PR watch reproduction | BASE
Commit: d756fed7aaac06134bc68cbe0047e6f14b0e9aa2
Environment: Linux; built T3 server; loopback RPC; fresh isolated database.
Synthetic GitHub API via gh Unix socket; real adapter, reactor, persistence and replay agent.
Terminal: 116 x 34, dark; no browser viewport applies.
A. Start watch while lint run 100 is pending on unchanged head aaaaaaa.
Run 100 fails before the first watch sweep.
First sweep delivered failure notification for run 100; failedChecks=["lint"]
- Checks failed on aaaaaaa: |   - lint https://github.com/evidence/demo/actions/runs/100/job/100
First awakened replay turn ended. Rerun starts and fails before next sweep.
Fixture now returns failed lint run 200, same check name and same head.
Waiting for unchanged production sweep interval: 30 seconds.
Waiting for unchanged production sweep interval: 60 seconds.
Waiting for unchanged production sweep interval: 90 seconds.
Waiting for unchanged production sweep interval: 120 seconds.
Waiting for unchanged production sweep interval: 150 seconds.
After rerun: failedChecks=["lint"]
Awakened agent turns: 1; statuses=completed
Persisted rerun failure notifications: 0
B. Same head: push lint fails; pull_request lint is running, then succeeds.
pull_request event = pending; server detail checks:
  lint | pending | required=true | https://github.com/evidence/demo/actions/runs/400/job/400
pull_request event = success; server detail checks:
  lint | success | required=true | https://github.com/evidence/demo/actions/runs/400/job/400
Observed result A: rerun failure absent
Observed result B: failure hidden by another event
AI-assisted; verified with the tools listed above.