diff --git a/bundles/analysis-tools.json b/bundles/analysis-tools.json index 12f25cf..5f9060c 100644 --- a/bundles/analysis-tools.json +++ b/bundles/analysis-tools.json @@ -30,6 +30,7 @@ "./skills/policyengine-writing", "./skills/policyengine-research-lookup", "./skills/policyengine-prior-scores", + "./skills/policyengine-comparables-review", "./skills/policyengine-calibration-diagnostics" ], "commands": [ @@ -41,7 +42,8 @@ "./commands/prior-analysis.md", "./commands/text-to-reform.md", "./commands/create-dashboard.md", - "./commands/deploy-dashboard.md" + "./commands/deploy-dashboard.md", + "./commands/comparables-review.md" ], "agents": [ "./agents/calibration-diagnostics.md", diff --git a/bundles/app-development.json b/bundles/app-development.json index 4c2bc3d..5f586f7 100644 --- a/bundles/app-development.json +++ b/bundles/app-development.json @@ -29,6 +29,7 @@ "./skills/policyengine-standards", "./skills/policyengine-cloud-costs", "./skills/policyengine-writing", + "./skills/policyengine-comparables-review", "./skills/fix-pr" ], "commands": [ @@ -37,7 +38,8 @@ "./commands/new-tool.md", "./commands/audit-seo.md", "./commands/audit-multizone.md", - "./commands/write-tests.md" + "./commands/write-tests.md", + "./commands/comparables-review.md" ], "agents": [ "./agents/app/component-test-writer.md", diff --git a/bundles/complete.json b/bundles/complete.json index 16a0619..946ddc3 100644 --- a/bundles/complete.json +++ b/bundles/complete.json @@ -33,6 +33,7 @@ "./skills/policyengine-user-guide", "./skills/policyengine-research-lookup", "./skills/policyengine-prior-scores", + "./skills/policyengine-comparables-review", "./skills/policyengine-calibration-diagnostics", "./skills/policyengine-content", "./skills/policyengine-github-agent", @@ -47,6 +48,7 @@ "./commands/audit-seo.md", "./commands/audit-state-tax.md", "./commands/backdate-program.md", + "./commands/comparables-review.md", "./commands/create-dashboard.md", "./commands/create-new-component.md", "./commands/create-pr.md", diff --git a/bundles/dashboard-builder.json b/bundles/dashboard-builder.json index 10cd985..f93743e 100644 --- a/bundles/dashboard-builder.json +++ b/bundles/dashboard-builder.json @@ -28,11 +28,13 @@ "./skills/policyengine-design", "./skills/policyengine-standards", "./skills/policyengine-cloud-costs", - "./skills/policyengine-writing" + "./skills/policyengine-writing", + "./skills/policyengine-comparables-review" ], "commands": [ "./commands/create-dashboard.md", - "./commands/deploy-dashboard.md" + "./commands/deploy-dashboard.md", + "./commands/comparables-review.md" ], "agents": [ "./agents/dashboard/dashboard-planner.md", diff --git a/dashboard/src/data/manifest.json b/dashboard/src/data/manifest.json index a468fb4..66bad46 100644 --- a/dashboard/src/data/manifest.json +++ b/dashboard/src/data/manifest.json @@ -1,9 +1,9 @@ { - "generated_at": "2026-11-13T14:31:18.877622Z", + "generated_at": "2026-11-14T06:04:44.366477Z", "counts": { - "skill": 25, + "skill": 26, "agent": 36, - "command": 20, + "command": 21, "bundle": 8, "analysis": 6 }, @@ -563,6 +563,59 @@ "references": {}, "body_length": 19305 }, + { + "id": "policyengine-comparables-review", + "kind": "skill", + "name": "policyengine-comparables-review", + "description": "The comparables review every PolicyEngine analysis passes before its dashboard or research post\nis listed on policyengine.org, and again before a newsletter, social campaign or press push\nfeatures it: search every producer on the scorecard's list for external estimates, stage each\nin PolicyEngine/policyengine-scorecard, compare it with the analysis's own figures, and explain\nevery gap over the threshold. CI in the scorecard, policyengine-app-v2 and the newsletters repo\nchecks parts of this procedure. Includes a brief for running the search as a background job.\nTriggers: \"comparables\", \"comparables review\", \"pre-publish review\", \"before publishing\",\n\"add to apps.json\", \"add to posts.json\", \"scorecard record\", \"external estimates\",\n\"comparables gate\", \"analysis-uk-\", \"analysis-us-\", \"re-harvest\", \"newsletter\",\n\"before a push\", \"promote an analysis\".\nNOT for: benchmarking a reform mid-analysis without publishing (policyengine-prior-scores,\n/prior-scores).", + "category": "process", + "path": "skills/policyengine-comparables-review/SKILL.md", + "tools": [], + "model": null, + "triggers": [ + "comparables", + "comparables review", + "pre-publish review", + "before publishing", + "add to apps.json", + "add to posts.json", + "scorecard record", + "external estimates", + "comparables gate", + "analysis-uk-", + "analysis-us-", + "re-harvest", + "newsletter", + "before a push", + "promote an analysis" + ], + "bundles": [ + "analysis-tools", + "app-development", + "complete", + "dashboard-builder" + ], + "target_repos": [ + "policyengine-app-v2", + "policyengine-app", + "newsletters", + "policyengine-us", + "policyengine-canada" + ], + "functional_role": null, + "functional_summary": null, + "functional_scope": [], + "functional_supersedes": [], + "registry_status": "recommended", + "registry_owner": "Frontend & product", + "recommended_for": [ + "Load this skill for work involving: The comparables review every PolicyEngine analysis passes before its dashboard or research post\nis listed on policyengine.org, and again before a newsletter, social campaign or press push\nfeatures it: search every producer on the scorecard's list for external estimates, stage each\nin PolicyEngine/policyengine-scorecard, compare it with the analysis's own figures, and explain\nevery gap over the threshold. CI in the scorecard, policyengine-app-v2 and the newsletters repo\nchecks parts of this procedure. Includes a brief for running the search as a background job.\nTriggers: \"comparables\", \"comparables review\", \"pre-publish review\", \"before publishing\",\n\"add to apps.json\", \"add to posts.json\", \"scorecard record\", \"external estimates\",\n\"comparables gate\", \"analysis-uk-\", \"analysis-us-\", \"re-harvest\", \"newsletter\",\n\"before a push\", \"promote an analysis\".\nNOT for: benchmarking a reform mid-analysis without publishing (policyengine-prior-scores,\n/prior-scores)." + ], + "use_instead": [], + "registry_notes": "Recommended for the listed scope.", + "references": {}, + "body_length": 25188 + }, { "id": "policyengine-content", "kind": "skill", @@ -585,7 +638,9 @@ "bundles": [ "complete" ], - "target_repos": [], + "target_repos": [ + "policyengine-us" + ], "functional_role": "process:content", "functional_summary": "Social assets for posts: image template, token usage, generation workflow.", "functional_scope": [ @@ -602,7 +657,7 @@ "use_instead": [], "registry_notes": "Recommended for the listed scope.", "references": {}, - "body_length": 3884 + "body_length": 5344 }, { "id": "policyengine-data", @@ -930,7 +985,7 @@ "use_instead": [], "registry_notes": "Recommended for the listed scope.", "references": {}, - "body_length": 8075 + "body_length": 8248 }, { "id": "policyengine-research-lookup", @@ -1080,7 +1135,7 @@ "use_instead": [], "registry_notes": "Recommended for the listed scope.", "references": {}, - "body_length": 19807 + "body_length": 20148 }, { "id": "policyengine-uk", @@ -1267,7 +1322,7 @@ "use_instead": [], "registry_notes": "Recommended for the listed scope.", "references": {}, - "body_length": 28693 + "body_length": 29270 }, { "id": "review-program", @@ -1473,7 +1528,8 @@ "complete" ], "target_repos": [ - "policyengine-us" + "policyengine-us", + "policyengine-app" ], "functional_role": "orchestrate:content", "functional_summary": "Generates marketing assets from blog posts \u2014 localized social images and LinkedIn/X post copy.", @@ -1488,8 +1544,15 @@ ], "use_instead": [], "registry_notes": "Agents are usually implementation details of commands; call directly only when maintaining or debugging the workflow.", - "references": {}, - "body_length": 2724 + "references": { + "skill": [ + "policyengine-comparables-review" + ], + "command": [ + "comparables-review" + ] + }, + "body_length": 4674 }, { "id": "ci-fixer", @@ -3154,15 +3217,17 @@ "skill": [ "policyengine-app", "policyengine-canada", + "policyengine-comparables-review", "policyengine-uk", "policyengine-us", "policyengine-writing" ], "command": [ - "analyze-policy" + "analyze-policy", + "comparables-review" ] }, - "body_length": 17168 + "body_length": 17439 }, { "id": "neutrality-reviewer", @@ -3448,6 +3513,52 @@ }, "body_length": 52434 }, + { + "id": "comparables-review", + "kind": "command", + "name": "comparables-review", + "description": "Run the pre-publish comparables review for a dashboard, research post or tracker score and record it in the scorecard", + "category": "command", + "path": "targets/claude/commands/comparables-review.md", + "tools": [], + "model": null, + "triggers": [], + "bundles": [ + "analysis-tools", + "app-development", + "complete", + "dashboard-builder" + ], + "target_repos": [ + "policyengine-us", + "policyengine-uk", + "policyengine-canada", + "policyengine-app" + ], + "functional_role": null, + "functional_summary": null, + "functional_scope": [], + "functional_supersedes": [], + "registry_status": "recommended", + "registry_owner": "Country model engineering", + "recommended_for": [ + "Use this command when you need to run the pre-publish comparables review for a dashboard, research post or tracker score and record it in the scorecard" + ], + "use_instead": [], + "registry_notes": "Recommended for the listed scope.", + "references": { + "skill": [ + "policyengine-comparables-review" + ], + "command": [ + "comparables-review", + "deploy-dashboard", + "prior-scores", + "reform-pipeline" + ] + }, + "body_length": 4107 + }, { "id": "create-dashboard", "kind": "command", @@ -3502,12 +3613,13 @@ ], "command": [ "analyze-policy", + "comparables-review", "create-dashboard", "deploy-dashboard", "new-tool" ] }, - "body_length": 22374 + "body_length": 22467 }, { "id": "create-new-component", @@ -3648,13 +3760,15 @@ "policyengine-api", "policyengine-app", "policyengine-cloud-costs", + "policyengine-comparables-review", "policyengine-tools" ], "command": [ + "comparables-review", "deploy-dashboard" ] }, - "body_length": 12765 + "body_length": 15547 }, { "id": "encode-reform", @@ -3744,14 +3858,16 @@ "neutrality-reviewer" ], "skill": [ + "policyengine-comparables-review", "policyengine-content", "policyengine-writing" ], "command": [ + "comparables-review", "generate-content" ] }, - "body_length": 1608 + "body_length": 3424 }, { "id": "household-calc", @@ -3888,15 +4004,17 @@ "policyengine-api-access", "policyengine-app", "policyengine-cloud-costs", + "policyengine-comparables-review", "policyengine-design", "policyengine-tools" ], "command": [ + "comparables-review", "create-dashboard", "new-tool" ] }, - "body_length": 17056 + "body_length": 19531 }, { "id": "prior-analysis", @@ -4018,10 +4136,12 @@ "report-logger" ], "skill": [ + "policyengine-comparables-review", "policyengine-uk" ], "command": [ "analyze-policy", + "comparables-review", "create-dashboard", "deploy-dashboard", "implement-structural", @@ -4029,7 +4149,7 @@ "reform-pipeline" ] }, - "body_length": 9938 + "body_length": 12550 }, { "id": "setup-verbs", @@ -4180,7 +4300,7 @@ "use_instead": [], "registry_notes": "Bundle install profile.", "references": {}, - "body_length": 2175 + "body_length": 2263 }, { "id": "api-development", @@ -4232,7 +4352,7 @@ "use_instead": [], "registry_notes": "Bundle install profile.", "references": {}, - "body_length": 1501 + "body_length": 1589 }, { "id": "complete", @@ -4258,7 +4378,7 @@ "use_instead": [], "registry_notes": "Bundle install profile.", "references": {}, - "body_length": 3770 + "body_length": 3858 }, { "id": "country-models", @@ -4310,7 +4430,7 @@ "use_instead": [], "registry_notes": "Bundle install profile.", "references": {}, - "body_length": 1366 + "body_length": 1454 }, { "id": "data-science", @@ -4654,6 +4774,36 @@ "target_kind": "command", "target": "backdate-program" }, + { + "source_kind": "command", + "source": "comparables-review", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, + { + "source_kind": "command", + "source": "comparables-review", + "target_kind": "command", + "target": "comparables-review" + }, + { + "source_kind": "command", + "source": "comparables-review", + "target_kind": "command", + "target": "deploy-dashboard" + }, + { + "source_kind": "command", + "source": "comparables-review", + "target_kind": "command", + "target": "prior-scores" + }, + { + "source_kind": "command", + "source": "comparables-review", + "target_kind": "command", + "target": "reform-pipeline" + }, { "source_kind": "command", "source": "create-dashboard", @@ -4726,6 +4876,12 @@ "target_kind": "command", "target": "analyze-policy" }, + { + "source_kind": "command", + "source": "create-dashboard", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "command", "source": "create-dashboard", @@ -4840,12 +4996,24 @@ "target_kind": "skill", "target": "policyengine-cloud-costs" }, + { + "source_kind": "command", + "source": "deploy-dashboard", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "command", "source": "deploy-dashboard", "target_kind": "skill", "target": "policyengine-tools" }, + { + "source_kind": "command", + "source": "deploy-dashboard", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "command", "source": "deploy-dashboard", @@ -4924,6 +5092,12 @@ "target_kind": "agent", "target": "neutrality-reviewer" }, + { + "source_kind": "command", + "source": "generate-content", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "command", "source": "generate-content", @@ -4936,6 +5110,12 @@ "target_kind": "skill", "target": "policyengine-writing" }, + { + "source_kind": "command", + "source": "generate-content", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "command", "source": "generate-content", @@ -5056,6 +5236,12 @@ "target_kind": "skill", "target": "policyengine-cloud-costs" }, + { + "source_kind": "command", + "source": "new-tool", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "command", "source": "new-tool", @@ -5068,6 +5254,12 @@ "target_kind": "skill", "target": "policyengine-tools" }, + { + "source_kind": "command", + "source": "new-tool", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "command", "source": "new-tool", @@ -5116,6 +5308,12 @@ "target_kind": "agent", "target": "report-logger" }, + { + "source_kind": "command", + "source": "reform-pipeline", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "command", "source": "reform-pipeline", @@ -5128,6 +5326,12 @@ "target_kind": "command", "target": "analyze-policy" }, + { + "source_kind": "command", + "source": "reform-pipeline", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "command", "source": "reform-pipeline", @@ -5308,6 +5512,12 @@ "target_kind": "skill", "target": "policyengine-app" }, + { + "source_kind": "agent", + "source": "content-orchestrator", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "agent", "source": "dashboard-architecture-validator", @@ -6010,6 +6220,12 @@ "target_kind": "skill", "target": "policyengine-canada" }, + { + "source_kind": "agent", + "source": "report-logger", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "agent", "source": "report-logger", @@ -6220,6 +6436,12 @@ "target_kind": "skill", "target": "policyengine-canada" }, + { + "source_kind": "bundle", + "source": "analysis-tools", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "bundle", "source": "analysis-tools", @@ -6274,6 +6496,12 @@ "target_kind": "command", "target": "analyze-policy" }, + { + "source_kind": "bundle", + "source": "analysis-tools", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "bundle", "source": "analysis-tools", @@ -6490,6 +6718,12 @@ "target_kind": "skill", "target": "policyengine-cloud-costs" }, + { + "source_kind": "bundle", + "source": "app-development", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "bundle", "source": "app-development", @@ -6544,6 +6778,12 @@ "target_kind": "command", "target": "audit-seo" }, + { + "source_kind": "bundle", + "source": "app-development", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "bundle", "source": "app-development", @@ -6838,6 +7078,12 @@ "target_kind": "skill", "target": "policyengine-cloud-costs" }, + { + "source_kind": "bundle", + "source": "complete", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "bundle", "source": "complete", @@ -6964,6 +7210,12 @@ "target_kind": "command", "target": "backdate-program" }, + { + "source_kind": "bundle", + "source": "complete", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "bundle", "source": "complete", @@ -7294,6 +7546,12 @@ "target_kind": "skill", "target": "policyengine-cloud-costs" }, + { + "source_kind": "bundle", + "source": "dashboard-builder", + "target_kind": "skill", + "target": "policyengine-comparables-review" + }, { "source_kind": "bundle", "source": "dashboard-builder", @@ -7330,6 +7588,12 @@ "target_kind": "skill", "target": "policyengine-writing" }, + { + "source_kind": "bundle", + "source": "dashboard-builder", + "target_kind": "command", + "target": "comparables-review" + }, { "source_kind": "bundle", "source": "dashboard-builder", @@ -7431,7 +7695,7 @@ "kind": "skill", "id": "fix-pr" }, - "score": 0.5013, + "score": 0.4939, "shared_terms": [ "review", "claude", @@ -7452,13 +7716,13 @@ "kind": "skill", "id": "policyengine-plugin-maintenance" }, - "score": 0.1867, + "score": 0.1884, "shared_terms": [ "claude", "codex", "routing", - "catalog", "skills", + "catalog", "tests", "source", "test" @@ -7473,7 +7737,7 @@ "kind": "skill", "id": "review-program" }, - "score": 0.5881, + "score": 0.5826, "shared_terms": [ "review", "canonical", @@ -7494,12 +7758,12 @@ "kind": "agent", "id": "ci-fixer" }, - "score": 0.2156, + "score": 0.2143, "shared_terms": [ - "review", "contract", - "worktree", "test", + "worktree", + "review", "tests", "fix", "run", @@ -7515,7 +7779,7 @@ "kind": "agent", "id": "edge-case-generator" }, - "score": 0.1705, + "score": 0.1674, "shared_terms": [ "tests", "review", @@ -7536,7 +7800,7 @@ "kind": "agent", "id": "implementation-validator" }, - "score": 0.2537, + "score": 0.2475, "shared_terms": [ "review", "fix", @@ -7545,7 +7809,7 @@ "read", "structural", "mode", - "worktree" + "scope" ] }, { @@ -7557,7 +7821,7 @@ "kind": "agent", "id": "model-worker" }, - "score": 0.3507, + "score": 0.3525, "shared_terms": [ "encode", "coordinator", @@ -7578,7 +7842,7 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.2276, + "score": 0.2199, "shared_terms": [ "review", "push", @@ -7586,8 +7850,8 @@ "tests", "encode", "draft", - "commit", - "approved" + "approved", + "commit" ] }, { @@ -7599,10 +7863,10 @@ "kind": "agent", "id": "program-reviewer" }, - "score": 0.2141, + "score": 0.2108, "shared_terms": [ - "review", "program", + "review", "tests", "code", "implementation", @@ -7620,12 +7884,12 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.3604, + "score": 0.3509, "shared_terms": [ "review", - "canonical", - "delegation", "program", + "delegation", + "canonical", "coordinator", "source", "completion", @@ -7641,7 +7905,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.1809, + "score": 0.1824, "shared_terms": [ "tests", "test", @@ -7662,13 +7926,13 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.212, + "score": 0.2077, "shared_terms": [ "program", "review", "worktree", - "arguments", "read", + "arguments", "research", "pdf", "dpi" @@ -7683,7 +7947,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.1659, + "score": 0.1671, "shared_terms": [ "tests", "test", @@ -7704,7 +7968,7 @@ "kind": "skill", "id": "policyengine-plugin-maintenance" }, - "score": 0.1811, + "score": 0.1844, "shared_terms": [ "claude", "codex", @@ -7725,7 +7989,7 @@ "kind": "skill", "id": "review-program" }, - "score": 0.5668, + "score": 0.5615, "shared_terms": [ "review", "canonical", @@ -7746,16 +8010,16 @@ "kind": "agent", "id": "ci-fixer" }, - "score": 0.2222, + "score": 0.2197, "shared_terms": [ "fix", - "failures", "files", + "failures", "workflow", "review", - "worktree", "root", - "fixes" + "fixes", + "worktree" ] }, { @@ -7767,7 +8031,7 @@ "kind": "agent", "id": "implementation-validator" }, - "score": 0.3385, + "score": 0.3358, "shared_terms": [ "fix", "review", @@ -7775,8 +8039,8 @@ "mode", "role", "code", - "only", - "verification" + "verification", + "only" ] }, { @@ -7788,11 +8052,11 @@ "kind": "agent", "id": "model-worker" }, - "score": 0.1589, + "score": 0.1593, "shared_terms": [ "role", - "canonical", "delegation", + "canonical", "workflow", "assigned", "read", @@ -7809,12 +8073,12 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.1825, + "score": 0.176, "shared_terms": [ "fix", "review", - "push", "workflow", + "push", "pusher", "files", "local", @@ -7830,7 +8094,7 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.2978, + "score": 0.2871, "shared_terms": [ "review", "canonical", @@ -7838,8 +8102,8 @@ "delegation", "workflow", "assigned", - "read", - "code" + "code", + "read" ] }, { @@ -7851,7 +8115,7 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.1893, + "score": 0.1841, "shared_terms": [ "review", "files", @@ -7872,12 +8136,12 @@ "kind": "skill", "id": "policyengine-api" }, - "score": 