REVIEW --> U6
  U7["Unmatched: factoryline/credential_role_constructor_provider.py"]
  REVIEW --> U7
  U8["Unmatched: factoryline/credential_role_identity_effects.py"]
  REVIEW --> U8
  U9["Unmatched: factoryline/credential_role_method_use_getter_audit.py"]
  REVIEW --> U9
  U10["Unmatched: factoryline/data/judge_framework.json"]
  REVIEW --> U10
  U11["Unmatched: factoryline/finite_capture.py"]
  REVIEW --> U11
  U12["Unmatched: factoryline/import_context.py"]
  REVIEW --> U12
  U13["Unmatched: factoryline/jev_classifier.py"]
  REVIEW --> U13
  U14["Unmatched: factoryline/primitive_observation_binding.py"]
  REVIEW --> U14
  U15["Unmatched: factoryline/pytest_collection_audit.py"]
  REVIEW --> U15
  U16["Unmatched: factoryline/review_audits.py"]
  REVIEW --> U16
  U17["Unmatched: factoryline/review_integrity.py"]
  REVIEW --> U17
  U18["Unmatched: factoryline/semantic_credential_sensor.py"]
  REVIEW --> U18
  U19["Unmatched: plans/release-record-reconciliation-20261027.md"]
  REVIEW --> U19
  U20["Unmatched: plans/shared-security-evidence.md"]
  REVIEW --> U20
  U21["Unmatched: scripts/evaluate_jev.mjs"]
  REVIEW --> U21
  U22["Unmatched: scripts/evaluate_jev.test.mjs"]
  REVIEW --> U22
  U23["Unmatched: scripts/evaluate_jev_evidence.test.mjs"]
  REVIEW --> U23
  U24["Unmatched: scripts/jev_evidence.mjs"]
  REVIEW --> U24
  U25["Unmatched: scripts/precision_cohort_preflight.py"]
  REVIEW --> U25
  U26["Unmatched: specs/shared-security-evidence.md"]
  REVIEW --> U26
  U27["Unmatched: tests/test_audit_rule_search.py"]
  REVIEW --> U27
  U28["Unmatched: tests/test_cli_bootstrap.py"]
  REVIEW --> U28
  U29["Unmatched: tests/test_context_efficiency.py"]
  REVIEW --> U29
  U30["Unmatched: tests/test_credential_role_identity_fields.py"]
  REVIEW --> U30
  U31["Unmatched: tests/test_deep_audit.py"]
  REVIEW --> U31
  U32["Unmatched: tests/test_finite_capture_origins.py"]
  REVIEW --> U32
  U33["Unmatched: tests/test_mock_fixture_provenance.py"]
  REVIEW --> U33
  U34["Unmatched: tests/test_review_audits.py"]
  REVIEW --> U34
  U35["Unmatched: tests/test_runtime_audit_integrity.py"]
  REVIEW --> U35
  U36["Unmatched: tests/test_runtime_audit_observations.py"]
  REVIEW --> U36
  F1["unmatched_changed_path"]
  REVIEW --> F1

Fix only the six files named by hosted Ruff-format failure. All six Python ASTs match the prior head; full lint and C901 pass. Preserve unrelated working-tree edits. No behavioral or accuracy claim.

Reject invalid pytest selector modes while retaining unsupported security-only collection as UNKNOWN without test-role authority. Reuse captured provider effect analysis before persisted request-origin claims. Preserve live dispatch bindings, document public methods and simplify existing control flow without new modules or cap increases. Validation: exact17 previously failing hosted tests pass; expanded storage family28/28; retrieval/preflight boundary5/5; scoped cache and CLI controls pass. CI-pinned Ruff format/lint/C901, strict architecture437/133 and exact candidate quality A99.1/99.9448 attribution/max10 pass. Hash-bound workflow audit PASS. Hosted checks still required. Plugin expectation alignment follows separately.

