Run the real quality:matrix baseline comparison against 0.30.17 (release 0.31 closeout item)

[EmersonBraun]

Goal

Close the last unchecked item from release 0.31's closeout checklist (#261): run the repo's own quality:matrix command against a real 0.30.17-vs-0.31.x comparison, with detection and precision numbers reported in the CHANGELOG, instead of leaving the check unverified.

Why this is still open

npm run quality:matrix -- --baseline <path> (scripts/evaluate-quality.mjs) compares two full QualityInput artifacts — a 14-area matrix (coverage, findings, comments, security, reliability, performance, tokens, batches, memory, configuration, integration, per src/quality-matrix.ts). Building one requires a real campaign/PR-level run through the full harness (createHarnessRun/recordBatchCompletion etc.), not just a fixture cycle benchmark (npm run benchmark:baseline, which only proves deterministic fail-closed process behavior per docs/baseline-protocol.md) and not just the 20-case detection/precision numbers from the quality/evals/default.json corpus (which is what ADR-0008's live A/B used — a narrower, real but partial signal).

Neither a 0.30.17 QualityInput baseline artifact nor a current 0.31.x one exists anywhere in this repo or as an external study artifact (quality/baselines/ doesn't exist; validateStudyOutputPath requires these live outside the repo by design). Fabricating either would misrepresent real evidence, so this needs a maintainer with provider credentials and campaign infra to actually run it.

Scope

• Capture a QualityInput artifact for v0.30.17 (or the closest available real-run/campaign evidence for that version) via a genuine PR-level or campaign-level run — not the credential-free fixture.
• Capture the equivalent for the current 0.31.x line under the same conditions.
• Run npm run quality:matrix -- --baseline <0.30.17-path> --input <0.31.x-path> and confirm no area drops more than 2 points (the checklist's stated threshold).
• Report the resulting detection/precision/area scores in CHANGELOG.md or a dedicated docs/ note, and store the raw QualityInput/QualityReport artifacts externally per validateStudyOutputPath's design (never committed).
• Update docs/baseline-protocol.md to record where the captured artifacts live and how to reproduce this comparison.

Related

• Release 0.31.0: port improvements from open-code-review (tracking) #261 (release 0.31 umbrella tracking issue, already closed — this issue exists so the one unchecked closeout item isn't silently lost)
• docs/adr/0008-verification-posture.md — the live posture A/B that was completed used a narrower, real signal (20-case corpus detection/precision) and should not be conflated with this broader 14-area matrix comparison.