0.2236, + "score": 0.2182, "shared_terms": [ "household", "api", - "country", "rest", + "country", "runs", "server", "python", @@ -7893,15 +8157,15 @@ "kind": "skill", "id": "policyengine-canada" }, - "score": 0.1725, + "score": 0.1678, "shared_terms": [ "household", - "microsimulation", "tax", + "microsimulation", "income", - "analysis", "credit", "full", + "analysis", "poverty" ] }, @@ -7914,16 +8178,16 @@ "kind": "skill", "id": "policyengine-model-development" }, - "score": 0.1729, + "score": 0.1698, "shared_terms": [ - "household", "model", - "analysis", + "household", "microsimulation", + "analysis", "variables", + "parameters", "reform", - "run", - "parameters" + "run" ] }, { @@ -7935,7 +8199,7 @@ "kind": "skill", "id": "policyengine-uk" }, - "score": 0.4089, + "score": 0.4048, "shared_terms": [ "household", "unit", @@ -7956,7 +8220,7 @@ "kind": "skill", "id": "policyengine-us" }, - "score": 0.3338, + "score": 0.3325, "shared_terms": [ "unit", "tax", @@ -7977,7 +8241,7 @@ "kind": "skill", "id": "policyengine-user-guide" }, - "score": 0.2867, + "score": 0.2852, "shared_terms": [ "household", "impact", @@ -7985,8 +8249,8 @@ "poverty", "income", "tax", - "inequality", - "losers" + "population", + "inequality" ] }, { @@ -7998,16 +8262,16 @@ "kind": "agent", "id": "calibration-diagnostics" }, - "score": 0.1521, + "score": 0.1529, "shared_terms": [ "state", - "microcosm", "tax", "unit", + "microcosm", "dataset", "income", "poverty", - "run" + "result" ] }, { @@ -8019,7 +8283,7 @@ "kind": "agent", "id": "dashboard-integrator" }, - "score": 0.1673, + "score": 0.1636, "shared_terms": [ "household", "api", @@ -8040,7 +8304,7 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.234, + "score": 0.235, "shared_terms": [ "dataset", "runs", @@ -8061,14 +8325,14 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1905, + "score": 0.192, "shared_terms": [ "tax", "income", "state", "model", - "reform", "ctc", + "reform", "parameter", "rate" ] @@ -8082,7 +8346,7 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.157, + "score": 0.1547, "shared_terms": [ "household", "scenarios", @@ -8103,7 +8367,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.2604, + "score": 0.2534, "shared_terms": [ "household", "microsimulation", @@ -8124,7 +8388,7 @@ "kind": "skill", "id": "policyengine-api-access" }, - "score": 0.5715, + "score": 0.5736, "shared_terms": [ "api", "client", @@ -8145,7 +8409,7 @@ "kind": "skill", "id": "policyengine-app" }, - "score": 0.162, + "score": 0.1631, "shared_terms": [ "api", "server", @@ -8166,15 +8430,15 @@ "kind": "agent", "id": "backend-builder" }, - "score": 0.3242, + "score": 0.3247, "shared_terms": [ "api", - "public", "client", + "public", "dashboard", "economy", "data", - "country", + "simulation", "access" ] }, @@ -8187,16 +8451,16 @@ "kind": "agent", "id": "dashboard-integrator" }, - "score": 0.3658, + "score": 0.3673, "shared_terms": [ "api", - "household", "client", + "household", "data", "error", "year", - "dashboard", - "request" + "request", + "dashboard" ] }, { @@ -8208,7 +8472,7 @@ "kind": "agent", "id": "dashboard-planner" }, - "score": 0.1782, + "score": 0.1778, "shared_terms": [ "api", "dashboard", @@ -8217,7 +8481,7 @@ "rest", "credential", "endpoint", - "household" + "impact" ] }, { @@ -8229,7 +8493,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.2028, + "score": 0.2039, "shared_terms": [ "api", "client", @@ -8238,7 +8502,7 @@ "org", "data", "household", - "public" + "cloud" ] }, { @@ -8250,11 +8514,11 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.3584, + "score": 0.3585, "shared_terms": [ "api", - "country", "economy", + "country", "region", "baseline", "org", @@ -8271,7 +8535,7 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.2041, + "score": 0.2034, "shared_terms": [ "api", "window", @@ -8292,7 +8556,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1697, + "score": 0.1627, "shared_terms": [ "household", "api", @@ -8313,15 +8577,15 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2186, + "score": 0.2152, "shared_terms": [ "api", "window", "public", "dashboard", "client", - "org", "calc", + "org", "country" ] }, @@ -8334,7 +8598,7 @@ "kind": "skill", "id": "policyengine-app" }, - "score": 0.2341, + "score": 0.2361, "shared_terms": [ "api", "app", @@ -8355,12 +8619,12 @@ "kind": "skill", "id": "policyengine-tools" }, - "score": 0.2207, + "score": 0.2169, "shared_terms": [ "api", - "public", "org", "dashboard", + "public", "backend", "app", "vercel", @@ -8376,12 +8640,12 @@ "kind": "agent", "id": "backend-builder" }, - "score": 0.4853, + "score": 0.4835, "shared_terms": [ "api", "dashboard", - "public", "backend", + "public", "client", "modal", "pattern", @@ -8397,7 +8661,7 @@ "kind": "agent", "id": "dashboard-integrator" }, - "score": 0.423, + "score": 0.43, "shared_terms": [ "api", "client", @@ -8418,7 +8682,7 @@ "kind": "agent", "id": "dashboard-planner" }, - "score": 0.2457, + "score": 0.2417, "shared_terms": [ "dashboard", "api", @@ -8439,15 +8703,15 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.3391, + "score": 0.3383, "shared_terms": [ "api", "dashboard", "client", "backend", "public", - "page", "org", + "page", "pattern" ] }, @@ -8460,16 +8724,16 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.1945, + "score": 0.197, "shared_terms": [ "api", "org", - "country", "runs", - "local", + "country", "economy", + "local", "call", - "run" + "url" ] }, { @@ -8481,11 +8745,11 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.2084, + "score": 0.2014, "shared_terms": [ "dashboard", - "page", "api", + "page", "household", "backend", "pages", @@ -8502,16 +8766,16 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.1784, + "score": 0.1839, "shared_terms": [ "dashboard", "modal", "app", "api", - "public", "backend", "authentication", - "token" + "vercel", + "public" ] }, { @@ -8523,7 +8787,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2965, + "score": 0.2914, "shared_terms": [ "api", "dashboard", @@ -8544,7 +8808,7 @@ "kind": "skill", "id": "policyengine-design" }, - "score": 0.1664, + "score": 0.1669, "shared_terms": [ "app", "tailwindcss", @@ -8565,7 +8829,7 @@ "kind": "skill", "id": "policyengine-tools" }, - "score": 0.474, + "score": 0.4706, "shared_terms": [ "app", "next", @@ -8586,7 +8850,7 @@ "kind": "agent", "id": "multizone-validator" }, - "score": 0.2266, + "score": 0.2265, "shared_terms": [ "app", "host", @@ -8607,7 +8871,7 @@ "kind": "agent", "id": "dashboard-architecture-validator" }, - "score": 0.283, + "score": 0.2801, "shared_terms": [ "app", "router", @@ -8628,7 +8892,7 @@ "kind": "agent", "id": "dashboard-design-validator" }, - "score": 0.1631, + "score": 0.1615, "shared_terms": [ "app", "tsx", @@ -8649,12 +8913,12 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.1628, + "score": 0.1639, "shared_terms": [ "api", "app", - "page", "tsx", + "page", "design", "next", "router", @@ -8670,7 +8934,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.2613, + "score": 0.2603, "shared_terms": [ "app", "kit", @@ -8691,15 +8955,15 @@ "kind": "command", "id": "audit-multizone" }, - "score": 0.241, + "score": 0.2397, "shared_terms": [ "app", "host", "next", "multizone", "website", - "rewrites", "zone", + "rewrites", "vercel" ] }, @@ -8712,12 +8976,12 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.2526, + "score": 0.2502, "shared_terms": [ "app", "component", - "src", "design", + "src", "kit", "tsx", "react", @@ -8733,15 +8997,15 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.2336, + "score": 0.2469, "shared_terms": [ "app", "host", - "rewrites", "website", - "workspace", + "rewrites", "vercel", "bun", + "workspace", "next" ] }, @@ -8754,11 +9018,11 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2835, + "score": 0.2794, "shared_terms": [ "app", - "calculator", "api", + "calculator", "bun", "page", "kit", @@ -8775,7 +9039,7 @@ "kind": "skill", "id": "policyengine-data" }, - "score": 0.318, + "score": 0.3147, "shared_terms": [ "microcosm", "calibration", @@ -8796,7 +9060,7 @@ "kind": "agent", "id": "calibration-diagnostics" }, - "score": 0.5191, + "score": 0.5185, "shared_terms": [ "calibration", "microcosm", @@ -8817,7 +9081,7 @@ "kind": "agent", "id": "model-corroborator" }, - "score": 0.2097, + "score": 0.2083, "shared_terms": [ "calibration", "reform", @@ -8838,7 +9102,7 @@ "kind": "skill", "id": "policyengine-uk" }, - "score": 0.2784, + "score": 0.2749, "shared_terms": [ "household", "canada", @@ -8859,7 +9123,7 @@ "kind": "skill", "id": "policyengine-us" }, - "score": 0.1808, + "score": 0.178, "shared_terms": [ "tax", "household", @@ -8880,7 +9144,7 @@ "kind": "skill", "id": "policyengine-user-guide" }, - "score": 0.1637, + "score": 0.1605, "shared_terms": [ "household", "income", @@ -8901,13 +9165,13 @@ "kind": "command", "id": "household-calc" }, - "score": 0.2317, + "score": 0.2267, "shared_terms": [ "household", "canada", "benefit", - "level", "income", + "level", "population", "net", "microsimulation" @@ -8922,13 +9186,13 @@ "kind": "agent", "id": "backend-builder" }, - "score": 0.1988, + "score": 0.1987, "shared_terms": [ "modal", "backend", "containers", - "dashboard", "api", + "dashboard", "plan", "cloud", "reason" @@ -8943,7 +9207,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.235, + "score": 0.2349, "shared_terms": [ "modal", "deploy", @@ -8952,7 +9216,112 @@ "run", "backend", "cloud", - "merge" + "shows" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-comparables-review" + }, + "b": { + "kind": "skill", + "id": "policyengine-content" + }, + "score": 0.2554, + "shared_terms": [ + "scorecard", + "analysis", + "comparables", + "post", + "push", + "social", + "review", + "promote" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-comparables-review" + }, + "b": { + "kind": "skill", + "id": "policyengine-research-lookup" + }, + "score": 0.1572, + "shared_terms": [ + "posts", + "json", + "analysis", + "post", + "research", + "app", + "search", + "filename" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-comparables-review" + }, + "b": { + "kind": "agent", + "id": "content-orchestrator" + }, + "score": 0.3101, + "shared_terms": [ + "analysis", + "scorecard", + "comparables", + "push", + "review", + "social", + "posts", + "record" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-comparables-review" + }, + "b": { + "kind": "command", + "id": "comparables-review" + }, + "score": 0.6049, + "shared_terms": [ + "scorecard", + "comparables", + "analysis", + "review", + "push", + "newsletter", + "search", + "gap" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-comparables-review" + }, + "b": { + "kind": "command", + "id": "generate-content" + }, + "score": 0.3458, + "shared_terms": [ + "scorecard", + "analysis", + "comparables", + "social", + "review", + "push", + "search", + "posts" ] }, { @@ -8964,16 +9333,16 @@ "kind": "skill", "id": "policyengine-design" }, - "score": 0.2879, + "score": 0.2569, "shared_terms": [ "theme", "teal", "css", + "var", "color", - "kit", + "primary", "chart", - "var", - "dark" + "kit" ] }, { @@ -8985,16 +9354,16 @@ "kind": "skill", "id": "policyengine-research-lookup" }, - "score": 0.1556, + "score": 0.1646, "shared_terms": [ "post", "blog", - "image", "published", + "date", + "image", + "analysis", "app", - "tone", - "live", - "research" + "search" ] }, { @@ -9006,16 +9375,16 @@ "kind": "skill", "id": "policyengine-writing" }, - "score": 0.2052, + "score": 0.2041, "shared_terms": [ "social", "post", "blog", "copy", + "analysis", + "not", "writing", - "neutrality", - "chart", - "make" + "content" ] }, { @@ -9027,16 +9396,16 @@ "kind": "agent", "id": "content-orchestrator" }, - "score": 0.5381, + "score": 0.5825, "shared_terms": [ "social", "content", "copy", - "chrome", - "images", - "logo", - "image", - "headless" + "push", + "analysis", + "scorecard", + "blog", + "announcement" ] }, { @@ -9045,19 +9414,19 @@ "id": "policyengine-content" }, "b": { - "kind": "agent", - "id": "frontend-builder" + "kind": "command", + "id": "comparables-review" }, - "score": 0.1823, + "score": 0.3141, "shared_terms": [ - "kit", - "teal", - "design", - "chart", - "css", - "system", - "app", - "import" + "post", + "scorecard", + "comparables", + "push", + "social", + "analysis", + "review", + "published" ] }, { @@ -9069,7 +9438,7 @@ "kind": "command", "id": "generate-content" }, - "score": 0.6299, + "score": 0.6985, "shared_terms": [ "social", "content", @@ -9077,50 +9446,8 @@ "post", "announcement", "image", - "copy", - "blog" - ] - }, - { - "a": { - "kind": "skill", - "id": "policyengine-content" - }, - "b": { - "kind": "command", - "id": "new-tool" - }, - "score": 0.1512, - "shared_terms": [ - "kit", - "css", - "theme", - "var", - "import", - "color", - "teal", - "copy" - ] - }, - { - "a": { - "kind": "skill", - "id": "policyengine-data" - }, - "b": { - "kind": "skill", - "id": "policyengine-uk" - }, - "score": 0.1513, - "shared_terms": [ - "microcosm", - "data", - "entity", - "person", - "unit", - "household", - "see", - "population" + "analysis", + "scorecard" ] }, { @@ -9132,7 +9459,7 @@ "kind": "agent", "id": "calibration-diagnostics" }, - "score": 0.2797, + "score": 0.2752, "shared_terms": [ "microcosm", "calibration", @@ -9153,15 +9480,15 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.1746, + "score": 0.1729, "shared_terms": [ "microcosm", "dataset", - "local", "data", + "local", "populace", - "default", "runs", + "default", "acs" ] }, @@ -9174,7 +9501,7 @@ "kind": "agent", "id": "report-logger" }, - "score": 0.1572, + "score": 0.1556, "shared_terms": [ "microcosm", "data", @@ -9195,15 +9522,15 @@ "kind": "skill", "id": "policyengine-tools" }, - "score": 0.3036, + "score": 0.3042, "shared_terms": [ "css", "theme", "postcss", "kit", "tailwindcss", - "app", "tailwind", + "app", "chart" ] }, @@ -9216,13 +9543,13 @@ "kind": "agent", "id": "dashboard-architecture-validator" }, - "score": 0.2715, + "score": 0.2714, "shared_terms": [ "css", "postcss", "tailwindcss", - "app", "theme", + "app", "kit", "tailwind", "import" @@ -9237,13 +9564,13 @@ "kind": "agent", "id": "dashboard-design-validator" }, - "score": 0.1556, + "score": 0.1558, "shared_terms": [ "kit", "css", "design", - "app", "font", + "app", "var", "colors", "node" @@ -9258,7 +9585,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.3831, + "score": 0.3848, "shared_terms": [ "kit", "chart", @@ -9279,13 +9606,13 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.2094, + "score": 0.21, "shared_terms": [ "kit", "design", "css", - "app", "theme", + "app", "component", "colors", "tailwind" @@ -9300,7 +9627,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2846, + "score": 0.2886, "shared_terms": [ "css", "theme", @@ -9321,10 +9648,10 @@ "kind": "skill", "id": "policyengine-uk" }, - "score": 0.2272, + "score": 0.2252, "shared_terms": [ - "household", "income", + "household", "tax", "unit", "net", @@ -9342,7 +9669,7 @@ "kind": "skill", "id": "policyengine-us" }, - "score": 0.3185, + "score": 0.3175, "shared_terms": [ "unit", "tax", @@ -9350,8 +9677,8 @@ "medicaid", "household", "aca", - "net", - "person" + "person", + "net" ] }, { @@ -9363,7 +9690,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1751, + "score": 0.1708, "shared_terms": [ "household", "income", @@ -9384,15 +9711,15 @@ "kind": "skill", "id": "policyengine-uk" }, - "score": 0.1566, + "score": 0.1545, "shared_terms": [ "household", "entity", "benunit", "model", "age", - "income", "person", + "income", "variables" ] }, @@ -9405,11 +9732,11 @@ "kind": "skill", "id": "policyengine-us" }, - "score": 0.1513, + "score": 0.1504, "shared_terms": [ "entity", - "household", "year", + "household", "gov", "period", "person", @@ -9426,7 +9753,7 @@ "kind": "agent", "id": "edge-case-generator" }, - "score": 0.1956, + "score": 0.1959, "shared_terms": [ "tests", "test", @@ -9435,7 +9762,7 @@ "year", "parameters", "periods", - "add" + "formula" ] }, { @@ -9447,7 +9774,7 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.1515, + "score": 0.1511, "shared_terms": [ "changelog", "tests", @@ -9468,7 +9795,7 @@ "kind": "agent", "id": "structural-implementer" }, - "score": 0.1513, + "score": 0.1517, "shared_terms": [ "tests", "parameter", @@ -9489,7 +9816,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.3082, + "score": 0.3095, "shared_terms": [ "tests", "variable", @@ -9510,7 +9837,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1987, + "score": 0.1994, "shared_terms": [ "gov", "parameter", @@ -9531,7 +9858,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.1846, + "score": 0.1843, "shared_terms": [ "formula", "contrib", @@ -9539,8 +9866,8 @@ "bracket", "parameters", "reform", - "run", - "model" + "model", + "run" ] }, { @@ -9552,7 +9879,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.1762, + "score": 0.1764, "shared_terms": [ "tests", "test", @@ -9561,7 +9888,7 @@ "run", "repo", "path", - "references" + "conventions" ] }, { @@ -9573,7 +9900,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.1966, + "score": 0.1968, "shared_terms": [ "test", "tests", @@ -9594,13 +9921,13 @@ "kind": "agent", "id": "model-corroborator" }, - "score": 0.2606, + "score": 0.2567, "shared_terms": [ "reform", "external", "scores", - "published", "scored", + "published", "benchmark", "shape", "stage" @@ -9615,10 +9942,10 @@ "kind": "agent", "id": "prior-scores-finder" }, - "score": 0.479, + "score": 0.474, "shared_terms": [ - "prior", "tier", + "prior", "scores", "scripts", "scored", @@ -9636,10 +9963,10 @@ "kind": "agent", "id": "reform-comparator" }, - "score": 0.2534, + "score": 0.247, "shared_terms": [ - "prior", "anchor", + "prior", "scores", "microsim", "reform", @@ -9657,7 +9984,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1819, + "score": 0.1816, "shared_terms": [ "reform", "preset", @@ -9678,7 +10005,7 @@ "kind": "command", "id": "prior-analysis" }, - "score": 0.2492, + "score": 0.2456, "shared_terms": [ "prior", "scripts", @@ -9687,7 +10014,7 @@ "scores", "analyses", "scored", - "score" + "find" ] }, { @@ -9699,15 +10026,15 @@ "kind": "command", "id": "prior-scores" }, - "score": 0.5996, + "score": 0.5968, "shared_terms": [ "scorekeepers", - "scores", "tier", + "scores", "prior", "presets", "scored", - "published", + "benchmark", "reform" ] }, @@ -9720,7 +10047,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2506, + "score": 0.2502, "shared_terms": [ "reform", "preset", @@ -9738,19 +10065,61 @@ "id": "policyengine-research-lookup" }, "b": { - "kind": "skill", - "id": "policyengine-writing" + "kind": "skill", + "id": "policyengine-writing" + }, + "score": 0.2193, + "shared_terms": [ + "posts", + "blog", + "post", + "research", + "writing", + "about", + "published", + "descriptions" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-research-lookup" + }, + "b": { + "kind": "agent", + "id": "content-orchestrator" + }, + "score": 0.18, + "shared_terms": [ + "posts", + "blog", + "date", + "analysis", + "published", + "post", + "title", + "written" + ] + }, + { + "a": { + "kind": "skill", + "id": "policyengine-research-lookup" + }, + "b": { + "kind": "agent", + "id": "report-logger" }, - "score": 0.2231, + "score": 0.1911, "shared_terms": [ "posts", - "blog", + "articles", + "app", "post", "research", - "writing", - "published", - "about", - "descriptions" + "src", + "data", + "has" ] }, { @@ -9759,19 +10128,19 @@ "id": "policyengine-research-lookup" }, "b": { - "kind": "agent", - "id": "report-logger" + "kind": "command", + "id": "comparables-review" }, - "score": 0.1967, + "score": 0.1616, "shared_terms": [ + "post", "posts", - "articles", + "published", + "date", "app", - "post", "research", - "src", - "has", - "data" + "analysis", + "json" ] }, { @@ -9783,16 +10152,16 @@ "kind": "command", "id": "generate-content" }, - "score": 0.1502, + "score": 0.1741, "shared_terms": [ "posts", "blog", "post", + "date", + "analysis", "image", - "research", - "title", - "png", - "tone" + "published", + "search" ] }, { @@ -9804,13 +10173,13 @@ "kind": "agent", "id": "issue-manager" }, - "score": 0.1628, + "score": 0.1612, "shared_terms": [ "repo", "number", "create", - "repository", "draft", + "repository", "issue", "title", "never" @@ -9825,7 +10194,7 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.2886, + "score": 0.2901, "shared_terms": [ "changelog", "repo", @@ -9846,7 +10215,7 @@ "kind": "agent", "id": "structural-implementer" }, - "score": 0.1544, + "score": 0.1537, "shared_terms": [ "draft", "changelog", @@ -9867,7 +10236,7 @@ "kind": "command", "id": "create-pr" }, - "score": 0.4055, + "score": 0.4032, "shared_terms": [ "echo", "wait", @@ -9888,7 +10257,7 @@ "kind": "agent", "id": "multizone-validator" }, - "score": 0.4052, + "score": 0.4058, "shared_terms": [ "zone", "basepath", @@ -9909,7 +10278,7 @@ "kind": "agent", "id": "backend-builder" }, - "score": 0.1748, + "score": 0.1733, "shared_terms": [ "modal", "backend", @@ -9930,10 +10299,10 @@ "kind": "agent", "id": "dashboard-architecture-validator" }, - "score": 0.4273, + "score": 0.4272, "shared_terms": [ - "app", "postcss", + "app", "css", "config", "kit", @@ -9951,7 +10320,7 @@ "kind": "agent", "id": "dashboard-design-validator" }, - "score": 0.1878, + "score": 0.1861, "shared_terms": [ "kit", "app", @@ -9960,7 +10329,7 @@ "tsx", "design", "dashboard", - "required" + "config" ] }, { @@ -9972,7 +10341,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.2573, + "score": 0.2566, "shared_terms": [ "dashboard", "api", @@ -9993,7 +10362,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.2785, + "score": 0.277, "shared_terms": [ "kit", "app", @@ -10001,8 +10370,8 @@ "css", "design", "next", - "chart", - "import" + "import", + "chart" ] }, { @@ -10014,7 +10383,7 @@ "kind": "command", "id": "audit-multizone" }, - "score": 0.3411, + "score": 0.342, "shared_terms": [ "zone", "next", @@ -10035,7 +10404,7 @@ "kind": "command", "id": "audit-seo" }, - "score": 0.192, + "score": 0.1917, "shared_terms": [ "vercel", "seo", @@ -10056,7 +10425,7 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.2145, + "score": 0.2135, "shared_terms": [ "kit", "app", @@ -10077,12 +10446,12 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.3244, + "score": 0.3264, "shared_terms": [ "modal", "host", - "app", "vercel", + "app", "dashboard", "config", "next", @@ -10098,7 +10467,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.3504, + "score": 0.3472, "shared_terms": [ "kit", "css", @@ -10107,7 +10476,7 @@ "next", "bun", "api", - "calculator" + "dashboard" ] }, { @@ -10119,14 +10488,14 @@ "kind": "skill", "id": "policyengine-us" }, - "score": 0.5313, + "score": 0.5305, "shared_terms": [ "unit", "tax", "household", "income", - "poverty", "person", + "poverty", "credit", "entity" ] @@ -10140,7 +10509,7 @@ "kind": "skill", "id": "policyengine-user-guide" }, - "score": 0.2785, + "score": 0.274, "shared_terms": [ "household", "income", @@ -10161,7 +10530,7 @@ "kind": "agent", "id": "calibration-diagnostics" }, - "score": 0.1536, + "score": 0.1541, "shared_terms": [ "microcosm", "unit", @@ -10182,7 +10551,7 @@ "kind": "agent", "id": "dashboard-integrator" }, - "score": 0.1564, + "score": 0.1505, "shared_terms": [ "household", "income", @@ -10203,7 +10572,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1878, + "score": 0.1891, "shared_terms": [ "income", "tax", @@ -10211,8 +10580,8 @@ "model", "canada", "personal", - "household", - "gov" + "gov", + "rate" ] }, { @@ -10224,7 +10593,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.2978, + "score": 0.2871, "shared_terms": [ "household", "income", @@ -10245,7 +10614,7 @@ "kind": "skill", "id": "policyengine-user-guide" }, - "score": 0.2238, + "score": 0.221, "shared_terms": [ "household", "tax", @@ -10266,7 +10635,7 @@ "kind": "agent", "id": "calibration-diagnostics" }, - "score": 0.1814, + "score": 0.1831, "shared_terms": [ "unit", "tax", @@ -10275,7 +10644,7 @@ "state", "poverty", "ctc", - "eitc" + "person" ] }, { @@ -10287,7 +10656,7 @@ "kind": "agent", "id": "dashboard-integrator" }, - "score": 0.1744, + "score": 0.1703, "shared_terms": [ "household", "unit", @@ -10308,7 +10677,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.3053, + "score": 0.3068, "shared_terms": [ "tax", "income", @@ -10329,7 +10698,7 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.2107, + "score": 0.2127, "shared_terms": [ "tax", "state", @@ -10350,15 +10719,15 @@ "kind": "command", "id": "household-calc" }, - "score": 0.2622, + "score": 0.2556, "shared_terms": [ "household", "tax", "income", "level", "net", - "reform", "year", + "reform", "benefits" ] }, @@ -10371,13 +10740,13 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.17, + "score": 0.1685, "shared_terms": [ "reform", "impact", "region", - "calculator", "baseline", + "calculator", "org", "poverty", "wide" @@ -10392,7 +10761,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.191, + "score": 0.1915, "shared_terms": [ "reform", "income", @@ -10413,12 +10782,12 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.1946, + "score": 0.1901, "shared_terms": [ "reform", "calculator", - "household", "charts", + "household", "impact", "poverty", "income", @@ -10434,10 +10803,10 @@ "kind": "command", "id": "household-calc" }, - "score": 0.3298, + "score": 0.3213, "shared_terms": [ - "household", "reform", + "household", "income", "population", "benefits", @@ -10455,16 +10824,16 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.1687, + "score": 0.166, "shared_terms": [ "reform", "impact", "analyze", - "link", "reforms", + "link", "org", "parameter", - "into" + "ages" ] }, { @@ -10476,16 +10845,16 @@ "kind": "agent", "id": "content-orchestrator" }, - "score": 0.1692, + "score": 0.1869, "shared_terms": [ "social", "blog", "posts", "copy", + "analysis", "content", - "post", - "keep", - "audience" + "public", + "not" ] }, { @@ -10497,15 +10866,15 @@ "kind": "agent", "id": "reform-describer" }, - "score": 0.1797, + "score": 0.1819, "shared_terms": [ "sentence", "blog", "description", "descriptions", "writing", - "posts", "neutral", + "posts", "advocacy" ] }, @@ -10518,7 +10887,7 @@ "kind": "agent", "id": "neutrality-reviewer" }, - "score": 0.2118, + "score": 0.2117, "shared_terms": [ "neutral", "blog", @@ -10539,16 +10908,16 @@ "kind": "command", "id": "generate-content" }, - "score": 0.2213, + "score": 0.2132, "shared_terms": [ "social", "blog", "posts", + "analysis", "post", "copy", "content", - "writing", - "neutrality" + "writing" ] }, { @@ -10560,15 +10929,15 @@ "kind": "agent", "id": "ci-fixer" }, - "score": 0.2085, + "score": 0.2089, "shared_terms": [ "root", "review", "prefix", "workflow", - "run", "test", "read", + "run", "local" ] }, @@ -10581,7 +10950,7 @@ "kind": "agent", "id": "implementation-validator" }, - "score": 0.2718, + "score": 0.2626, "shared_terms": [ "review", "root", @@ -10602,12 +10971,12 @@ "kind": "agent", "id": "model-worker" }, - "score": 0.2504, + "score": 0.2541, "shared_terms": [ - "canonical", "delegation", - "coordinator", + "canonical", "role", + "coordinator", "work", "own", "model", @@ -10623,7 +10992,7 @@ "kind": "agent", "id": "program-reviewer" }, - "score": 0.2105, + "score": 0.2058, "shared_terms": [ "review", "program", @@ -10631,8 +11000,8 @@ "code", "regulatory", "workflow", - "reviewer", - "read" + "read", + "reviewer" ] }, { @@ -10644,7 +11013,7 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.4805, + "score": 0.4682, "shared_terms": [ "review", "canonical", @@ -10653,7 +11022,7 @@ "completion", "workflow", "reviewers", - "source" + "read" ] }, { @@ -10665,16 +11034,16 @@ "kind": "agent", "id": "reference-validator" }, - "score": 0.162, + "score": 0.1627, "shared_terms": [ "pdf", "review", "reference", "references", - "source", "validation", + "source", "program", - "missing" + "text" ] }, { @@ -10686,7 +11055,7 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.1807, + "score": 0.1842, "shared_terms": [ "pdf", "audit", @@ -10707,7 +11076,7 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.2543, + "score": 0.2508, "shared_terms": [ "review", "program", @@ -10719,6 +11088,27 @@ "phase" ] }, + { + "a": { + "kind": "skill", + "id": "review-program" + }, + "b": { + "kind": "command", + "id": "comparables-review" + }, + "score": 0.1585, + "shared_terms": [ + "review", + "source", + "dispatch", + "local", + "resolves", + "search", + "run", + "repository" + ] + }, { "a": { "kind": "agent", @@ -10728,7 +11118,7 @@ "kind": "agent", "id": "edge-case-generator" }, - "score": 0.1935, + "score": 0.1953, "shared_terms": [ "test", "tests", @@ -10749,7 +11139,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.2379, + "score": 0.2402, "shared_terms": [ "test", "tests", @@ -10770,7 +11160,7 @@ "kind": "agent", "id": "dashboard-architecture-validator" }, - "score": 0.1674, + "score": 0.1671, "shared_terms": [ "test", "app", @@ -10791,7 +11181,7 @@ "kind": "agent", "id": "dashboard-design-validator" }, - "score": 0.1927, + "score": 0.1919, "shared_terms": [ "tsx", "components", @@ -10812,7 +11202,7 @@ "kind": "agent", "id": "dashboard-plan-validator" }, - "score": 0.153, + "score": 0.1532, "shared_terms": [ "component", "tsx", @@ -10833,7 +11223,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.2267, + "score": 0.2256, "shared_terms": [ "component", "button", @@ -10854,7 +11244,7 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.3289, + "score": 0.3284, "shared_terms": [ "component", "tsx", @@ -10875,7 +11265,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.3842, + "score": 0.3859, "shared_terms": [ "test", "tests", @@ -10896,7 +11286,7 @@ "kind": "agent", "id": "dashboard-plan-validator" }, - "score": 0.2204, + "score": 0.2206, "shared_terms": [ "zone", "path", @@ -10917,16 +11307,16 @@ "kind": "command", "id": "audit-multizone" }, - "score": 0.5812, + "score": 0.5867, "shared_terms": [ "zone", "path", "host", "next", - "repo", "config", - "app", - "basepath" + "repo", + "basepath", + "app" ] }, { @@ -10938,16 +11328,16 @@ "kind": "command", "id": "audit-seo" }, - "score": 0.1932, + "score": 0.1943, "shared_terms": [ "zone", "repo", "url", "path", "legacy", - "zones", "host", - "mounted" + "mounted", + "zones" ] }, { @@ -10959,7 +11349,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.221, + "score": 0.2251, "shared_terms": [ "host", "zone", @@ -10980,7 +11370,7 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.1789, + "score": 0.1786, "shared_terms": [ "dataset", "microcosm", @@ -11001,7 +11391,7 @@ "kind": "agent", "id": "model-corroborator" }, - "score": 0.168, + "score": 0.1674, "shared_terms": [ "calibration", "reform", @@ -11022,7 +11412,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.175, + "score": 0.1753, "shared_terms": [ "state", "income", @@ -11043,15 +11433,15 @@ "kind": "agent", "id": "reform-comparator" }, - "score": 0.1962, + "score": 0.1946, "shared_terms": [ "dataset", "microsim", "anchor", "prior", "calibration", - "result", "comparator", + "result", "cost" ] }, @@ -11064,18 +11454,39 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.1714, + "score": 0.1703, "shared_terms": [ "calibration", "microsim", - "microcosm", "dataset", + "microcosm", "hypothesis", "cost", - "prior", + "frs", "default" ] }, + { + "a": { + "kind": "agent", + "id": "content-orchestrator" + }, + "b": { + "kind": "command", + "id": "comparables-review" + }, + "score": 0.3753, + "shared_terms": [ + "scorecard", + "comparables", + "analysis", + "push", + "record", + "social", + "review", + "slug" + ] + }, { "a": { "kind": "agent", @@ -11085,16 +11496,16 @@ "kind": "command", "id": "generate-content" }, - "score": 0.6314, + "score": 0.7386, "shared_terms": [ "social", "content", + "analysis", + "scorecard", "images", - "quote", - "image", - "blog", - "generate", - "announcement" + "comparables", + "copy", + "blog" ] }, { @@ -11106,7 +11517,7 @@ "kind": "agent", "id": "edge-case-generator" }, - "score": 0.2358, + "score": 0.2369, "shared_terms": [ "test", "files", @@ -11127,7 +11538,7 @@ "kind": "agent", "id": "implementation-validator" }, - "score": 0.3471, + "score": 0.3446, "shared_terms": [ "fix", "root", @@ -11135,8 +11546,8 @@ "run", "prefix", "read", - "review", - "fixes" + "fixes", + "review" ] }, { @@ -11148,13 +11559,13 @@ "kind": "agent", "id": "model-worker" }, - "score": 0.2064, + "score": 0.208, "shared_terms": [ "test", "root", "run", - "blocked", "role", + "blocked", "contract", "manifest", "encode" @@ -11169,7 +11580,7 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.2599, + "score": 0.2594, "shared_terms": [ "caller", "fix", @@ -11190,12 +11601,12 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.1626, + "score": 0.1633, "shared_terms": [ "test", - "review", "blocked", "root", + "review", "role", "read", "evidence", @@ -11211,7 +11622,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.2739, + "score": 0.2755, "shared_terms": [ "test", "tests", @@ -11232,10 +11643,10 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.2748, + "score": 0.2702, "shared_terms": [ - "worktree", "root", + "worktree", "files", "prefix", "program", @@ -11253,7 +11664,7 @@ "kind": "command", "id": "encode-reform" }, - "score": 0.151, + "score": 0.1511, "shared_terms": [ "prefix", "test", @@ -11274,7 +11685,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.2628, + "score": 0.2618, "shared_terms": [ "test", "tests", @@ -11295,14 +11706,14 @@ "kind": "agent", "id": "program-reviewer" }, - "score": 0.2316, + "score": 0.2319, "shared_terms": [ "implementations", "regulations", "program", "income", - "disregards", "government", + "disregards", "benefit", "deductions" ] @@ -11316,7 +11727,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.1624, + "score": 0.1612, "shared_terms": [ "month", "months", @@ -11337,16 +11748,16 @@ "kind": "agent", "id": "reference-validator" }, - "score": 0.1797, + "score": 0.1801, "shared_terms": [ "pdf", "citation", "text", "source", - "page", "doc", + "page", "authoritative", - "references" + "document" ] }, { @@ -11358,16 +11769,16 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.2044, + "score": 0.2046, "shared_terms": [ "pdf", "state", "source", "tmp", - "sources", "income", + "sources", "documents", - "document" + "download" ] }, { @@ -11379,11 +11790,11 @@ "kind": "agent", "id": "implementation-validator" }, - "score": 0.2896, + "score": 0.2882, "shared_terms": [ "mode", - "review", "fix", + "review", "read", "program", "only", @@ -11400,7 +11811,7 @@ "kind": "agent", "id": "program-reviewer" }, - "score": 0.2226, + "score": 0.221, "shared_terms": [ "income", "tests", @@ -11421,7 +11832,7 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.1602, + "score": 0.1566, "shared_terms": [ "test", "review", @@ -11442,7 +11853,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.468, + "score": 0.4682, "shared_terms": [ "test", "tests", @@ -11463,12 +11874,12 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.1651, + "score": 0.1621, "shared_terms": [ "files", "program", - "review", "backdate", + "review", "read", "existing", "only", @@ -11484,7 +11895,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.3743, + "score": 0.377, "shared_terms": [ "test", "tests", @@ -11505,7 +11916,7 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.1651, + "score": 0.1638, "shared_terms": [ "fix", "review", @@ -11526,7 +11937,7 @@ "kind": "agent", "id": "program-reviewer" }, - "score": 0.2137, + "score": 0.2108, "shared_terms": [ "mode", "program", @@ -11547,7 +11958,7 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.2036, + "score": 0.1957, "shared_terms": [ "review", "program", @@ -11555,8 +11966,8 @@ "read", "code", "root", - "only", - "verification" + "verification", + "only" ] }, { @@ -11568,7 +11979,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.1638, + "score": 0.1645, "shared_terms": [ "fix", "program", @@ -11576,8 +11987,8 @@ "read", "not", "findings", - "only", - "variable" + "variable", + "only" ] }, { @@ -11589,15 +12000,15 @@ "kind": "agent", "id": "reference-validator" }, - "score": 0.1575, + "score": 0.1561, "shared_terms": [ "reference", "pdf", - "validator", - "review", "mode", "fix", + "validator", "program", + "review", "issues" ] }, @@ -11610,11 +12021,11 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.3015, + "score": 0.2962, "shared_terms": [ "program", - "worktree", "root", + "worktree", "review", "read", "only", @@ -11631,7 +12042,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.1724, + "score": 0.1736, "shared_terms": [ "structural", "mode", @@ -11652,14 +12063,14 @@ "kind": "agent", "id": "pr-pusher" }, - "score": 0.3155, + "score": 0.3115, "shared_terms": [ "repo", "branch", - "push", "head", "issue", "number", + "push", "base", "repository" ] @@ -11673,16 +12084,16 @@ "kind": "agent", "id": "structural-implementer" }, - "score": 0.1914, + "score": 0.1912, "shared_terms": [ "issue", "repo", "branch", "base", "draft", - "push", "url", - "open" + "push", + "number" ] }, { @@ -11694,7 +12105,7 @@ "kind": "agent", "id": "report-logger" }, - "score": 0.1961, + "score": 0.1973, "shared_terms": [ "issue", "repo", @@ -11703,7 +12114,7 @@ "has", "github", "url", - "run" + "git" ] }, { @@ -11715,7 +12126,7 @@ "kind": "command", "id": "audit-seo" }, - "score": 0.1819, + "score": 0.1823, "shared_terms": [ "url", "repo", @@ -11736,7 +12147,7 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.1533, + "score": 0.1532, "shared_terms": [ "state", "number", @@ -11757,7 +12168,7 @@ "kind": "command", "id": "create-pr" }, - "score": 0.2621, + "score": 0.2638, "shared_terms": [ "branch", "number", @@ -11778,7 +12189,7 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.6196, + "score": 0.6253, "shared_terms": [ "your", "assigned", @@ -11799,16 +12210,16 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.1793, + "score": 0.1803, "shared_terms": [ "test", "tests", "assigned", - "contract", "model", + "contract", "cases", "owned", - "actual" + "manifest" ] }, { @@ -11820,7 +12231,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.1701, + "score": 0.1713, "shared_terms": [ "test", "tests", @@ -11841,14 +12252,14 @@ "kind": "agent", "id": "structural-implementer" }, - "score": 0.198, + "score": 0.196, "shared_terms": [ "changelog", "issue", "draft", "commit", - "push", "tests", + "push", "caller", "repo" ] @@ -11862,14 +12273,14 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.1614, + "score": 0.1611, "shared_terms": [ "tests", - "commit", "caller", + "commit", "pusher", - "files", "fix", + "files", "contract", "create" ] @@ -11883,15 +12294,15 @@ "kind": "command", "id": "create-pr" }, - "score": 0.2475, + "score": 0.246, "shared_terms": [ "branch", "git", "draft", - "commit", "number", - "create", + "commit", "head", + "create", "base" ] }, @@ -11904,10 +12315,10 @@ "kind": "agent", "id": "review-worker" }, - "score": 0.2211, + "score": 0.2171, "shared_terms": [ - "review", "program", + "review", "your", "code", "assigned", @@ -11925,7 +12336,7 @@ "kind": "agent", "id": "rules-engineer" }, - "score": 0.1835, + "score": 0.1864, "shared_terms": [ "program", "tanf", @@ -11946,7 +12357,7 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.2453, + "score": 0.2469, "shared_terms": [ "tests", "test", @@ -11967,7 +12378,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1742, + "score": 0.1747, "shared_terms": [ "income", "state", @@ -11988,16 +12399,16 @@ "kind": "agent", "id": "reference-validator" }, - "score": 0.1982, + "score": 0.1986, "shared_terms": [ "requirements", "program", "income", "correct", - "missing", "reference", - "review", - "implementations" + "missing", + "implementations", + "manual" ] }, { @@ -12009,7 +12420,7 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.178, + "score": 0.176, "shared_terms": [ "program", "review", @@ -12030,16 +12441,16 @@ "kind": "agent", "id": "test-creator" }, - "score": 0.1576, + "score": 0.1593, "shared_terms": [ "test", "assigned", "program", "model", - "read", "needed", - "actual", - "development" + "read", + "development", + "actual" ] }, { @@ -12051,15 +12462,15 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.1675, + "score": 0.1634, "shared_terms": [ "program", "review", "read", "root", "workflow", - "only", "role", + "you", "task" ] }, @@ -12072,11 +12483,11 @@ "kind": "agent", "id": "reference-validator" }, - "score": 0.3645, + "score": 0.3641, "shared_terms": [ "href", - "page", "title", + "page", "section", "pdf", "manual", @@ -12093,15 +12504,15 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.1825, + "score": 0.1819, "shared_terms": [ "pdf", "oregon", "page", "gov", "full", - "https", "income", + "https", "parameter" ] }, @@ -12114,7 +12525,7 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.179, + "score": 0.1763, "shared_terms": [ "program", "page", @@ -12135,7 +12546,7 @@ "kind": "agent", "id": "policy-text-researcher" }, - "score": 0.1655, + "score": 0.164, "shared_terms": [ "enacted", "effective", @@ -12143,8 +12554,8 @@ "bill", "structural", "gov", - "date", - "enrolled" + "enrolled", + "date" ] }, { @@ -12156,7 +12567,7 @@ "kind": "agent", "id": "reform-classifier" }, - "score": 0.1941, + "score": 0.1953, "shared_terms": [ "structural", "backlog", @@ -12177,14 +12588,14 @@ "kind": "agent", "id": "report-logger" }, - "score": 0.2029, + "score": 0.2031, "shared_terms": [ "issue", "draft", "repo", "structural", - "country", "verdict", + "country", "report", "body" ] @@ -12198,7 +12609,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.6504, + "score": 0.653, "shared_terms": [ "structural", "backlog", @@ -12219,7 +12630,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.3952, + "score": 0.3981, "shared_terms": [ "test", "tests", @@ -12240,7 +12651,7 @@ "kind": "agent", "id": "dashboard-architecture-validator" }, - "score": 0.2093, + "score": 0.2101, "shared_terms": [ "modal", "backend", @@ -12248,8 +12659,8 @@ "dashboard", "package", "pattern", - "json", - "data" + "data", + "json" ] }, { @@ -12261,7 +12672,7 @@ "kind": "agent", "id": "dashboard-integrator" }, - "score": 0.4004, + "score": 0.4037, "shared_terms": [ "api", "backend", @@ -12282,7 +12693,7 @@ "kind": "agent", "id": "dashboard-plan-validator" }, - "score": 0.2101, + "score": 0.2107, "shared_terms": [ "plan", "api", @@ -12303,7 +12714,7 @@ "kind": "agent", "id": "dashboard-planner" }, - "score": 0.3376, + "score": 0.3348, "shared_terms": [ "dashboard", "api", @@ -12324,7 +12735,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.5143, + "score": 0.5151, "shared_terms": [ "api", "dashboard", @@ -12345,7 +12756,7 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.1858, + "score": 0.1875, "shared_terms": [ "api", "data", @@ -12366,14 +12777,14 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.203, + "score": 0.1977, "shared_terms": [ "dashboard", "plan", - "reform", "api", - "page", + "reform", "backend", + "page", "precomputed", "json" ] @@ -12387,7 +12798,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.3927, + "score": 0.3989, "shared_terms": [ "modal", "dashboard", @@ -12395,8 +12806,8 @@ "pattern", "backend", "precomputed", - "plan", - "data" + "data", + "plan" ] }, { @@ -12408,7 +12819,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2207, + "score": 0.2177, "shared_terms": [ "api", "dashboard", @@ -12417,7 +12828,7 @@ "custom", "client", "precomputed", - "page" + "pattern" ] }, { @@ -12429,13 +12840,13 @@ "kind": "agent", "id": "dashboard-design-validator" }, - "score": 0.4117, + "score": 0.4107, "shared_terms": [ "grep", "app", "kit", - "tsx", "include", + "tsx", "components", "css", "node" @@ -12450,12 +12861,12 @@ "kind": "agent", "id": "dashboard-plan-validator" }, - "score": 0.2755, + "score": 0.2744, "shared_terms": [ "grep", "app", - "fail", "include", + "fail", "tsx", "components", "test", @@ -12471,13 +12882,13 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.2281, + "score": 0.2286, "shared_terms": [ "backend", "css", - "dashboard", "custom", "modal", + "dashboard", "app", "kit", "tsx" @@ -12492,7 +12903,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.2578, + "score": 0.2558, "shared_terms": [ "app", "kit", @@ -12513,7 +12924,7 @@ "kind": "command", "id": "audit-multizone" }, - "score": 0.1799, + "score": 0.1783, "shared_terms": [ "app", "config", @@ -12522,7 +12933,7 @@ "bash", "package", "json", - "validator" + "echo" ] }, { @@ -12534,7 +12945,7 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.2105, + "score": 0.2089, "shared_terms": [ "app", "kit", @@ -12555,7 +12966,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.2152, + "score": 0.2143, "shared_terms": [ "modal", "app", @@ -12576,7 +12987,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2586, + "score": 0.2591, "shared_terms": [ "css", "app", @@ -12597,7 +13008,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.1728, + "score": 0.1729, "shared_terms": [ "test", "app", @@ -12618,7 +13029,7 @@ "kind": "agent", "id": "dashboard-plan-validator" }, - "score": 0.4587, + "score": 0.4595, "shared_terms": [ "grep", "components", @@ -12639,7 +13050,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.1591, + "score": 0.1583, "shared_terms": [ "components", "tsx", @@ -12660,7 +13071,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.3516, + "score": 0.3507, "shared_terms": [ "kit", "components", @@ -12681,14 +13092,14 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.3462, + "score": 0.3461, "shared_terms": [ "components", "kit", "grep", "tsx", - "app", "design", + "app", "component", "include" ] @@ -12702,7 +13113,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.162, + "score": 0.1624, "shared_terms": [ "kit", "app", @@ -12723,7 +13134,7 @@ "kind": "agent", "id": "dashboard-plan-validator" }, - "score": 0.222, + "score": 0.2236, "shared_terms": [ "api", "components", @@ -12744,7 +13155,7 @@ "kind": "agent", "id": "dashboard-planner" }, - "score": 0.1887, + "score": 0.1885, "shared_terms": [ "api", "data", @@ -12752,8 +13163,8 @@ "step", "plan", "backend", - "input", - "components" + "components", + "input" ] }, { @@ -12765,7 +13176,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.3195, + "score": 0.3222, "shared_terms": [ "api", "client", @@ -12786,11 +13197,11 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.1764, + "score": 0.1744, "shared_terms": [ "components", - "builder", "frontend", + "builder", "api", "component", "design", @@ -12807,7 +13218,7 @@ "kind": "agent", "id": "microsim-runner" }, - "score": 0.2167, + "score": 0.2189, "shared_terms": [ "api", "data", @@ -12828,7 +13239,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1504, + "score": 0.1513, "shared_terms": [ "api", "state", @@ -12840,27 +13251,6 @@ "household" ] }, - { - "a": { - "kind": "agent", - "id": "dashboard-integrator" - }, - "b": { - "kind": "command", - "id": "household-calc" - }, - "score": 0.1502, - "shared_terms": [ - "household", - "api", - "income", - "values", - "year", - "tax", - "net", - "pattern" - ] - }, { "a": { "kind": "agent", @@ -12870,7 +13260,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.1777, + "score": 0.1791, "shared_terms": [ "api", "client", @@ -12891,7 +13281,7 @@ "kind": "agent", "id": "dashboard-planner" }, - "score": 0.2117, + "score": 0.2111, "shared_terms": [ "plan", "dashboard", @@ -12912,7 +13302,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.3176, + "score": 0.3189, "shared_terms": [ "plan", "dashboard", @@ -12933,7 +13323,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.2543, + "score": 0.2517, "shared_terms": [ "component", "plan", @@ -12954,7 +13344,7 @@ "kind": "command", "id": "audit-multizone" }, - "score": 0.2391, + "score": 0.2392, "shared_terms": [ "zone", "path", @@ -12975,7 +13365,7 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.2142, + "score": 0.2147, "shared_terms": [ "component", "components", @@ -12996,11 +13386,11 @@ "kind": "command", "id": "new-tool" }, - "score": 0.1531, + "score": 0.1526, "shared_terms": [ - "dashboard", "api", "embedding", + "dashboard", "app", "plan", "multi", @@ -13017,7 +13407,7 @@ "kind": "agent", "id": "dashboard-scaffold" }, - "score": 0.3288, + "score": 0.3254, "shared_terms": [ "dashboard", "plan", @@ -13038,16 +13428,16 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.3506, + "score": 0.345, "shared_terms": [ "dashboard", "planner", "plan", "reform", "tabs", - "page", "natural", - "plans" + "plans", + "validation" ] }, { @@ -13059,15 +13449,15 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.189, + "score": 0.196, "shared_terms": [ "dashboard", "modal", "step", + "data", "plan", "custom", "pattern", - "data", "needs" ] }, @@ -13080,7 +13470,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.1645, + "score": 0.1621, "shared_terms": [ "dashboard", "api", @@ -13088,8 +13478,8 @@ "step", "design", "plan", - "page", - "embedding" + "embedding", + "purpose" ] }, { @@ -13101,10 +13491,10 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.1561, + "score": 0.1576, "shared_terms": [ - "natural", "description", + "natural", "language", "api", "reform", @@ -13122,7 +13512,7 @@ "kind": "agent", "id": "frontend-builder" }, - "score": 0.2529, + "score": 0.2538, "shared_terms": [ "kit", "plan", @@ -13143,7 +13533,7 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.296, + "score": 0.2894, "shared_terms": [ "dashboard", "plan", @@ -13164,11 +13554,11 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.1619, + "score": 0.1617, "shared_terms": [ "kit", - "dashboard", "design", + "dashboard", "components", "tsx", "component", @@ -13185,16 +13575,16 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.3113, + "score": 0.3181, "shared_terms": [ "dashboard", "modal", "custom", "plan", "pattern", - "main", "host", - "step" + "step", + "main" ] }, { @@ -13206,7 +13596,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.3408, + "score": 0.3393, "shared_terms": [ "dashboard", "api", @@ -13227,7 +13617,7 @@ "kind": "command", "id": "create-new-component" }, - "score": 0.4584, + "score": 0.4594, "shared_terms": [ "component", "kit", @@ -13248,7 +13638,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.2938, + "score": 0.2967, "shared_terms": [ "kit", "tailwind", @@ -13269,7 +13659,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1744, + "score": 0.1728, "shared_terms": [ "api", "state", @@ -13281,27 +13671,6 @@ "tracker" ] }, - { - "a": { - "kind": "agent", - "id": "microsim-runner" - }, - "b": { - "kind": "agent", - "id": "prior-scores-finder" - }, - "score": 0.1508, - "shared_terms": [ - "year", - "dataset", - "reform", - "legislative", - "country", - "local", - "tracker", - "returns" - ] - }, { "a": { "kind": "agent", @@ -13311,7 +13680,7 @@ "kind": "agent", "id": "reform-comparator" }, - "score": 0.2806, + "score": 0.2815, "shared_terms": [ "dataset", "year", @@ -13332,7 +13701,7 @@ "kind": "agent", "id": "report-logger" }, - "score": 0.1976, + "score": 0.1946, "shared_terms": [ "dataset", "local", @@ -13353,13 +13722,13 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.2989, + "score": 0.2987, "shared_terms": [ "dataset", "api", "year", - "default", "region", + "default", "baseline", "microsim", "local" @@ -13374,10 +13743,10 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1823, + "score": 0.1828, "shared_terms": [ - "reform", "baseline", + "reform", "level", "dict", "api", @@ -13395,15 +13764,15 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.1793, + "score": 0.1786, "shared_terms": [ "reform", "dict", "api", "runs", "url", - "country", "microsim", + "country", "baseline" ] }, @@ -13416,7 +13785,7 @@ "kind": "agent", "id": "parameter-locator" }, - "score": 0.1763, + "score": 0.1747, "shared_terms": [ "reform", "parameter", @@ -13437,13 +13806,13 @@ "kind": "agent", "id": "prior-scores-finder" }, - "score": 0.2337, + "score": 0.2322, "shared_terms": [ "external", "reform", "comparator", - "scores", "source", + "scores", "tier", "our", "scored" @@ -13458,7 +13827,7 @@ "kind": "agent", "id": "reform-classifier" }, - "score": 0.1616, + "score": 0.1598, "shared_terms": [ "reform", "parameter", @@ -13479,16 +13848,16 @@ "kind": "agent", "id": "reform-comparator" }, - "score": 0.2368, + "score": 0.2364, "shared_terms": [ "our", "microsim", "comparator", "result", "reform", - "scores", "anchor", - "calibration" + "calibration", + "scores" ] }, { @@ -13500,7 +13869,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1867, + "score": 0.1848, "shared_terms": [ "reform", "dict", @@ -13508,8 +13877,8 @@ "external", "single", "cap", - "household", - "sources" + "sources", + "household" ] }, { @@ -13521,10 +13890,10 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.1584, + "score": 0.1578, "shared_terms": [ - "reform", "original", + "reform", "model", "issue", "law", @@ -13542,7 +13911,7 @@ "kind": "command", "id": "prior-scores" }, - "score": 0.2836, + "score": 0.2808, "shared_terms": [ "shape", "external", @@ -13550,8 +13919,8 @@ "benchmark", "scores", "microsim", - "published", - "runs" + "runs", + "scored" ] }, { @@ -13563,7 +13932,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2633, + "score": 0.2613, "shared_terms": [ "reform", "dict", @@ -13584,7 +13953,7 @@ "kind": "agent", "id": "prior-scores-finder" }, - "score": 0.1916, + "score": 0.1911, "shared_terms": [ "analyses", "comparator", @@ -13593,7 +13962,7 @@ "scores", "blind", "magnitude", - "provisions" + "you" ] }, { @@ -13605,7 +13974,7 @@ "kind": "agent", "id": "policy-text-researcher" }, - "score": 0.2556, + "score": 0.2563, "shared_terms": [ "gov", "state", @@ -13626,13 +13995,13 @@ "kind": "agent", "id": "reform-classifier" }, - "score": 0.4488, + "score": 0.448, "shared_terms": [ "parameter", "gov", "deployed", - "reform", "provision", + "reform", "lag", "credits", "parametric" @@ -13647,7 +14016,7 @@ "kind": "agent", "id": "reform-describer" }, - "score": 0.3446, + "score": 0.3452, "shared_terms": [ "parameter", "reform", @@ -13668,7 +14037,7 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.2959, + "score": 0.2979, "shared_terms": [ "state", "tax", @@ -13710,7 +14079,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.2188, + "score": 0.2196, "shared_terms": [ "reform", "income", @@ -13731,14 +14100,14 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.2035, + "score": 0.2023, "shared_terms": [ "gov", "reform", "states", "rate", - "deployed", "phase", + "deployed", "structural", "change" ] @@ -13752,7 +14121,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.294, + "score": 0.2915, "shared_terms": [ "reform", "dict", @@ -13773,7 +14142,7 @@ "kind": "agent", "id": "reform-classifier" }, - "score": 0.1506, + "score": 0.1515, "shared_terms": [ "gov", "ctc", @@ -13794,15 +14163,15 @@ "kind": "agent", "id": "reform-describer" }, - "score": 0.3058, + "score": 0.3096, "shared_terms": [ "ctc", "provisions", + "description", "federal", "ages", - "reform", "gov", - "description", + "reform", "text" ] }, @@ -13815,16 +14184,16 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.1535, + "score": 0.1527, "shared_terms": [ "bill", "year", "arpa", "federal", "ctc", - "reform", + "description", "phase", - "number" + "reform" ] }, { @@ -13836,7 +14205,7 @@ "kind": "command", "id": "encode-reform" }, - "score": 0.3301, + "score": 0.3257, "shared_terms": [ "bill", "reform", @@ -13857,7 +14226,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.1654, + "score": 0.1645, "shared_terms": [ "enacted", "gov", @@ -13866,7 +14235,7 @@ "bill", "reform", "draft", - "date" + "baseline" ] }, { @@ -13878,7 +14247,7 @@ "kind": "command", "id": "reform-pipeline" }, - "score": 0.2736, + "score": 0.2703, "shared_terms": [ "bill", "reform", @@ -13899,7 +14268,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.3537, + "score": 0.3512, "shared_terms": [ "bill", "reform", @@ -13920,7 +14289,7 @@ "kind": "agent", "id": "reform-comparator" }, - "score": 0.2828, + "score": 0.2787, "shared_terms": [ "year", "prior", @@ -13928,8 +14297,8 @@ "scores", "microsim", "comparator", - "result", - "our" + "our", + "result" ] }, { @@ -13941,14 +14310,14 @@ "kind": "agent", "id": "report-logger" }, - "score": 0.1518, + "score": 0.153, "shared_terms": [ "analyses", "archive", "verdict", "local", - "country", "jurisdiction", + "country", "run", "tracker" ] @@ -13962,16 +14331,16 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.1755, + "score": 0.1732, "shared_terms": [ "year", "microsim", "prior", "archive", "horizon", - "scores", + "single", "reform", - "single" + "scores" ] }, { @@ -13983,7 +14352,7 @@ "kind": "command", "id": "prior-analysis" }, - "score": 0.399, + "score": 0.3965, "shared_terms": [ "analyses", "prior", @@ -14004,16 +14373,16 @@ "kind": "command", "id": "prior-scores" }, - "score": 0.4006, + "score": 0.3983, "shared_terms": [ "tier", "prior", "scores", "analyses", "external", + "microsim", "cbo", - "jct", - "microsim" + "jct" ] }, { @@ -14025,16 +14394,16 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.1567, + "score": 0.1548, "shared_terms": [ "reform", "microsim", "prior", - "bill", "reforms", + "bill", "scores", - "external", - "not" + "same", + "external" ] }, { @@ -14046,16 +14415,16 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.2398, + "score": 0.2394, "shared_terms": [ "pdf", "reference", - "source", "number", + "source", "full", - "page", "parameter", - "audit" + "audit", + "flag" ] }, { @@ -14067,16 +14436,16 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.1571, + "score": 0.1561, "shared_terms": [ "reference", "pdf", - "page", "program", - "review", + "page", "href", "phase", - "parameter" + "parameter", + "review" ] }, { @@ -14088,7 +14457,7 @@ "kind": "agent", "id": "reform-describer" }, - "score": 0.3083, + "score": 0.3098, "shared_terms": [ "provisions", "parameter", @@ -14109,7 +14478,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.2683, + "score": 0.2687, "shared_terms": [ "structural", "backlog", @@ -14130,7 +14499,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.371, + "score": 0.3693, "shared_terms": [ "reform", "provision", @@ -14151,7 +14520,7 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.3632, + "score": 0.3629, "shared_terms": [ "year", "microsim", @@ -14172,7 +14541,7 @@ "kind": "command", "id": "prior-scores" }, - "score": 0.1883, + "score": 0.1847, "shared_terms": [ "microsim", "prior", @@ -14193,7 +14562,7 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.1642, + "score": 0.1637, "shared_terms": [ "dashboard", "reform", @@ -14201,8 +14570,8 @@ "provisions", "description", "child", - "json", - "multi" + "multi", + "json" ] }, { @@ -14214,7 +14583,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1648, + "score": 0.1658, "shared_terms": [ "reform", "description", @@ -14235,7 +14604,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2343, + "score": 0.236, "shared_terms": [ "reform", "description", @@ -14256,7 +14625,7 @@ "kind": "command", "id": "analyze-policy" }, - "score": 0.2919, + "score": 0.2928, "shared_terms": [ "log", "archive", @@ -14277,7 +14646,7 @@ "kind": "command", "id": "audit-seo" }, - "score": 0.1514, + "score": 0.1525, "shared_terms": [ "issue", "repo", @@ -14298,7 +14667,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.2033, + "score": 0.2032, "shared_terms": [ "issue", "structural", @@ -14306,8 +14675,8 @@ "verdict", "run", "country", - "archive", - "report" + "report", + "archive" ] }, { @@ -14319,7 +14688,7 @@ "kind": "command", "id": "prior-analysis" }, - "score": 0.2477, + "score": 0.25, "shared_terms": [ "archive", "analyses", @@ -14328,7 +14697,7 @@ "frontmatter", "local", "repo", - "country" + "jurisdiction" ] }, { @@ -14340,7 +14709,7 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.1682, + "score": 0.1706, "shared_terms": [ "auto", "horizon", @@ -14361,7 +14730,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1623, + "score": 0.1637, "shared_terms": [ "single", "year", @@ -14382,7 +14751,7 @@ "kind": "command", "id": "reform-pipeline" }, - "score": 0.2224, + "score": 0.2219, "shared_terms": [ "bill", "auto", @@ -14403,7 +14772,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2204, + "score": 0.2201, "shared_terms": [ "microsim", "bill", @@ -14424,7 +14793,7 @@ "kind": "command", "id": "audit-seo" }, - "score": 0.2894, + "score": 0.2903, "shared_terms": [ "audit", "zone", @@ -14445,7 +14814,7 @@ "kind": "command", "id": "create-dashboard" }, - "score": 0.1865, + "score": 0.187, "shared_terms": [ "repo", "multi", @@ -14466,7 +14835,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.3263, + "score": 0.3308, "shared_terms": [ "host", "zone", @@ -14487,7 +14856,7 @@ "kind": "command", "id": "write-tests" }, - "score": 0.1547, + "score": 0.1553, "shared_terms": [ "path", "target", @@ -14508,7 +14877,7 @@ "kind": "command", "id": "audit-state-tax" }, - "score": 0.1825, + "score": 0.1803, "shared_terms": [ "audit", "url", @@ -14529,7 +14898,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.1565, + "score": 0.1573, "shared_terms": [ "vercel", "deploy", @@ -14550,7 +14919,7 @@ "kind": "command", "id": "backdate-program" }, - "score": 0.1945, + "score": 0.1941, "shared_terms": [ "pdf", "state", @@ -14559,7 +14928,7 @@ "arguments", "files", "only", - "page" + "instruction" ] }, { @@ -14571,16 +14940,16 @@ "kind": "command", "id": "encode-reform" }, - "score": 0.1764, + "score": 0.1776, "shared_terms": [ "pdf", "state", "gov", - "arguments", "tax", "full", + "arguments", "phase", - "audit" + "number" ] }, { @@ -14592,16 +14961,58 @@ "kind": "command", "id": "encode-reform" }, - "score": 0.2536, + "score": 0.2509, "shared_terms": [ "prefix", "arguments", - "short", "orchestrator", + "short", "files", "agents", + "page", + "pdf" + ] + }, + { + "a": { + "kind": "command", + "id": "comparables-review" + }, + "b": { + "kind": "command", + "id": "generate-content" + }, + "score": 0.4082, + "shared_terms": [ + "scorecard", + "comparables", + "analysis", + "social", "review", - "page" + "push", + "record", + "post" + ] + }, + { + "a": { + "kind": "command", + "id": "comparables-review" + }, + "b": { + "kind": "command", + "id": "reform-pipeline" + }, + "score": 0.1908, + "shared_terms": [ + "bill", + "tracker", + "publish", + "search", + "score", + "dashboard", + "trackers", + "publication" ] }, { @@ -14613,7 +15024,7 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.1945, + "score": 0.1987, "shared_terms": [ "dashboard", "repo", @@ -14634,13 +15045,13 @@ "kind": "command", "id": "encode-reform" }, - "score": 0.1544, + "score": 0.1531, "shared_terms": [ "reform", "page", "multi", - "arguments", "implement", + "arguments", "phase", "skip", "workflow" @@ -14655,7 +15066,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1678, + "score": 0.1661, "shared_terms": [ "reform", "household", @@ -14676,7 +15087,7 @@ "kind": "command", "id": "new-tool" }, - "score": 0.3098, + "score": 0.3004, "shared_terms": [ "dashboard", "create", @@ -14697,7 +15108,7 @@ "kind": "command", "id": "reform-pipeline" }, - "score": 0.2227, + "score": 0.2235, "shared_terms": [ "dashboard", "reform", @@ -14726,8 +15137,8 @@ "export", "design", "dashboard", - "src", - "import" + "import", + "css" ] }, { @@ -14739,14 +15150,14 @@ "kind": "command", "id": "deploy-dashboard" }, - "score": 0.1849, + "score": 0.1864, "shared_terms": [ "branch", "git", "main", "check", - "step", "bash", + "step", "json", "feature" ] @@ -14760,16 +15171,16 @@ "kind": "command", "id": "new-tool" }, - "score": 0.1545, + "score": 0.1581, "shared_terms": [ "dashboard", "app", "modal", - "step", "bash", + "step", "embedding", "bun", - "multi" + "next" ] }, { @@ -14781,7 +15192,7 @@ "kind": "command", "id": "household-calc" }, - "score": 0.1513, + "score": 0.1508, "shared_terms": [ "reform", "level", @@ -14802,7 +15213,7 @@ "kind": "command", "id": "implement-structural" }, - "score": 0.1805, + "score": 0.1819, "shared_terms": [ "reform", "contrib", @@ -14823,7 +15234,7 @@ "kind": "command", "id": "reform-pipeline" }, - "score": 0.3156, + "score": 