FixMap v0.9.0: complete evidence-first agent workflow

[aryamthecodebreaker]

Summary

• add a separate evidence-backed Impact Graph using direct imports, reverse dependents, routed tests, and repeated bounded Git co-change history
• add fixmap context for deterministic, budget-aware source ranges and fixmap graph for Mermaid or JSON exports
• add fixmap watch for working-tree drift feedback and impact recalculation, plus compact --format agent handoffs
• add fixmap benchmark for same-corpus BM25 / FixMap / Impact backtests on historical parent snapshots, with leak cohorts, Wilson intervals, maintained-source targets, raw cases, and hostile checkout-filter protection
• expose Plan, Context, Graph, Explain, Compare, Verify, and Doctor consistently through CLI, seven-tool MCP, Action/setup documentation, browser exports, website, and live demo
• publish the benchmark integrity correction and a frozen controlled agent-study protocol without claiming measured agent, time, or token savings
• add the approved 32-second 1080p agent-comparison launch film with original no-vocals music, UI sound cues, poster, and README GIF
• prepare the v0.9.0 metadata, changelog, release notes, social drafts, verification ledger, and post-release checklist

Evidence

• clean npm ci: 451 packages installed; zero vulnerabilities
• full npm run ci: 727 tests (42 Action, 247 CLI, 438 core), typecheck, lint, optimized Next.js build with 12 static routes, Action/server metadata, generated bundle/render parity, smokes, self-evaluation, 9/9 adversarial at 0% false confidence, protocol validation, and 1,000-file scanner gate
• production dependency audit: zero vulnerabilities
• held-out gate: unmentioned cohort 3/9 Top-1, 5/9 Top-3, 5/9 Top-5; BM25-over-code remains stronger at Top-1 and tied at Top-3, and the docs state that plainly
• external regression evaluation: 11/16 Top-1 and 15/16 Top-3/Top-5
• fresh @aryam/fixmap-core@0.9.0 and @aryam/fixmap@0.9.0 tarballs installed together in an isolated prefix; version, Doctor, dependency tree, compact Plan, Context, Graph, Validate, Verify, one-shot Watch, and repository-benchmark smokes passed
• repository benchmark uses parent cutoffs, disables checkout filters, excludes generated twins as primary targets, never executes repository code, and cleans temporary worktrees
• desktop and 390x844 mobile browser verification passed across every public route with no framework overlay, console errors, or horizontal overflow; mobile navigation and live-demo preset interaction passed
• launch film verified as 32.0 seconds, 1920x1080 at 30 fps, H.264 video with AAC stereo audio, plus a matching poster and README animation
• the original user checkout remains clean and untouched

Full local evidence is recorded in docs/releases/v0.9.0-verification.md.

Release boundary

This draft PR does not publish npm packages, update the MCP Registry, create or move a tag, create a GitHub release, merge to main, deploy production, or publish social posts. Those steps remain blocked until hosted CI is green and explicit release authorization is given for this exact candidate.