…y mode Unsupported external pytest collection remains UNKNOWN only in the explicit security-only profile, with empty patterns and no oracle authority. Strict default mode still raises; malformed import-mode values continue to raise in both modes. Keep expectation alignment separate from scanner behavior repairs. Also directly exercise sealed preflight path containment and bundle tampering boundaries.

Remove the overbroad early source-capture veto. Existing role-specific collection guards still reject fixture/test authority and the policy remains BLOCKED with exit2. Source-owned shell/file/worker context can independently survive native receipt resealing. Validation:27 focused cases including the five latest hosted failures, strict architecture, Ruff and hash-bound workflow audit pass.

Assert validated source availability while retaining UNKNOWN/INCOMPLETE blocked policy and fixture-authority rejection. Malformed non-string pytest options must raise the exact invalid configuration error; unsupported plugin configuration remains a separate rejected category. Scanner behavior and expectation alignment remain separate commits.

Replace custom filtered mutation indexing with the existing shared provider-stability validator. Bind the existing Path-keyed capture map only after every snapshot validates, so foreign request escapes and overrides cannot gain persisted input authority. Reuse source resources and remove redundant filtering without a new helper or module. Validation:30source-bound storage controls pass including cross-file escape/override and3positive writer sinks; independent bounded Sol review confirms P1 resolved. Pinned format/lint, strict architecture and hash-bound workflow audit pass; public production AST inventory unchanged from passing quality head. No accuracy claim.

zrk222 added 30 commits October 27, 2026 07:08

fix: separate source metadata from pytest collection authority

455421d

fix: preserve source-bound constructor parameter literals

7571cc5

fix: retain security scans with unmodeled builtin pytest collection

a8765d3

fix: reject ancestor replacements in shared provider identity checks

978382c

docs: freeze next accuracy cohort before blind evaluation

18f21ca

docs: reconcile v0.49.0 publication record

77d5935

zrk222 added 10 commits October 28, 2026 03:26

test: expect judge framework criterion version 1.2.7

8cc9c30

plan: record frozen current-baseline evaluation boundaries

9bab45c

Record fixed source exclusions, source/receipt preflight, explicit sampling scope and preserved startup corrections. No actionable accuracy or publication claim.

docs: preserve existing plan text encoding

7ab04a9

zrk222 enabled auto-merge (squash) October 28, 2026 10:44

zrk222 disabled auto-merge October 28, 2026 10:49

zrk222 enabled auto-merge October 28, 2026 10:49

zrk222 added 6 commits October 28, 2026 17:49

Merge branch 'main' into codex/precision-study-20261027

b642ffb

style: apply CI-pinned formatting to accumulated repairs

c014ed6

Fix only the six files named by hosted Ruff-format failure. All six Python ASTs match the prior head; full lint and C901 pass. Preserve unrelated working-tree edits. No behavioral or accuracy claim.

zrk222 disabled auto-merge October 28, 2026 12:39

zrk222 enabled auto-merge October 28, 2026 12:45

zrk222 merged commit acd8256 into main Oct 28, 2026
68 of 69 checks passed

Navigation Menu

Repository navigation

Share captured security evidence and harden evaluation boundaries - #260

Share captured security evidence and harden evaluation boundaries#260
zrk222 merged 48 commits into
mainfrom
codex/precision-study-20261027

zrk222 commented Oct 28, 2026 •

edited

Loading

Uh oh!

coderabbitai Bot commented Oct 28, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Oct 28, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

zrk222 commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coderabbitai Bot commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

FactoryLine Proof Review

Changed-scope walkthrough

Fact-derived next action

Findings

Unproven claims

Authority boundary

Existing Diff-to-Proof map

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

zrk222 commented Oct 28, 2026 •

edited

Loading

coderabbitai Bot commented Oct 28, 2026 •

edited

Loading

github-actions Bot commented Oct 28, 2026 •

edited

Loading