0.3144, "shared_terms": [ "bill", "reform", @@ -14844,7 +15255,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.3149, + "score": 0.3097, "shared_terms": [ "bill", "reform", @@ -14865,16 +15276,16 @@ "kind": "command", "id": "prior-scores" }, - "score": 0.1525, + "score": 0.152, "shared_terms": [ "reform", "analyze", "consult", - "scores", "external", - "prior", + "scores", "benchmark", - "microsim" + "microsim", + "prior" ] }, { @@ -14886,12 +15297,12 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2845, + "score": 0.2847, "shared_terms": [ "reform", "dict", - "preset", "analyze", + "preset", "description", "entry", "microsim", @@ -14907,14 +15318,14 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.1766, + "score": 0.174, "shared_terms": [ "reform", "structural", "analyze", "bill", - "deployed", "formula", + "deployed", "extension", "pre" ] @@ -14928,7 +15339,7 @@ "kind": "command", "id": "prior-scores" }, - "score": 0.2166, + "score": 0.2134, "shared_terms": [ "analyses", "prior", @@ -14937,7 +15348,7 @@ "archive", "change", "archived", - "scores" + "runs" ] }, { @@ -14949,7 +15360,7 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2538, + "score": 0.2529, "shared_terms": [ "reform", "scorekeepers", @@ -14958,7 +15369,7 @@ "presets", "bill", "consult", - "scores" + "runs" ] }, { @@ -14970,12 +15381,12 @@ "kind": "command", "id": "text-to-reform" }, - "score": 0.2762, + "score": 0.2724, "shared_terms": [ "bill", "reform", - "coverage", "api", + "coverage", "pipeline", "end", "analyze", @@ -15115,6 +15526,7 @@ "./skills/policyengine-writing", "./skills/policyengine-research-lookup", "./skills/policyengine-prior-scores", + "./skills/policyengine-comparables-review", "./skills/policyengine-calibration-diagnostics" ], "commands": [ @@ -15126,7 +15538,8 @@ "./commands/prior-analysis.md", "./commands/text-to-reform.md", "./commands/create-dashboard.md", - "./commands/deploy-dashboard.md" + "./commands/deploy-dashboard.md", + "./commands/comparables-review.md" ], "agents": [ "./agents/calibration-diagnostics.md", @@ -15223,6 +15636,7 @@ "./skills/policyengine-standards", "./skills/policyengine-cloud-costs", "./skills/policyengine-writing", + "./skills/policyengine-comparables-review", "./skills/fix-pr" ], "commands": [ @@ -15231,7 +15645,8 @@ "./commands/new-tool.md", "./commands/audit-seo.md", "./commands/audit-multizone.md", - "./commands/write-tests.md" + "./commands/write-tests.md", + "./commands/comparables-review.md" ], "agents": [ "./agents/app/component-test-writer.md", @@ -15279,6 +15694,7 @@ "./skills/policyengine-user-guide", "./skills/policyengine-research-lookup", "./skills/policyengine-prior-scores", + "./skills/policyengine-comparables-review", "./skills/policyengine-calibration-diagnostics", "./skills/policyengine-content", "./skills/policyengine-github-agent", @@ -15293,6 +15709,7 @@ "./commands/audit-seo.md", "./commands/audit-state-tax.md", "./commands/backdate-program.md", + "./commands/comparables-review.md", "./commands/create-dashboard.md", "./commands/create-new-component.md", "./commands/create-pr.md", @@ -15432,11 +15849,13 @@ "./skills/policyengine-design", "./skills/policyengine-standards", "./skills/policyengine-cloud-costs", - "./skills/policyengine-writing" + "./skills/policyengine-writing", + "./skills/policyengine-comparables-review" ], "commands": [ "./commands/create-dashboard.md", - "./commands/deploy-dashboard.md" + "./commands/deploy-dashboard.md", + "./commands/comparables-review.md" ], "agents": [ "./agents/dashboard/dashboard-planner.md", @@ -17992,6 +18411,7 @@ "skills": [ "policyengine-api-access", "policyengine-app", + "policyengine-comparables-review", "policyengine-research-lookup", "policyengine-tools", "policyengine-user-guide" @@ -18009,7 +18429,7 @@ "deploy-dashboard", "write-tests" ], - "total": 14 + "total": 15 }, { "name": "policyengine-canada", @@ -18024,6 +18444,7 @@ "policyengine-app", "policyengine-calibration-diagnostics", "policyengine-canada", + "policyengine-comparables-review", "policyengine-data", "policyengine-model-development", "policyengine-prior-scores", @@ -18044,6 +18465,7 @@ "reform-describer" ], "commands": [ + "comparables-review", "deploy-dashboard", "encode-reform", "household-calc", @@ -18051,7 +18473,7 @@ "prior-scores", "reform-pipeline" ], - "total": 26 + "total": 28 }, { "name": "policyengine-claude", @@ -18615,6 +19037,7 @@ "policyengine-api", "policyengine-api-access", "policyengine-app", + "policyengine-comparables-review", "policyengine-design", "policyengine-plugin-maintenance", "policyengine-prior-scores", @@ -18625,6 +19048,7 @@ "agents": [ "component-test-writer", "multizone-validator", + "content-orchestrator", "backend-builder", "dashboard-architecture-validator", "dashboard-design-validator", @@ -18641,6 +19065,7 @@ "analyze-policy", "audit-multizone", "audit-seo", + "comparables-review", "create-dashboard", "create-new-component", "create-pr", @@ -18651,7 +19076,7 @@ "text-to-reform", "write-tests" ], - "total": 34 + "total": 37 }, { "name": "policyengine.dev", @@ -18844,6 +19269,7 @@ ], "commands": [ "analyze-policy", + "comparables-review", "create-pr", "generate-content", "household-calc", @@ -18854,7 +19280,7 @@ "text-to-reform", "write-tests" ], - "total": 31 + "total": 32 }, { "name": "policyengine-us", @@ -18873,6 +19299,8 @@ "policyengine-calibration-diagnostics", "policyengine-canada", "policyengine-cloud-costs", + "policyengine-comparables-review", + "policyengine-content", "policyengine-data", "policyengine-design", "policyengine-github-agent", @@ -18928,6 +19356,7 @@ "audit-seo", "audit-state-tax", "backdate-program", + "comparables-review", "create-dashboard", "create-pr", "deploy-dashboard", @@ -18941,7 +19370,7 @@ "text-to-reform", "write-tests" ], - "total": 71 + "total": 74 }, { "name": "policyengine-ve", @@ -19318,11 +19747,12 @@ "visibility": "PUBLIC", "pushed_at": "2026-05-28T13:36:24Z", "skills": [ + "policyengine-comparables-review", "policyengine-writing" ], "agents": [], "commands": [], - "total": 1 + "total": 2 }, { "name": "plugin-blog", diff --git a/docs/when-to-use.md b/docs/when-to-use.md index d160fa0..f84cfd5 100644 --- a/docs/when-to-use.md +++ b/docs/when-to-use.md @@ -1,6 +1,6 @@ # When to use what -The PolicyEngine plugin ships 20 command files, 36 agents, and 25 skills (22 knowledge skills plus the three workflow skills, consolidated from 53 in the July 2026 rebuild). The `/encode-policy-v2`, `/review-program`, and `/fix-pr` entries are provided directly by skills: each name resolves to a skill whose `references/workflow.md` is the canonical definition, with no duplicate command stub. Most days you'll only use 3-5 entry points. This guide maps common work to the right one. +The PolicyEngine plugin ships 21 command files, 36 agents, and 26 skills (23 knowledge skills plus the three workflow skills, consolidated from 53 in the July 2026 rebuild). The `/encode-policy-v2`, `/review-program`, and `/fix-pr` entries are provided directly by skills: each name resolves to a skill whose `references/workflow.md` is the canonical definition, with no duplicate command stub. Most days you'll only use 3-5 entry points. This guide maps common work to the right one. ## Quick decision matrix @@ -12,6 +12,7 @@ The PolicyEngine plugin ships 20 command files, 36 agents, and 25 skills (22 kno | Convert a bill / URL / description into a reform-dict | `/text-to-reform` | Submit to PE API or pass to `/analyze-policy` | | Check whether we already analyzed this reform | `/prior-analysis` | Read the archived analysis or run fresh | | Consult external scorekeepers (JCT/CBO/OBR/IFS/CRFB/TPC/etc.) | `/prior-scores` | Cite in your writeup | +| Record every external estimate of an analysis in the scorecard before listing it, or before a newsletter or press push features it | `/comparables-review` | Merge the scorecard PR, then add the block to apps.json or posts.json | | Implement a NEW state benefit program | `/encode-policy-v2` | Inspect its draft PR and included review; address remaining findings | | Implement a STRUCTURAL reform the model can't express as parameters | `/implement-structural` | `/review-program` → `/create-pr` | | Change parameters on an EXISTING program | `/encode-reform` | `/review-program` → `/create-pr` | diff --git a/presets/scorekeepers.yaml b/presets/scorekeepers.yaml index ca4552a..89b8fdd 100644 --- a/presets/scorekeepers.yaml +++ b/presets/scorekeepers.yaml @@ -18,6 +18,11 @@ # search_hints (optional) — extra query hints for WebSearch (e.g., "site:crfb.org") # caveats (optional) — known biases, publication cadence, dataset differences # +# Pre-publish comparables (policyengine-comparables-review): the producers a published +# analysis must be searched against are PolicyEngine/policyengine-scorecard's +# data/comparables/producers.json, not this file. This registry stays the wider pool +# prior-scores-finder consults while an analysis is being built. +# # Tier 2 vs Tier 3 rule: Tier 2 sources have official standing (produce estimates # used in legislation or budget documents). Tier 3 sources are independent analysts # whose estimates are widely cited but not officially binding. When in doubt, use Tier 3. diff --git a/skills/README.md b/skills/README.md index 9d2ba65..588b168 100644 --- a/skills/README.md +++ b/skills/README.md @@ -63,6 +63,7 @@ lints the whole tree against the failure modes that rotted the previous catalog | **policyengine-writing** | Editorial law: neutrality, quantitative language, sentence case | | **policyengine-standards** | uv/bun, ruff, towncrier changelogs, CI gates, PR discipline | | **policyengine-research-lookup** | Finding existing PolicyEngine research and proof points | +| **policyengine-comparables-review** | The required pre-publish search for external estimates, recorded in the scorecard, with every large gap explained | | **policyengine-user-guide** | Using the policyengine.org web app | | **policyengine-content** | Social image/copy generation for posts | | **policyengine-github-agent** | GitHub bot operating rules | diff --git a/skills/policyengine-comparables-review/SKILL.md b/skills/policyengine-comparables-review/SKILL.md new file mode 100644 index 0000000..db3d3a8 --- /dev/null +++ b/skills/policyengine-comparables-review/SKILL.md @@ -0,0 +1,413 @@ +--- +name: policyengine-comparables-review +description: | + The comparables review every PolicyEngine analysis passes before its dashboard or research post + is listed on policyengine.org, and again before a newsletter, social campaign or press push + features it: search every producer on the scorecard's list for external estimates, stage each + in PolicyEngine/policyengine-scorecard, compare it with the analysis's own figures, and explain + every gap over the threshold. CI in the scorecard, policyengine-app-v2 and the newsletters repo + checks parts of this procedure. Includes a brief for running the search as a background job. + Triggers: "comparables", "comparables review", "pre-publish review", "before publishing", + "add to apps.json", "add to posts.json", "scorecard record", "external estimates", + "comparables gate", "analysis-uk-", "analysis-us-", "re-harvest", "newsletter", + "before a push", "promote an analysis". + NOT for: benchmarking a reform mid-analysis without publishing (policyengine-prior-scores, + /prior-scores). +metadata: + category: process +--- + +# Pre-publish comparables review + +Before PolicyEngine publishes an analysis, every external estimate of what it estimates is +found, recorded in PolicyEngine/policyengine-scorecard, compared with the analysis's own +figures, and every large gap is explained. The rule dates from October 2026. The £100,000 +childcare dashboard went live at £0.43bn for the 30 hours in 2027-28, while CenTax had published +£0.80bn from HMRC data a month earlier. The cause, a Microcosm hours draw (microcosm#1126), had +already been filed. A search before launch would have put the two figures side by side, and the +gate would have required the gap to be explained. + +The contract lives in the scorecard: `docs/COMPARABLES.md`. Read it before your first review. +This skill is the operating procedure. + +## What checks it + +| Where | What its CI fails | +|---|---| +| policyengine-scorecard | Missing required producer rows, malformed fields, inconsistent declared identities or results pins, and unexplained over-threshold comparisons computed from those fields; a stale `data/comparables/index.json`; feeds the build rewrites that are not committed. These checks trust the submitted sources and metadata | +| policyengine-app-v2 | A research entry (an apps.json app shown with research or tagged `policy`, or any post) that is not on the list of entries live before the gate (`website/src/data/comparablesGrandfathered.json`) and has no `comparables` block naming a search the scorecard index on main lists. The search must be dated on or before the entry's date, no more than 30 days before it, and not be a backfill. On the day a change lists or re-lists an entry, the search must also be no more than 30 days old that day, have compared the results the analysis reads now, and leave no gap pending. A card a change adds to the featured banner needs a fresh search (see "Before a push"). Evaluated rewrites classified as pages require registration or a listed exception; asset heuristics leave coverage gaps described below | +| PolicyEngine/newsletters | A changed edition whose recognized analysis links fail freshness (see "Before a push"), using app-v2's generated route map and the scorecard records' pages and `deployments`; `upload-newsletter` refuses those failures before it touches Mailchimp. Unknown links can pass without a freshness check | + +The route gate currently treats any filename extension and the entire `/_zones/` prefix as +assets. Rewrites such as `/uk/ungated-analysis.html` and `/_zones/ungated-analysis` can therefore +serve an unregistered analysis without failing it. Audit all new page rewrites manually and +register each analysis even when CI exempts its route. The route builder also drops absolute +redirect destinations, including `/redirect` → `https://policyengine.org/uk/existing`. +Newsletter checks do not follow redirects: resolve aliases to the analysis yourself, verify +the record, and run `fresh` for the intended push date. A green link check does not prove that +it recognized the analysis. + +A green scorecard validator does not establish source truth. It can accept omitted published +estimates, a null metric that avoids comparison, invented or irrelevant cause citations, +matching fake hashes, token negative-search queries, or a printed figure mislabeled as a chart +reading. The reviewer checklist below is required even when every automated check passes. + +The order is: scorecard PR merged first, then the apps.json or posts.json entry. A post that is +plainly not an analysis (tagged `org`, `event`, `giving`, `conferences`, `year-in-review` or +`april-fools`, and no analysis tag) may declare +`"comparables": {"status": "no_estimate", "reason": "..."}` instead. An app never may: a +household calculator is searched like any other analysis (step 2). + +These are CI checks. Until the repositories make them required status checks, a red check is +what stops a merge, so never merge past one. + +## The review, step by step + +Work in a worktree of the scorecard off `origin/main`. Run commands with `uv run` and +`PYTHONPATH=.`. The worked example to copy is `sources/analysis-uk-childcare-100k-threshold/`. + +### 1. Scaffold + +```bash +PYTHONPATH=. uv run python -m scorecard_db.comparables scaffold --country uk --kind app --slug +``` + +`` is the analysis's policyengine.org slug: the apps.json `slug`, or the post's file name +up to its first dot, lowercased, with underscores as hyphens (`UK_Two_Child_Limit.md` → +`uk-two-child-limit`). `--today` defaults to your local date. Only the UK and US have producer +lists so far; for another country, add its list to `data/comparables/producers.json` first. The +skeleton's `search_log.json` lists every producer the search must log. + +For a bill tracker, use `--kind app --slug -` to scaffold: the CLI only accepts +`app` and `post`. Then set `analysis.json`'s `registry.kind` to `tracker` and its `url` to the +actual tracker page. Keep `published: null` until the bill-review merge makes the score public; +complete the date handoff under "Complete publication metadata" afterwards. + +### 2. Pin the analysis's own figures (`analysis.json`) + +Every figure the page publishes that a producer might also estimate goes in `pe_estimates`, +read from the results file the page builds from, never retyped from the page. Leaving a +published figure out is how a miss hides, so list them all. Each estimate has: + +- `key`, `label`, `world`, `metric`, `unit_concept`, `period`, `time_basis`, `value`, + `value_raw` (as the file prints it), `normalization`; +- `conditions.basis`: one word, `static` (no behavioural response), `post_behavioural`, + `forecast` (a projected level, such as current-law revenue) or `outturn`; a fuller account goes + in `conditions.basis_note`; +- `conditions.program`: a slug for what is costed, the programmes or taxes whose change the + figure counts (`thirty_hours`, `both_schemes`, `state_pension_with_income_tax`); a longer + description goes in `conditions.program_note`; +- `conditions.prices: real` for a money figure in a base year's prices (absent means cash); +- for a fiscal year, `conditions.fy` ("2027-28"), with `period` its end year and `time_basis` + `fiscal_year`; anything more about the year goes in `conditions.fy_note`; +- `provenance`: the analysis repo, the commit the live page builds from, the results file's + path and SHA-256, the field path, `computed_at`, the engine versions and the dataset release; +- the value in PolicyEngine's orientation for its metric (`scorecard_db/signs.py` + `ORIENTATION`): `revenue_change` is positive = yield, `benefit_cost_change` positive = more + spending, household income changes positive = gain. A file that prints a cost as a positive + number gets `"sign": -1` and a `sign_basis`. + +`worlds` has one entry per policy world the analysis or a producer scores, with a `policy` slug +and a `description`; reuse a slug the scorecard already uses for the same world. + +`deployments` lists the other https URLs the analysis is served from, such as its own Vercel +deployment. The newsletter resolver uses those hosts even after an analysis leaves the registry. +The validator accepts an empty list, so add URLs already known and complete the publication +metadata after first deployment; a registry entry alone does not preserve that coverage. + +A household calculator publishes no fixed estimate. It declares +`"calculator": {"reason": ""}`, pins no estimates, and its searches +record `results_pins: []`. It is still searched. + +### 3. Search every producer + +For each producer in the skeleton, run its `search_hints` with the analysis's `search_terms`, +check the places its `where_to_look` names, and log what happened: + +| Result | When | +|---|---| +| `found` | You staged its estimates; list the claim keys | +| `none_found` | Nothing; list the queries run and pages checked | +| `found_not_staged` | An estimate exists, but its primary cannot be read: a figure seen only in press, or released under FOI with no published response. The note says what it is and `checked` lists where it was seen. It does not count in `found` | +| `not_applicable` | Only when the producer's `applies_to` excludes the analysis's `scope` | + +`open_search` (any producer not on the list: other think tanks, consultancies, parties, select +committees, the Commons Library) and `academic` are never skipped. A comparable from an +unlisted producer is found under them and keeps its producer's own slug as `source`; a +comparable from a listed producer is logged under that producer's own entry. If an open search +keeps finding one producer, add it to `producers.json` with a `why` and an `added` date of the +day you add it. + +Search engines often answer from secondary coverage. Trace each figure to the producer's own +publication before staging it. + +Each search records `results_pins`, the results files it compared: + +```bash +PYTHONPATH=. uv run python -m scorecard_db.comparables pins analysis-- +``` + +### 4. Stage each estimate (`claims_staged.jsonl`, `manifest.jsonl`) + +Download each primary into a new empty directory, hash it, and add one `manifest.jsonl` line +(url, title, date, publisher, sha256, retrieved, access). Hash the documents your annotations +cite and the pages your searches checked too; a manifest line nothing cites, links or lists as +checked fails the check. Then one claim per published number. +Copy a claim's shape from the worked example: `key`, `source`, `found_by`, `found_on`, +`source_model`, `metric`, `unit_concept`, `value`, `value_raw`, `normalization`, `value_kind`, +`period`, `time_basis`, `conditions`, `world`, `reform_hint`, `attribution`, `benchmark_class`, +`baseline_policy`, `publication` (title, url, date, publisher, primary_sha256, locator, access), +`source_table`, `source_column`, `quote`, `parse_confidence`, `status: "ok"`, and `compare_to` +or `pe_missing`. + +Of the harvest rules (`sources/harvest-uk-ab2025-2026-09-24/README.md`), these apply: verbatim +values (1), registered vocabulary, named baselines (7) and unreadable primaries (8). The fields +specific to the Autumn Budget harvest (`measure_key`, `fiscal_event`) do not. + +- `value_raw` exactly as printed (`"-800m"`, `"£0.80bn"`, `"(0.4)"`, `"−4.2pp"`), `value` + normalised under a normalisation whose scale matches the printed one (`NORMALIZATIONS` in + `scorecard_db/comparables.py`; `percent as printed` for a figure printed with `%` or `pp`), + and a `quote` that prints the value as a whole number with its sign (a dash between two + numbers, as in "£3-3.4bn", is a range, not a sign). One number per claim. A printed range is + two claims, `value_kind` `range_low` and `range_high`, sharing a `conditions.range_group`; a + one-sided figure ("over £1 billion") is `lower_bound` or `upper_bound`, and like a `cumulative` + total it is set beside an estimate, never compared with one. +- Never re-sign a value. Put its convention in `conditions.sign_convention`, starting + `positive =` and saying one thing, in words `signs.py` reads: `positive = yield to the + Exchequer`, `positive = cost to the Exchequer`, `positive = gain to households`, + `positive = higher government revenue`, `positive = higher government expenditure`. It does + not read `negative = ...