Bounded Reproduction and Validation
Audience: software-engineer. Updated: 2026-10-11.
Use temporary fixture Missions, never damage this Mission's own status history. Prior investigator reproduction is in research.md, E1.
1. In a TemporaryDirectory, write {not json\n to status.events.jsonl. Call existing run_doctor for that directory. On baseline without snapshot it falsely reports healthy. 2. Add a valid status.json with empty work_packages; rerun. Baseline raises StoreError during drift replay. Both are bugs, not expected final results. 3. New failing-first acceptance tests must demand ERROR/unhealthy without exception for both; run explicitly and commit before production edits. If transitional regression marker is used, it must not hide RED evidence and must be removed on the fix. 4. Extend five-class/two-snapshot matrix, truthful physical lines, CLI text/JSON, byte preservation and repaired-history controls. No writes beyond temporary fixtures. 5. After fix, run commands in plan.md, every changed test file and status owning-module fast tier, plus the shared make test-fast baseline once. No duplicate concurrent runs or heavy full suites. 6. Independent reviewer checks red→green commit order and actual complete diff. Parent opens PR targeting main. Leave mission approved/accepted pending PR integration; never merge/consolidate or push main.