List closed issues under Fixed only when a repair is recorded
Closed Issue/Bug/Bugfix/Defect records previously entered Fixed from type alone, turning factual closures into unsupported repair claims. The shared tracker-evidence policy now keeps factual closures under Closed as Closed: <title> and omits explicit non-change dispositions. Features and Tasks retain their category rules. Generation, summaries, structured JSON, semver, release-note exports and workflow receipts share the policy, including real trackers without git.
This review round fixes both unresolved findings:
| Thread | Result |
|---|---|
PRRT_kwDOSfkApc6ubWS1 | Require completed repair wording. “Resolve in a follow-up release” and “Fix after the next deploy” stay neutral unless independent linked files/tests, a completed resolution or delivery evidence exists. Related Repair/Correct instructions and both evidence fields are covered. |
PRRT_kwDOSfkApc6ubWS7 | Route neutral items while building field groups, preserving selected item order and first-appearance group order. Mixed repairs/neutral closures retain one ordered Closed group under status, type and label grouping. |
The complete review inventory and every unresolved comment were read with pagination exhausted. The latest CodeRabbit review has no outside-diff findings. Whole-diff self-review also covered type/label heading collisions and related future-repair wording. GitHub thread resolution states remain unchanged.
Behavioral revert proofs use independently rebuilt disposable copies of the same source:
- Restore only
REPAIR: 12 of 17 future-repair tests fail; all five independent-evidence controls still pass. - Restore only
groupSectionItems: all three ordering regressions fail. - Restore only
groupByField: type/label ordering regressions fail; the status control passes.
Working source checksums remained unchanged during the probes. Final linked closure tests pass 165/165, including SDK-created plain trackers, CLI and extension exports without git. Committed dist is rebuilt, existing pm history is append-only, and changelog generation/checking follows tracker writes.
Validation on final head 8bd72c63e3aa26a3f752c6f5f2f1d670eeb7e3ad: CI Node 22 and Node 26 each pass 782/782, with zero failures/skips/cancellations, 100% measured lines/branches/functions across all 14 required sources, 175 documented declarations, zero production vulnerabilities and every release component green.
Full local npm run release:check exits 1 with 774/782 and eight existing 5000ms worker-startup timeouts; full local bun run release:check exits 1 with 779/782 and three such timeouts. Both report 100% measured lines/branches/functions and zero skipped/cancelled tests. Every timed-out case passes in both exact-code CI jobs, satisfying this round’s explicit host-load exception. Thresholds and startup/CPU/watchdog budgets are unchanged. Both local attempts use isolated HOME/Git configuration; Bun runs in a separate clone verified against every tracked working file.
All remaining release components also pass separately under both npm and Bun: production audit, 52-file package dry-run, changelog drift/links, version-derived date and publish attestation. Post-push identity/privacy passes 4/4. The final receipt changes tracker metadata only; tested source, tests, dist and dependencies are unchanged.
Tracking and durable receipts: pmc-txk1. The claim is released and the item remains in progress for orchestrator verification. Linked result storage reports tracking_disabled, so results are recorded in comments. No new PM CLI/SDK defect was found. Existing CI legacy-provenance and independently measured statement-coverage limitations remain unchanged.