`, a two-sided sentence, or a phrase turned around + (`positive = yield forgone`). +- The same identity fields as an estimate: `period`, `time_basis`, `conditions.basis`, + `conditions.program`, `conditions.prices`. A producer that does not say whether its figure + allows for behaviour gets basis `unstated`, and one that does not say whether it is in cash or + today's prices gets prices `unstated`: such a claim is never the same quantity as an estimate, + so it is set beside one, not compared. Check each producer's year label: CenTax labels tax + years by their later year. +- `attribution`: `own` when the producer computed it, `restated` when it repeats another's + figure. Only `own` estimates are compared. + +### 5. Compare + +- A claim on the same quantity as an estimate (same metric, unit, period, time basis, basis and + programme, the producer's own, not cumulative, same baseline) names it in + `compare_to: {"pe_estimate": ""}`. If the two score different policy worlds, + `compare_to.across_worlds` gives the reason the quantities still match and its basis, and the + comparison is recorded as `constructed`. +- A claim with an estimate near it (the same period, and the same metric and unit where the + claim names them) that is a different quantity (net of behaviour beside static, a fiscal year + beside a calendar year) names that estimate in `nearest_estimate`, gives `pe_missing`, and + gets an annotation with status `concept_mismatch` (or `pe_gap`) and a basis. A claim on the + same quantity must be compared. The validator detects matches from the declared fields; + it cannot establish that those fields accurately describe the published quantities. +- Any other claim gives `pe_missing`. + +**Never do these to keep a gap from being compared**, even when the validator cannot tell: + +- leave a published figure out of `pe_estimates`; +- set a claim's metric or unit to null when an estimate covers the quantity; +- stage a figure as `approx_chart_reading` when the producer prints it; +- choose a period, time basis, basis or programme to avoid a match; +- label an explained miss `concept_mismatch`: a miss on the same quantity is `comparable` or + `constructed`, with the cause. + +### 6. Explain every gap over the threshold (`annotations.json`) + +```bash +PYTHONPATH=. uv run python -m scorecard_db.comparables check sources/analysis-- +PYTHONPATH=. uv run python -m scorecard_db.comparables report analysis-- --since +``` + +`check` lists every error, and the unexplained gaps once nothing else is wrong. `report` lists +over-threshold rows for valid records and tolerates the missing-annotation error +`no annotation explains it` (printing it on stderr). It still exits 1 with empty stdout when +an existing annotation has `explains_gap: false` and must explain a gap. Preserve that +annotation and the `check` diagnostics in the draft PR; assemble the gap table from the staged +claims and estimates if `report` refuses it. Do not mark the gap explained to obtain a report. +Other validation errors also block the report. +A gap over `gap_threshold` (`data/comparables/policy.json`, 25% by +default), a sign disagreement, or a nonzero estimate against a published zero needs an +annotation with an `id`, the `claims`, a `status`, `explains_gap: true`, a `cause`, and a +`basis`: a GitHub issue or pull, engine metadata pinned to a commit, or a document or +diagnostic URL (or a file in the scorecard). A range is over only when our figure lies outside +it by more than the threshold. + +Without a basis you have read, the gap is not explained. Find the cause first (most often our +data or rules: file the issue), or hold the analysis. A number in the cause must come from a +real run: never scale or interpolate other runs to argue the gap away. + +A gap that a search after publication finds (`backfill`, `reharvest`, `prepromotion`) may be +`pending` while the cause is found: open an issue for the analysis's owner with the producer, +the figures, the gap and what you checked, put it in `tracking_issue`, and write the cause as +"Unexplained; tracked in " with the issue as the basis. A pending gap blocks every push +of the analysis until it is replaced by the cause. Nothing found before publication can be +pending. + +### 7. Land it + +If `check` still fails, commit the record files, diagnostics and gap table and open a draft +scorecard PR. Defer index/database/feed generation, merge and publication until the errors +are resolved: the index builder also refuses unexplained gaps. Keep the unexplained +annotation intact. Once `check` passes, run: + +```bash +PYTHONPATH=. uv run python -m scorecard_db.comparables index +rm -f data/scorecard.db && PYTHONPATH=. uv run python -m scorecard_db.build_db data/scorecard.db +PYTHONPATH=. uv run python -m scorecard_db.export_populations data/scorecard.db +for f in lanes.json populations.json comparison.json exhibits.json; do cp data/$f app/public/data/$f; done +PYTHONPATH=. uv run --python 3.12 --with pytest --with hypothesis --with pypdf --with pyyaml --with numpy --with psutil --with rfc8785==0.1.4 --with jsonschema pytest tests/ -q +``` + +The suite runs on Python 3.12, as the scorecard's CI does (one of its tests needs a 3.12 +interpreter). Commit everything the build rewrites: `git status` must be clean afterwards, or CI's drift check +fails. Open the scorecard PR, have it reviewed with the checklist below, and merge it on its +gates. For apps and posts, print the registry block and paste it into the apps.json or posts.json entry: + +```bash +PYTHONPATH=. uv run python -m scorecard_db.comparables record analysis-- +``` + +It cites the record's latest search (`--search-date` and `--search-kind` pick another). A +revised analysis returning to the site cites the new search its results were compared in, and +gives the entry its new listing date. + +For trackers, link the merged record and gap table from the bill-review PR instead of adding +an app-v2 entry. Publication is not complete until its actual date is recorded below. + +### Reviewer checklist + +A reviewer opens the record and the primaries, not just the validator's output: + +- every figure on the page, or in the results file the page reads, is in `pe_estimates`; +- each `pe_missing` reason is true, and each `nearest_estimate` mismatch is real; +- each annotation's basis, once opened, supports its cause; +- a sample of `none_found` results is searched again, and every `found_not_staged` note is + checked; +- the manifest's hashes are recomputed from the URLs; +- each staged value and quote is found at its locator in the primary. + +## United States + +- **Sign conventions.** Use `positive = higher government revenue` for `revenue_change` + (CBO and JCT print revenue increases as positive), `positive = increase in borrowing` or + `positive = cost to the Exchequer` for a deficit increase stated against revenue, and + `positive = higher government expenditure` for `benefit_cost_change`. A convention that names the opposite quantity ("positive = increase in federal revenue" for a + spending metric) is refused. An empty convention never orients a change. +- **Fiscal and calendar years.** CBO and JCT report federal fiscal years (October to September), + with `period` the year the fiscal year ends and `time_basis` `fiscal_year`. PolicyEngine's + federal estimates are calendar years. The two are near but not the same quantity: name the + calendar-year estimate in `nearest_estimate` with a `concept_mismatch` annotation, unless the + analysis publishes a fiscal-year figure to compare. +- **Budget windows.** A ten-year total is `value_kind: cumulative`. The validator does not + compare a cumulative figure with anything yet, a window total of the same years included: stage + it uncompared, beside the nearest single-year estimate with a `concept_mismatch` annotation + saying it is a window total, and never restate it as an annual figure. +- **State fiscal notes** are `state_fiscal_note` (`_fiscal_note` as `source`), searched + for state and local analyses. + +## Running the search as a background job + +`references/search-brief.md` is a self-contained brief with placeholders to fill. It works for +any agent with web search, a shell and `gh` authenticated to push a branch and open a draft PR +in the scorecard. The job writes, downloads primaries and opens a PR, so it needs a writable +workspace and network access. On PolicyEngine's internal Subfleet lanes: + +```bash +subfleet run --task research --tier standard -s workspace-write -C -p brief.md -o out.md +``` + +The brief has the job open a draft scorecard PR, even when a gap is unexplained, and report the +producer table and every gap. Review the PR with the checklist above before merging. A job's +report is a claim, not proof. + +## Before a push + +Before an analysis is featured in a newsletter, a social campaign, a webinar or a press push, its +search must be fresh: + +- a search dated on the push date or in the `promotion_window_days` before it + (`data/comparables/policy.json`, 7 by default, counting both ends); +- that compared the results files the analysis reads now; +- and no gap found by then still `pending`. + +```bash +PYTHONPATH=. uv run python -m scorecard_db.comparables fresh analysis-- --on +``` + +Run it in a checkout of the scorecard's `origin/main`, which is what the newsletters check +reads. Resolve the analysis and record regardless of whether the source is a canonical URL, +deployment URL, redirect, or local file, and whether it publishes fixed figures or is a +household calculator. Ask for the analysis identity if a local source does not identify it; +an unknown link or empty estimate list does not exempt an analysis. An analysis with no record +gets its first search: `prepublish` if not yet public, `backfill` if already published. For an +unpublished analysis, complete the prepublish workflow and publication metadata before public +promotion; `prepromotion` requires a known publication date. For an already published analysis +that is not fresh, append a search with `kind: prepromotion` and today's date that logs every +producer, stage what is new, and explain every gap over the threshold. A pending gap blocks the push: explain it, or leave +the analysis out. Once it is explained, the push may state the comparable beside our figure. +The same rule holds for a card added to policyengine.org's featured research banner, which +app-v2's gate checks on the day the change runs. + +## After publishing + +### Complete publication metadata + +As soon as the production URLs and actual first-publication date are known, update the +scorecard record's `deployments` and `published`. Keep earlier serving URLs in `deployments` +when an analysis is delisted or moves. The prepublish review may merge with no deployment and +`published: null`; the publication owner must land a follow-up scorecard PR with these fields, +regenerate the index, database and exported feeds using step 7, and merge it before listing or +promoting a new deployment. Carry this follow-up in the publication handoff until it merges. + +Verify every direct serving URL resolves to the merged scorecard record independently of +the app registry. In a newsletters checkout, with the scorecard checkout on `origin/main`, +run this completion check (substitute the index path, source and production URL): + +```bash +PYTHONPATH=src uv run python - /path/to/scorecard/data/comparables/index.json analysis-uk-example https://example.vercel.app/ <<'PY' +import json +import sys +from pathlib import Path +from newsletter_uploader.comparables import Registry, Routes, linked + +index = json.loads(Path(sys.argv[1]).read_text()) +registry = Registry([], [], index["entry_rules"]) +routes = Routes({"paths": [], "hosts": []}, registry, index) +targets, refused = linked(sys.argv[3], routes) +assert not refused, refused +assert ("record", sys.argv[2]) in targets, targets +print("Direct URL resolves to the scorecard record") +PY +``` + +This verifies recognition; separately run `fresh` for any promotion date. A newsletter CLI +exit of 0 alone cannot prove recognition because it also accepts unknown URLs. + +For trackers, assign the bill-review publication owner the follow-up before dispatch. Keep +`published: null` while the score is hidden in review, then set the actual date when the +bill-review merge publishes it and merge the regenerated scorecard metadata. Verify +`scorecard_db.comparables due --apps --posts --today ` +includes the tracker; records with no `published` date are skipped. Do not substitute dispatch +or planned publication dates for the actual date. + +For six months after an analysis is published, a PolicyEngine scheduled job searches again each +week (`kind: reharvest`) once the scorecard tooling is on main. A new comparable becomes a row in +the analysis's directory. A new gap over the threshold gets a `pending` annotation and an issue +for the owner (the app-v2 author). Close the issue by replacing `pending` with the cause. + +## Bill-tracker scores + +A bill scored through `/reform-pipeline`'s tracker route is reviewed before dispatch like any +other analysis. Its record has `registry.kind: tracker` (slug `-`, the tracker page +as `url`), and its bill-review PR carries the record's link and gap table. No registry lists +tracker bills, so app-v2 does not check them; the scorecard's re-harvest finds them by their +record's `published` date. Scaffold them via the app path in step 1 and finish the publication +metadata handoff above after the bill-review PR publishes the score. + +## Related + +- `policyengine-prior-scores`: benchmarks while an analysis is being built. Its scorekeepers + registry is the wider pool, but the producers this review must log are the scorecard's + `data/comparables/producers.json`. +- `policyengine-writing`: the publishing checklist this review is part of. +- `policyengine-tools`: registering a dashboard in apps.json. diff --git a/skills/policyengine-comparables-review/references/search-brief.md b/skills/policyengine-comparables-review/references/search-brief.md new file mode 100644 index 0000000..3fb2a16 --- /dev/null +++ b/skills/policyengine-comparables-review/references/search-brief.md @@ -0,0 +1,186 @@ +# Comparables search brief + +Fill every `{placeholder}`, delete the sections for the search kinds you are not running, and +pass the result as the job's prompt. The brief works for any agent with web search, a shell and +`gh` authenticated to push a branch and open a draft PR in PolicyEngine/policyengine-scorecard. +The job writes files, downloads primaries and opens a PR, so it needs a writable workspace and +network access. + +--- + +You are running the comparables review for one PolicyEngine analysis. Read +`docs/COMPARABLES.md` in this repository (PolicyEngine/policyengine-scorecard) first; it is the +contract. `python -m scorecard_db.comparables check` checks its structural rules, but trusts +declared identities, source content, causes, searches and hashes; a reviewer must verify them. +The worked example to copy is `sources/analysis-uk-childcare-100k-threshold/`. + +## The analysis + +- Title: {title} +- Page: {url} (kind: {app | post | tracker}; registry: {apps.json | posts.json | none for tracker}, + slug `{slug}`, country `{country}`; tracker slug: `-`) +- Other serving URLs: {known https deployment URLs, or "not deployed yet"} +- What it estimates: {one or two sentences: the reform and the quantities it publishes} +- Its results file: `{repo}` at commit `{commit}`, path `{results path}`. Pin every estimate to + this file and its SHA-256. (A household calculator with no fixed estimate: say so here.) +- Published: {YYYY-MM-DD, or "not yet"} +- Publication metadata handoff: {owner and publication/bill-review PR that will supply actual + URLs and publication date once public; for an already published analysis, complete them now} +- Owner (app-v2 author key): {author} +- Scope: {uk_wide | great_britain | england | scotland | wales | northern_ireland | local | + federal | state} +- Search terms to start from: {terms} +- Search date: {YYYY-MM-DD}. Search kind: {prepublish | backfill | reharvest | prepromotion} +- Push date (prepromotion only): {YYYY-MM-DD} + +## Do this + +Work in a branch `comparables/{country}-{slug}` off `origin/main` of this repository. Run every +command as `PYTHONPATH=. uv run python -m ...`. + +### For a first search (prepublish or backfill) + +1. `scorecard_db.comparables scaffold --country {country} --kind {app|post} --slug {slug} --today {date}`. + For a tracker, scaffold with `--kind app`, then set `analysis.json`'s `registry.kind` to + `tracker` and `url` to the actual tracker page; the CLI rejects `--kind tracker`. + Set the search's `kind` to `{kind}` and `by` to this job's id. A prepublish search is dated + on or before publication; a backfill on or after it, with `published` set in + `analysis.json`. Keep `published: null` until actual publication, including while a tracker + score is hidden in bill review; dispatch is not publication. +2. Fill `analysis.json`: title, url, owner, scope, subject, search terms, the policy worlds (each + with a `policy` slug and a `description`), and one `pe_estimates` entry per figure the page + publishes that a producer could also estimate, read from the results file (never from the + page). Leave none out. Each has the identity fields (`metric`, `unit_concept`, `period`, + `time_basis`, `conditions.basis`, `conditions.program`, and `conditions.fy` for a fiscal + year), `value_raw` as the file prints it, a `normalization`, and `provenance`. State each in + PolicyEngine's orientation for its metric (`scorecard_db/signs.py` `ORIENTATION`), with + `sign: -1` and `sign_basis` if the file prints it the other way. A household calculator + declares `"calculator": {"reason": "..."}`, pins no estimates, and records + `results_pins: []`. + Add every known serving URL to `deployments` and retain earlier URLs when an analysis moves + or leaves the app registry. Empty deployments validate but leave unregistered hosts + unrecognized by newsletter checks. Set `published` to the actual date if already public. +3. Set the search's `results_pins` to what `scorecard_db.comparables pins + analysis-{country}-{slug}` prints. + +### For a re-harvest or a search before a push (reharvest or prepromotion) + +1. If the analysis's results file has changed since the last search, first move the old + estimates to `superseded_estimates` with `superseded_on`, and pin the new ones. +2. Then append a search to `sources/analysis-{country}-{slug}/search_log.json`: `kind: {kind}`, + today's date (after publication and after the first search), `results_pins` from + `scorecard_db.comparables pins` (run after step 1, so they name the new results), and one + entry for every producer on `data/comparables/producers.json` for {country} added on or + before today. +3. Stage only what earlier searches did not find (check `claims_staged.jsonl` and the earlier + searches' notes first). + +### Every search + +4. Search every producer in the search entry, in list order. Run its `search_hints` combined + with the search terms, and check the places `where_to_look` names as well as search results. + Log `found`, `none_found`, `found_not_staged` or `not_applicable` with the queries run and + pages checked. `not_applicable` only when the producer's `applies_to` excludes the scope. + Never skip `open_search` or `academic`. A listed producer's estimate is logged under its own + entry, never under `open_search`. +5. Trace every figure to its producer's own publication. Download each primary into a new empty + directory outside this repository, `shasum -a 256` it, and add a `manifest.jsonl` line. A + figure whose primary you cannot read (press-only, FOI with no published response) is + `found_not_staged` with a note saying what it is and where it was seen, never a claim. +6. Stage each number as one line of `claims_staged.jsonl`, with the fields the worked example + uses. Of the harvest rules (`sources/harvest-uk-ab2025-2026-09-24/README.md`), verbatim + values (1), registered vocabulary, named baselines (7) and unreadable primaries (8) apply; + `measure_key` and `fiscal_event` do not. + - `value_raw` as printed, a registered `normalization` whose scale matches the printed one, + and a `quote` that prints the value as a whole number with its sign. A printed range is two + claims (`range_low`, `range_high`) sharing a `conditions.range_group`. + - The sign convention in `conditions.sign_convention`, starting `positive =` and saying one + thing, in words `signs.py` reads (the skill lists them, with US phrasings). + - The identity fields (`conditions.basis` one word, with any account in `basis_note`; + `conditions.program` a slug, with any description in `program_note`; `conditions.prices: + real` for a base year's prices; `unstated` basis or prices when the producer does not say), + the world it scores, attribution, benchmark class, and the publication's locator. + - A one-sided figure ("over £1 billion") is `lower_bound` or `upper_bound`, set beside an + estimate, never compared. +7. Link each claim to the PolicyEngine estimate it matches (`compare_to`: the same quantity, + identity and baseline; `across_worlds` with a reason and basis if the worlds differ). A claim + near an estimate that is a different quantity (net beside static, a fiscal year beside a + calendar year) names it in `nearest_estimate`, gives `pe_missing`, and gets a + `concept_mismatch` annotation with a basis. Any other claim gives `pe_missing`. + + **Never** do these to keep a gap from being compared: leave a published figure out of + `pe_estimates`; set a metric or unit to null when an estimate covers the quantity; stage a + printed figure as `approx_chart_reading`; choose a period, time basis, basis or programme to + avoid a match; label an explained miss `concept_mismatch`. +8. Run `scorecard_db.comparables check sources/analysis-{country}-{slug}` until it passes, apart + from unexplained gaps, and `scorecard_db.comparables report analysis-{country}-{slug} --since + {date}`. The report tolerates missing annotations (`no annotation explains it`), but exits 1 + with empty stdout for an existing annotation that must explain a gap and has + `explains_gap: false`. Preserve that annotation and `check` diagnostics in the draft PR; + assemble the gap table from staged claims and estimates if needed. Never claim a gap is + explained just to obtain a report. For every gap over the threshold, find the cause: read the analysis's code and data, + the producer's method, and open issues (microcosm, policyengine-{country}, the analysis repo). + Write an annotation that cites a basis you read. If you cannot find one: + - prepublish: leave the gap unexplained. Do not invent a cause. CI stays red, which is + correct: the analysis is not ready to publish. + - backfill, reharvest or prepromotion: add a `pending` annotation, open an issue for the owner + on the analysis's repository with the producer, the figures, the gap and what you checked, + put it in `tracking_issue`, and write the cause as "Unexplained; tracked in " with the + issue as the basis. For prepromotion, the push stays blocked until the cause is written. +9. Land it: + + If `check` still fails, commit the record files, gap table and diagnostics and open the + source-only draft PR described below. Skip index/database/feed generation: the index + builder rejects unexplained gaps too. Preserve the unexplained annotation. Do not merge + or proceed to publication while errors remain. When `check` passes, run: + + ```bash + PYTHONPATH=. uv run python -m scorecard_db.comparables index + rm -f data/scorecard.db && PYTHONPATH=. uv run python -m scorecard_db.build_db data/scorecard.db + PYTHONPATH=. uv run python -m scorecard_db.export_populations data/scorecard.db + for f in lanes.json populations.json comparison.json exhibits.json; do cp data/$f app/public/data/$f; done + PYTHONPATH=. uv run --python 3.12 --with pytest --with hypothesis --with pypdf --with pyyaml --with numpy --with psutil --with rfc8785==0.1.4 --with jsonschema pytest tests/test_comparables.py tests/test_export_populations.py tests/test_harvest_ingest.py -q + ``` + + Commit everything the build rewrites (`git status` clean), push, and open a draft PR titled + "Comparables for {title}", even when a gap is unexplained; its body states every gap and its + status. + +### Complete the publication handoff + +For apps/posts, the caller adds the reviewed block to app-v2 after the scorecard PR merges. +For trackers, carry the merged record and gap table into the bill-review PR; do not add an +app-v2 registry entry. If production URLs or the actual publication date are still unknown, +report the assigned follow-up owner and publication PR explicitly. Once known, that owner +must update `deployments` and `published`, repeat the index/database/feed builds above and +merge the scorecard follow-up before listing or promoting a new deployment. Keep this +handoff open until it merges. For trackers, update the date on actual publication by the +bill-review merge, then verify `scorecard_db.comparables due --apps --posts + --today ` includes the tracker; a null date is skipped. + +The caller must run the skill's "Complete publication metadata" recognition check for every +direct URL against the merged scorecard index with an empty app registry/route map. The +newsletter resolver must explicitly return this record; exit 0 from the newsletter CLI can +also mean it recognized no analysis. Redirect aliases need manual resolution because the +generated route map drops absolute destinations. Run `fresh` separately on the intended push +date, including for calculators and deployment/local-file sources resolved to this analysis. + +## Report + +Write to the output file: + +- the PR URL and head commit; +- one line per producer: result, claims staged, and for `found_not_staged` what and why; +- every compared claim: producer, value, the PolicyEngine estimate, the gap, and the annotation + (or "unexplained"); +- every claim set beside a nearest estimate, and why; +- the issues opened; +- metadata completed (`deployments`, actual `published`) or the outstanding publication + handoff owner/PR; for trackers, the bill-review record/gap-table handoff and date update; +- for prepromotion, what `scorecard_db.comparables fresh analysis-{country}-{slug} --on {push + date}` prints; +- anything you could not check, stated plainly. + +Do not merge. Do not edit policyengine-app-v2. A person reviews the PR with the skill's reviewer +checklist first: every published figure pinned, each `pe_missing` true, each basis supporting +its cause, `none_found` spot-checked, hashes recomputed. diff --git a/skills/policyengine-content/SKILL.md b/skills/policyengine-content/SKILL.md index 6784840..05ec1cc 100644 --- a/skills/policyengine-content/SKILL.md +++ b/skills/policyengine-content/SKILL.md @@ -25,6 +25,24 @@ runs through a command and agent. - The **`neutrality-reviewer`** agent then checks the generated copy for advocacy, speculation, or one-sided framing before anything is published. +Before promoting any PolicyEngine analysis, including a household calculator with no fixed +estimates, its comparables search must be fresh (policyengine-comparables-review, "Before a push"). +The orchestrator resolves every analysis being promoted to its scorecard record whether the input +is a canonical URL, deployment URL or local file. It asks for the analysis identity if the source +does not identify it, and for the intended push date if unknown. An unrecognized link or missing +record does not exempt an analysis. + +Fetch scorecard `origin` and run `PYTHONPATH=. uv run python -m scorecard_db.comparables fresh +analysis-- --on ` in a checkout of the updated `origin/main` before generating +assets. If the record is absent, complete its first search as `prepublish` if the analysis is not +yet public, or `backfill` if it is already published. For any unpublished analysis, complete the +prepublish review and publication metadata handoff before public promotion; do not append another +`prepublish` or a `prepromotion` search while `published` is null. After actual publication, +append `prepromotion` only if freshness needs renewal. Merge the review and metadata into scorecard +main, fetch and rerun `fresh`; pending gaps block the push. Fetch and rerun it immediately before +the push using the actual push date. This explicit identity check is necessary because automatic +link checks do not establish complete coverage. + Do not hand-render assets when the command exists — invoke `/generate-content` and let the agents own the mechanics. Follow the **policyengine-writing** skill for the copy's tone and neutrality. diff --git a/skills/policyengine-prior-scores/SKILL.md b/skills/policyengine-prior-scores/SKILL.md index 5f457e0..f4a6502 100644 --- a/skills/policyengine-prior-scores/SKILL.md +++ b/skills/policyengine-prior-scores/SKILL.md @@ -168,5 +168,7 @@ the caveats above. ## Related skills - `policyengine-research-lookup` — broader blog/post discovery beyond scored reforms. +- `policyengine-comparables-review` — the required pre-publish search; it logs every producer + on the scorecard's list and records what it finds in policyengine-scorecard. - `policyengine` — actually running the microsim to compare against these anchors. - `policyengine-calibration-diagnostics` — when a new run mismatches a prior, diagnose why. diff --git a/skills/policyengine-tools/SKILL.md b/skills/policyengine-tools/SKILL.md index 178dba6..449e81e 100644 --- a/skills/policyengine-tools/SKILL.md +++ b/skills/policyengine-tools/SKILL.md @@ -106,6 +106,10 @@ Mandatory rules: ### Deployment gotchas +- **A research listing needs a comparables record.** A new apps.json entry shown with research + or tagged `policy` must carry a `comparables` block naming a complete scorecard record, and + every app route must belong to such an entry, or app-v2's CI fails + (policyengine-comparables-review). Only entries live before the gate are exempt. - **Vercel auto-assigns a non-deterministic URL suffix** (e.g. `marriage-zeta-beryl.vercel.app`). Read the actual URL after first deploy and hardcode *that* in the host rewrite / apps.json — never guess it. diff --git a/skills/policyengine-writing/SKILL.md b/skills/policyengine-writing/SKILL.md index 10d79b7..90e08c5 100644 --- a/skills/policyengine-writing/SKILL.md +++ b/skills/policyengine-writing/SKILL.md @@ -759,6 +759,13 @@ Before publishing, verify: - [ ] Public copy describes the policy, the result, and what the reader can do; cut production, process, and work-in-progress notes - [ ] Remove self-referential clauses, empty contrasts, and unintroduced names or roles - [ ] Check mechanism claims against the code, test, or primary source +- [ ] Before publishing an estimate, run the comparables review and explain every gap over + the threshold (the `policyengine-comparables-review` skill in the analysis-tools bundle, or + `docs/COMPARABLES.md` in PolicyEngine/policyengine-scorecard); app-v2 CI requires its record +- [ ] Before a newsletter, social campaign, webinar or press push features an analysis, its + comparables search is fresh (on the push date or within the promotion window before it, + against the current results, no gap pending); the newsletters repo's CI checks every + analysis an edition links ## Resources diff --git a/targets/claude/agents/content/content-orchestrator.md b/targets/claude/agents/content/content-orchestrator.md index cb10202..f934196 100644 --- a/targets/claude/agents/content/content-orchestrator.md +++ b/targets/claude/agents/content/content-orchestrator.md @@ -24,6 +24,29 @@ You orchestrate the generation of marketing content from PolicyEngine blog posts - Extract: title, subtitle, key quotes, author, main points - Identify the primary announcement or finding +1b. **Check comparables are fresh (required before any push)** + - Determine whether the source concerns a PolicyEngine analysis, whether the input is a + policyengine.org URL, another deployment URL or a local file. Apply this check to every + analysis the content promotes, including household calculators with no fixed estimates. + - Resolve each analysis to its scorecard record and source directory (`analysis--`) from its canonical page, + deployment metadata or source repository. Do not treat an unrecognized URL or an absent + record as an exemption. Ask for the analysis identity if the source, especially a local + file, does not identify it; ask for the intended push date if it is unknown. + - Fetch PolicyEngine/policyengine-scorecard's `origin` and use a checkout of the updated + `origin/main`. Run `PYTHONPATH=. uv run python -m scorecard_db.comparables fresh + analysis-- --on ` for each resolved record before generating assets, + then fetch and rerun it immediately before the posts go out using their actual push date. + - If the record is absent or not fresh, stop and say so: follow the + `policyengine-comparables-review` skill. For an absent record, create its first search as + `prepublish` if the analysis is not yet public, or `backfill` if it is already published. + For any unpublished analysis, complete the prepublish review and publication metadata + handoff before public promotion; do not append another `prepublish` or a `prepromotion` + search while `published` is null. After actual publication, append `prepromotion` only + if freshness needs renewal. Merge the review and metadata into scorecard main, then fetch + and rerun `fresh`. A pending gap blocks the push until its cause is written; never write + copy around it. Resolve the analysis explicitly rather than relying on automatic link + coverage. + 2. **Gather missing information** - Use AskUserQuestion to get any missing required info: - Pull quote for social image diff --git a/targets/claude/agents/report-logger.md b/targets/claude/agents/report-logger.md index d6a0301..0734b03 100644 --- a/targets/claude/agents/report-logger.md +++ b/targets/claude/agents/report-logger.md @@ -349,8 +349,9 @@ Skip this destination unless `--log-to tracker` is explicitly passed AND credent `--log-to draft:/` opens a PR with the report rendered as a blog-post-style markdown file. The canonical PolicyEngine research-post target is `policyengine-app-v2` at `app/src/data/posts/articles/{slug}.md` (also register the slug in `app/src/data/posts/posts.json`). Note: `policyengine-app` v1 is archived — never target it. 1. Use the `policyengine-writing` skill for tone polish (the raw report uses neutral-mechanical language; a draft needs lead, sub-heads, and a TL;DR). -2. `gh pr create --draft` against the target repo, branch named `analyze-policy/{policy_id}`. -3. Surface the PR URL in the output. +2. Before the post can leave draft, it needs a comparables review: app-v2's CI requires a `comparables` block on every new research post. Note in the PR body that `/comparables-review` (the `policyengine-comparables-review` skill) is still to run, unless it already has. +3. `gh pr create --draft` against the target repo, branch named `analyze-policy/{policy_id}`. +4. Surface the PR URL in the output. ## Output diff --git a/targets/claude/commands/comparables-review.md b/targets/claude/commands/comparables-review.md new file mode 100644 index 0000000..527cc93 --- /dev/null +++ b/targets/claude/commands/comparables-review.md @@ -0,0 +1,63 @@ +--- +description: Run the pre-publish comparables review for a dashboard, research post or tracker score and record it in the scorecard +--- + +# Comparables review: $ARGUMENTS + +Required before a dashboard or research post is published or listed in `apps.json` or +`posts.json`, before a tracker score is dispatched through `/reform-pipeline`, and before a +newsletter, social campaign, webinar or press push features an analysis already published. +Calculators with no fixed estimates are included. app-v2 checks registry entries subject to +its gate; newsletter CI checks recognized analysis links. Route and redirect discovery has +gaps documented in the skill, so green CI does not replace resolving the analysis yourself. + +Load the `policyengine-comparables-review` skill and follow it. In short: + +1. Identify the analysis from `$ARGUMENTS`: its slug, country, registry kind (`app`, `post` or + `tracker`), page URL, other serving URLs, the repository and commit its page builds from, + and the results file. Resolve deployment URLs, redirects and local files to that identity; + ask for the identity if a local source does not provide it. Only a post + that is plainly not an analysis (an event, an announcement) may skip the review by declaring + `{"status": "no_estimate", "reason": "..."}`; an app never may. A household calculator is + searched, and declares `calculator` in its record instead of pinning estimates. +2. In a PolicyEngine/policyengine-scorecard worktree off `origin/main`, scaffold + `sources/analysis--/` (or append a search to it), pin every figure the page + publishes, and search every producer the skeleton lists, logging each result. For trackers, + scaffold with `--kind app --slug -`, then set `registry.kind: tracker` and the + actual tracker URL: `scaffold` only accepts app/post. Keep `published: null` until public. +3. Stage each estimate found, compare it with the figure it matches or set it beside the + nearest one with a stated mismatch, and run `python -m scorecard_db.comparables check`. + Explain every gap over the threshold with a cited cause. Never hide a gap by leaving a + figure out or relabelling a claim. +4. If `check` fails, commit the source record, gap table and diagnostics in a draft scorecard + PR, preserving unexplained annotations. Defer index/database/feed builds, merge and + publication until the errors are resolved. When it passes, build, commit everything the + build rewrites and open the scorecard PR; have it reviewed with the skill's checklist; + the validator cannot establish source truth. For apps/posts, + once it merges, print the block with + `python -m scorecard_db.comparables record analysis--` and add it to the + app-v2 entry. For trackers, link the record and gap table in the bill-review PR instead. +5. Follow the skill's "Complete publication metadata" step: once deployed, land the serving + URLs in `deployments` and the actual first-publication date in `published`, rebuild the + index and feeds, merge the follow-up and verify each direct URL resolves to this record. + For trackers, assign the date update to the bill-review publication owner before dispatch + and complete it when the bill-review merge publishes the score, then verify `due` finds it. + +To run the search as a background job instead, fill the skill's `references/search-brief.md` +and dispatch it to an agent with web search, a writable workspace and `gh` push access. Review +its PR before merging. + +## Arguments + +`$ARGUMENTS`: the dashboard, post, calculator or tracker score (its URL, slug, repository or +local source plus analysis identity), plus `--kind +prepublish|backfill|reharvest|prepromotion` (default `prepublish`). Use `prepromotion` +before a newsletter, social campaign, webinar or press push features an analysis that is +already published: the push needs a search on the push date or in the scorecard's promotion +window before it, against the results the page reads now, with no gap pending +(`python -m scorecard_db.comparables fresh --on `). + +## Related + +- `/prior-scores`: benchmarking while the analysis is being built. +- `/deploy-dashboard`: Step 5c runs this review before the apps.json entry. diff --git a/targets/claude/commands/create-dashboard.md b/targets/claude/commands/create-dashboard.md index f8db030..75f8830 100644 --- a/targets/claude/commands/create-dashboard.md +++ b/targets/claude/commands/create-dashboard.md @@ -562,7 +562,8 @@ make build # Production build ### Next steps -After reviewing locally, commit and push. Then use /deploy-dashboard to deploy. +After reviewing locally, commit and push. Then use /deploy-dashboard to deploy. It runs +/comparables-review before the dashboard is public: app-v2's CI requires its record. ``` ``` diff --git a/targets/claude/commands/deploy-dashboard.md b/targets/claude/commands/deploy-dashboard.md index 575cfba..c65e92f 100644 --- a/targets/claude/commands/deploy-dashboard.md +++ b/targets/claude/commands/deploy-dashboard.md @@ -48,12 +48,12 @@ Extract: - `dashboard.zone_path` — for multi-zone host rewrite verification - `data_pattern` — determines if Modal deploy is needed (`custom-modal` needs it; `precomputed`, `precomputed-csv`, and `policyengine-api` do not) - `tech_stack.framework` — should be `react-nextjs` (env var prefix: `NEXT_PUBLIC_*`) -- `embedding.register_in_apps_json` — determines if apps.json update is needed +- `embedding.register_in_apps_json` — whether the dashboard is listed with research (`displayWithResearch`); its apps.json entry is needed either way - `embedding.slug` — the URL slug for policyengine.org ### 2a. Pre-flight: check for existing host rewrites (informational) -If this dashboard has deployed before, the host (`policyengine-app-v2/website/next.config.ts`) may already have rewrites pointing at it. Check so we know whether Step 5 needs to add them or they're already in place. +If this dashboard has deployed before, the host may already route to it (`policyengine-app-v2/website/src/data/appZoneRoutes.ts`, which `website/next.config.ts` turns into rewrites). Check so we know whether Step 5 needs to add the route or it is already in place. ```bash # Clone or pull the host repo @@ -167,6 +167,10 @@ vercel env add NEXT_PUBLIC_API_URL production # Enter: https://policyengine--DASHBOARD_NAME-calculate.modal.run ``` +## Before Step 4: comparables review + +Run `/comparables-review` now, before the frontend deploy and the host wiring, whatever `embedding.register_in_apps_json` says. Every dashboard needs a research entry with a `comparables` block naming a complete record in PolicyEngine/policyengine-scorecard; the listing flag does not exempt it. App-v2 checks registered entries and proxy routes it identifies as pages, with the discovery limits described in Step 5. The review searches every producer for external estimates of the dashboard's figures, records them in the scorecard and explains any gap over the threshold (`policyengine-comparables-review` skill). Merge the scorecard PR before deploying the frontend. + ## Step 4: Deploy Frontend to Vercel See the `policyengine-tools` skill ("Deployment gotchas") for the full Vercel deployment reference. @@ -193,14 +197,22 @@ curl -s -o /dev/null -w "%{http_code}" https://VERCEL_PRODUCTION_URL/ **IMPORTANT:** Use the auto-assigned Vercel production URL, not a custom alias. Custom aliases may have deployment protection issues. +### Complete the scorecard publication metadata + +After the first public deployment, return to the scorecard record before opening the host-wiring PR or promoting the direct URL. Add each full HTTPS production URL to `analysis.json`'s `deployments`, including the path where the analysis is served, and set `published` to the actual date the analysis first became public. Preserve an existing earlier publication date and any other deployment URL that still serves the analysis. + +Follow **Complete publication metadata** in the `policyengine-comparables-review` skill: regenerate the comparables index, database, exported feeds and their `app/public/data` copies, run the scorecard checks, and commit and merge the follow-up PR. Run that section's explicit direct-URL recognition check against the merged scorecard index; a successful newsletter CLI check alone does not prove it recognized the URL. Repeat this metadata update when production URLs change, including while the dashboard is delisted from app-v2. + ## Step 5: Wire the zone into the host (policyengine-app-v2) After the Vercel deploy in Step 4, we have a production URL. Now wire the zone into `policyengine-app-v2` so it's reachable at `policyengine.org/`. This is a PR to `PolicyEngine/policyengine-app-v2`. -Two parts, both in the same PR: +Two parts, both required, in the same PR: -1. **Host rewrites** — required for every multi-zone dashboard. These make the zone reachable at `policyengine.org/`. -2. **apps.json entry** — optional, only if `embedding.register_in_apps_json: true`. This is research-listing metadata; it does NOT affect routing. +1. **The zone route** in `website/src/data/appZoneRoutes.ts`, which makes the zone reachable at `policyengine.org/`. +2. **The apps.json entry** with its `comparables` block. The entry is required even when the dashboard is hidden from the research listing. + +App-v2 reads evaluated rewrites and checks external proxy sources it classifies as pages against research deployments and explicit exceptions. Its asset heuristic skips filename extensions, including `.html`, and the entire `/_zones/` prefix. A route such as `/uk/ungated-analysis.html` or `/_zones/ungated-analysis` can therefore pass without registration. Inspect every route that serves the dashboard and register the analysis regardless of that heuristic; a passing test does not establish complete route coverage. ### 5a. Create the PR branch @@ -216,43 +228,41 @@ git pull git checkout -b wire-DASHBOARD_NAME-zone ``` -### 5b. Add host rewrites to `website/next.config.ts` +### 5b. Add the zone route to `website/src/data/appZoneRoutes.ts` -Add these entries to `rewrites().beforeFiles` in `website/next.config.ts` (use `VERCEL_PRODUCTION_URL` captured from Step 4 — e.g. `my-dashboard-policy-engine.vercel.app`): +Add one entry to `appZoneRoutes` (use `VERCEL_PRODUCTION_URL` captured from Step 4, e.g. `my-dashboard-policy-engine.vercel.app`). `next.config.ts` turns it into the page rewrite and the `/:path*` rewrite for deeper pages: ```ts -// Path-mounted zone: two route rewrites preserve the basePath in the destination -{ source: '', destination: 'https://' }, -{ source: '/:path*', destination: 'https:///:path*' }, - -// Asset rewrite (required for zones using assetPrefix: '/_zones/') -{ source: '/_zones//:path*', destination: 'https:///_zones//:path*' }, +{ + source: "", + destination: "https://", +}, ``` -Skip the asset rewrite only if the zone does not set `assetPrefix`. For root-served zones (no `basePath`), keep the route rewrites but map destinations to the zone root instead: `` → `https://` and `/:path*` → `https:///:path*`. See the `policyengine-tools` skill for the two valid patterns: path-mounted and root-served. +For a root-served zone (no `basePath`), map the destination to the zone root: `"https://"`. If the zone sets `assetPrefix: '/_zones/'`, also add its asset rewrite (`/_zones//:path*` to the same path on the deployment) where the file's other asset routes are. See the `policyengine-tools` skill for the two valid patterns: path-mounted and root-served. -### 5c. Add apps.json entry (only if `embedding.register_in_apps_json: true`) +### 5c. Add the apps.json entry -This is research-listing metadata. The dashboard is reachable at `policyengine.org/` from Step 5b alone; this entry adds it to the research/interactives listing. - -Add to `app/src/data/apps/apps.json`: +Add to `app/src/data/apps/apps.json`, with the block that `python -m scorecard_db.comparables record analysis--` printed in the comparables review before Step 4: ```json { + "type": "iframe", "slug": "SLUG", "title": "TITLE", "description": "DESCRIPTION", - "path": "/COUNTRY/SLUG", + "source": "https://", "tags": ["COUNTRY", "policy", "interactives"], "countryId": "COUNTRY", "displayWithResearch": true, "image": "SLUG-cover.png", "date": "CURRENT_DATE 12:00:00", - "authors": ["AUTHOR_SLUG"] + "authors": ["AUTHOR_SLUG"], + "comparables": { "...": "the block /comparables-review printed" } } ``` -The `path` field is the multi-zone route on `policyengine.org` (the host rewrites from 5b make it reachable). Do not use `"type": "iframe"` — that's the legacy embedding model and does not apply to multi-zone deployments. +`source` is the deployment the zone route proxies to, as the existing zone dashboards record it. With `embedding.register_in_apps_json: false`, keep the entry but set `"displayWithResearch": false`: it leaves the research listing, but the `policy` tag still makes it a research entry, so it still needs the block. Then run `bun run scripts/build-analysis-routes.ts` in `website/` and commit the regenerated `website/src/data/analysisRoutes.json` (a test fails while it is stale). Use `AskUserQuestion` to gather required metadata: @@ -277,18 +287,21 @@ options: ### 5d. Commit and open the PR ```bash -git add website/next.config.ts app/src/data/apps/apps.json # second path only if 5c ran +git add website/src/data/appZoneRoutes.ts app/src/data/apps/apps.json website/src/data/analysisRoutes.json git commit -m "Wire DASHBOARD_NAME zone" git push -u origin wire-DASHBOARD_NAME-zone gh pr create --repo PolicyEngine/policyengine-app-v2 \ --title "Wire DASHBOARD_NAME zone" \ - --body "Adds host rewrites (and apps.json entry, if applicable) so DASHBOARD_NAME is reachable at policyengine.org/COUNTRY/SLUG. + --body "Adds the zone route and apps.json entry so DASHBOARD_NAME is reachable at policyengine.org/COUNTRY/SLUG. Zone URL: VERCEL_PRODUCTION_URL -Public path: /COUNTRY/SLUG" +Public path: /COUNTRY/SLUG +Comparables: the scorecard record named in the entry's block" ``` +The app-v2 PR's comparables gate checks, on the day it runs, that the block's search is no more than 30 days old, compared the results the dashboard reads now, and has no gap pending. If the PR waits longer than that, run a new search and update the block. + ## Step 6: Smoke Test Direct-URL checks can run immediately. Public-path checks require the Step 5 PR to merge first. @@ -318,7 +331,8 @@ Present deployment summary to the user: > - **API endpoint:** https://policyengine--DASHBOARD_NAME-calculate.modal.run > - **Modal environment:** SELECTED_ENV > - **Running while idle:** nothing, or each declared exception with its cost (from the README's "Running costs") -> - **Host-wiring PR:** PR_URL (adds rewrites, and apps.json entry if applicable) +> - **Host-wiring PR:** PR_URL (adds rewrites, the apps.json entry and the route map) +> - **Scorecard publication metadata PR:** PR_URL and merge status (production URLs and actual publication date) > > ### Verify > - [ ] Dashboard loads at the zone URL diff --git a/targets/claude/commands/generate-content.md b/targets/claude/commands/generate-content.md index 275e092..dc1ddca 100644 --- a/targets/claude/commands/generate-content.md +++ b/targets/claude/commands/generate-content.md @@ -24,6 +24,24 @@ Orchestrates the content-generation pipeline through two dedicated agents: 1. `content-orchestrator` (agent) — parses the source, generates localized variants (UK/US spelling, references, framing), and renders the outputs (social images via Chrome headless, social copy for LinkedIn/X). 2. `neutrality-reviewer` (agent) — reviews the generated copy for advocacy language, speculation, or one-sided framing before publish. See `targets/claude/agents/shared/neutrality-reviewer.md`. +Before generating assets, the orchestrator determines whether the source concerns a PolicyEngine +analysis, including household calculators with no fixed estimates. This applies to canonical URLs, +deployment URLs and local files. It resolves each analysis being promoted to its scorecard record; +if a local file or other source does not identify the analysis, it asks for that identity. An +unrecognized link or missing record does not exempt the analysis. + +For each record, fetch scorecard `origin` and run `PYTHONPATH=. uv run python -m +scorecard_db.comparables fresh analysis-- --on ` in a checkout of the updated +`origin/main`. The search must be within the promotion window, compare the current results and +leave no gap pending. If the record is absent, follow `policyengine-comparables-review` to create +its first search as `prepublish` if the analysis is not yet public, or `backfill` if it is already +published. For any unpublished analysis, complete the prepublish review and publication metadata +handoff before public promotion; do not append another `prepublish` or a `prepromotion` search +while `published` is null. After actual publication, run `/comparables-review --kind prepromotion` +only if freshness needs renewal. Merge the review and metadata into scorecard main, then fetch +and rerun `fresh`. Fetch and rerun it immediately before the push using the actual push date. +Resolve the analysis explicitly; automatic link checks do not establish complete coverage. + The command is the entry point; the agents own the mechanics. Loads `content-generation` skill for templates and `policyengine-writing` skill for the tone standard. ## Usage @@ -50,6 +68,10 @@ output/ ## Required information The command will prompt for any missing information: + +- **Analysis identity**: Canonical page, source repository or scorecard source directory (`analysis--`) for each + PolicyEngine analysis being promoted, including when the input is a local file +- **Push date**: Intended date the social posts will go out, used for comparables freshness - **Headline**: Main announcement headline - **Quote**: Pull quote for social image - **Quote attribution**: Name and title of person quoted diff --git a/targets/claude/commands/new-tool.md b/targets/claude/commands/new-tool.md index f2e95c7..da349cb 100644 --- a/targets/claude/commands/new-tool.md +++ b/targets/claude/commands/new-tool.md @@ -318,7 +318,9 @@ export async function pollStatus(jobId) { } ``` -## Step 5: Initialize git and deploy +## Step 5: Initialize git, review comparables, then deploy + +Nothing goes public before its comparables review: create the repository private, review, then publish. ```bash # Initialize repo @@ -326,9 +328,15 @@ git init git add -A git commit -m "Initial scaffold for TOOL_TITLE" -# Create GitHub repo under PolicyEngine org -gh repo create PolicyEngine/TOOL_NAME --public --source=. +# Create the GitHub repo under the PolicyEngine org, private until the review +gh repo create PolicyEngine/TOOL_NAME --private --source=. git push -u origin main +``` + +Run `/comparables-review` now (`policyengine-comparables-review` skill) against the results the tool reads at this commit, and merge its scorecard PR. A household calculator that publishes no fixed estimate is still searched; its record declares `calculator`. Then publish: + +```bash +gh repo edit PolicyEngine/TOOL_NAME --visibility public --accept-visibility-change-consequences # Deploy to Vercel vercel link --scope policy-engine @@ -348,6 +356,10 @@ vercel env add NEXT_PUBLIC_API_URL production vercel --prod --force --yes --scope policy-engine ``` +Capture the production URL after deployment. Before app-v2 registration or promotion, return to the scorecard record: add each full HTTPS production URL (including its analysis path) to `deployments` and set `published` to the actual date the analysis first became public. Preserve an earlier publication date and any other deployment URL that still serves the analysis. + +Follow **Complete publication metadata** in the `policyengine-comparables-review` skill: regenerate the index, database, exported feeds and their `app/public/data` copies, run the scorecard checks, and commit and merge this follow-up. Run that section's explicit direct-URL recognition check against the merged index; a successful newsletter CLI check alone does not prove recognition. Repeat the update when production URLs change, including after app-v2 delisting. + ## Step 6: Add pull-request CI Create `.github/workflows/ci.yml` so every new tool gets build validation before deploy. Replace `CHECKER_COMMIT_SHA` with the output of `gh api repos/PolicyEngine/policyengine-skills/commits/main --jq .sha`, so the cost checker's rules change only when the tool's maintainers move that commit: @@ -399,9 +411,13 @@ Every embeddable tool should have migration guardrails and the cost check, plus ## Step 7: Register in apps.json -Add the tool entry to `website/src/data/apps.json` in `policyengine-app-v2`. Use the auto-assigned Vercel production URL (not a custom alias). +The comparables review and publication metadata follow-up ran in Step 5. App-v2's CI requires a `comparables` block on every new research entry (an app shown with research or tagged `policy`). Every analysis tool needs this entry, including a calculator without fixed estimates. Proxy discovery skips filename extensions, including `.html`, and all `/_zones/` paths, so inspect every public route rather than treating a passing route test as complete coverage. Print the block with `python -m scorecard_db.comparables record analysis--` in the scorecard. Apps never declare `no_estimate`. If the registry PR is opened more than 30 days after that search, or the results have changed since, search again first. + +Then add the tool entry, with that block, to `app/src/data/apps/apps.json` in `policyengine-app-v2`. Use the auto-assigned Vercel production URL (not a custom alias). + +If the tool needs a proxy path on policyengine.org rather than a direct iframe source URL, add it to `website/src/data/appZoneRoutes.ts` in the same PR. -If the tool needs a rewrite/proxy path rather than a direct iframe source URL, update `website/next.config.ts` in the same repo as part of the launch. +After every research registry addition or `source` change, run `bun run scripts/build-analysis-routes.ts` in `website/` and commit `website/src/data/analysisRoutes.json`. This step is required for iframe-only tools too: each research app contributes its canonical path and external source host to the map without needing a proxy route. ## Step 8: Verify diff --git a/targets/claude/commands/reform-pipeline.md b/targets/claude/commands/reform-pipeline.md index ece190b..ac96b0f 100644 --- a/targets/claude/commands/reform-pipeline.md +++ b/targets/claude/commands/reform-pipeline.md @@ -122,6 +122,19 @@ Apply the standard routing rules (identical to the CRM publication router): - otherwise (routine state bill) → **bill tracker** - `--publish tracker|dashboard` overrides; `--publish none` stops here. +**Comparables review, both routes.** Every published score goes through the comparables +review (`/comparables-review`, the `policyengine-comparables-review` skill) before it is public: + +- the dashboard route through `/deploy-dashboard`, which runs it before frontend deployment + and host wiring, then completes the scorecard deployment URLs and actual publication date; +- the tracker route before dispatch: the scaffold CLI supports `app` and `post`, so scaffold + with `--kind app`, then set `registry.kind: tracker` in `analysis.json`, slug + `-`, and `url` to the tracker page. Leave `published: null` before publication. + Search every listed producer including `state_fiscal_note` and merge the scorecard PR. + Put the record's link, its gap table and the publication metadata handoff below in + `review_notes`. Tracker records do not need an app-v2 registry block. No registry lists + tracker bills; scorecard's `due` skips their records until `published` has an actual date. + **Publication preconditions (from Phase 2's `enactment_reconciliation`):** consult the block before dispatching — it says whether `baseline_json` is required, and whether this publication SUPERSEDES an earlier entry (a @@ -157,11 +170,17 @@ alone. Compose it from the analysis you just ran: was used), headline framing choice, what was scored vs held out of scope, model version, route choice. - **Validation record**: the /analyze-policy verdict with a link to the - archive entry; the benchmark table (each pre-registered external source, - its framing, estimate, and delta vs ours); this run's computed number vs + archive entry; the comparables record in policyengine-scorecard (every + producer searched, each comparable, its gap and any cause); the benchmark + table (each pre-registered external source, its framing, estimate, and + delta vs ours); this run's computed number vs the analysis number with the dataset difference explained; the dry-run rehearsal link; known limitations (e.g. unscored provisions and why). - **Notable findings** from the Phase 5.7 interrogation, when any exist. +- **Publication metadata handoff** for a tracker: identify the scorecard source and the + person or agent who will confirm publication and complete its date after the bill-review + PR merges and its publication workflow succeeds. Include this follow-up in the + bill-review checklist, so the handoff survives workflow completion. Example: state-legislative-tracker PR #215 (GA HB 463). ``` @@ -180,12 +199,26 @@ gh run watch --repo --exit-status gh run download --name publication-result --dir /tmp/pub-result ``` +For the tracker route, workflow completion leaves the entry hidden in `in_review`; the +bill-review PR's merge triggers publication. Confirm that the publication workflow succeeds +and the entry is public, then complete the handoff in a scorecard follow-up PR: set the +record's `published` to the actual public date, add any +additional live analysis URLs to `deployments`, and follow **Complete publication metadata** +in the `policyengine-comparables-review` skill to regenerate the index, database, feeds and +their `app/public/data` copies, run checks, and merge the update. Run scorecard `due` with +the current app/post registries and the publication day, and verify that it returns this +tracker source. Do not use dispatch or hidden-review creation as the publication date. +If publication awaits a later human merge, report this metadata handoff as outstanding and +carry it in the bill-review PR; the re-harvest discovery step remains incomplete until the +dated scorecard follow-up merges. A dry run leaves `published: null`. + ## Final output A summary the analyst can forward: the reform, the verdict with headline numbers, the calibration-check note, what was published where (PR link / repo link), and the remaining human step (merge the bill-review PR, or -`/deploy-dashboard`). +`/deploy-dashboard`). Include the scorecard publication metadata PR and its status, or the +outstanding tracker handoff that must run after actual publication. ## Boundaries