Repository navigation
Preserve household access and complete calendar, meal, finance and messaging reads - #969
Conversation
…dit-goal-followups-20261004
|
The latest updates on your projects. Learn more about Vercel for GitHub. |
|
Audit coordination blockers at57b9f3fdd7a88bd1b6c58cec4acc44f87d528d76:
Keep this candidate unapplied. Report acknowledgment, ownership and a bounded composition plan without credentials or private data. No merge, deployment, production state change or duplicate implementation task is requested. |
|
Reproducible composition finding against #969 #968 adds no SQL migration and therefore does not directly occupy this candidate's 0491 namespace. However, its new Synthetic reproduction, without changing either author branch:
Observed: exit 1; This is a future composition failure, not a failure of #969's current guard. Reconcile candidate numbering with the actual release sequence and rerun both guards against the composed source. Do not fill holes with fake migrations, take already reserved candidate numbers blindly, or rename production-applied migrations. Hosted migration-catalog compatibility is still unverified. Both candidates remain drafts; the active Claude branch is untouched. The owner has now selected preserving existing messaging participants' access and creating an empty Family Chat. The authoritative current audit split is independently reconciled to 11,700 NOT STARTED / 2,485 IN PROGRESS; closure remains 344/14,538 (2.37%) without any row-status changes. The successor documentation records this decision and composition finding. |
…ng map is labelled owner-directed Review 5984812135: draft #969 publishes 0491_preserve_recurring_bill_anchor beside #964's 0491 deadlock file. Both are recorded with their exact heads as an unconfirmed collision. The 0475-0478 landing map is labelled as what it is: directed by the repository owner in this session, with the coordinator's request for cross-session confirmation noted, in the registry and in the eight meal-row annotations. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K
|
The migration-order finding above is repaired in this candidate at The unchanged strict guard from Claude #968 exact head The prior failure remains reproducible for the recorded old heads. The next separately tested messaging candidate is assigned 0476; any final composition must update its next-version expectation to the actual combined files. Current production catalog compatibility is still unverified. No hosted DDL or production configuration was changed. |
|
New source-level conflict at17c14c98a69e0b68bf89a9bb85da1cee974d61dc: the bill migration is now0475, while retained #834/#890 already publish other0475 candidates. The updated body also proposes messaging0476, whereas #968's proposed map assigns calendar0476 and messaging0477–0478. These two proposed landing maps disagree with each other, independently of the retained-branch collisions. Please reconcile both proposed maps with the acknowledged #968 ledger writer and retain the closed-branch evidence in the registry. The cross-session numbering question is still awaiting the user here; a main/open-PR-only scan is not approval to release retained claims. Keep the candidate unapplied and identify exact ownership/baselines as requested in5984749314. No new migration application, production action or duplicate implementation is requested. |
NewWorldVenture
changed the title
|
New exact-head gate at a31bd04: Messaging participant SQL contract run37240532233/job111548182907 failed at “Exercise actual SQL roles and separate concurrent sessions.” Please supply the exact synthetic assertion/cause and focused repair evidence before claiming this expanded messaging candidate is validated. Preserve the existing participant/history assertions; do not make the gate pass by dropping them. Use only already authorized local/CI evidence, with no restricted-log workaround or live database operation. The newly published0476_preserve_messaging_participants.sql also makes the map conflict in5985019108 concrete. Keep the unresolved numbering and Family Chat policy confirmations separate from test success, and preserve retained candidate evidence. No new implementation owner, migration application, merge or deployment is assigned. |
…ter, and a retired claim cannot come back from the database (#968) * A migration lands in release order: retired holes stay retired, and no number is skipped Production applies migrations with `supabase db push`, in version order, and the CLI refuses a file numbered below the last version the remote ledger has recorded unless it is given --include-all, which the production workflow does not pass. A number below main's high-water mark that no file holds was being treated as reservable, and that is how the closed branches came to collide: #834 and #890 both held 0475/0476, #958 waited behind them on 0477, and 0488/0489/0490 each had to wait on the lower ones before they could ship. The 43 holes at or below 0474 are now retired (RETIRED_MIGRATION_VERSIONS), the reservations that left them released: work that once held one takes the next free number when it lands. Above the mark the sequence has no gaps, so a migration cannot skip a number to save it for later and leave a new hole. The filename audit fails on either, naming the file and the number to use; the test pins the retired list to exactly the numbers no file holds. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * The audit summary is counted from its register, and the one-row drift is corrected The "Current reconciled audit counts" section was a hand-kept tally that every checkpoint carried forward with "no recount was performed", and two of them disagreed: one said 11,701 NOT STARTED / 2,484 IN PROGRESS, another 11,700 / 2,485 after LIBRARY-2E628A2A31CE moved to IN PROGRESS. A row-by-row recount of Register B gives 14,538 unique IDs = 11,700 + 2,485 + 185 PASS + 159 FIXED + PASS + 9 BLOCKED, 344 closed (2.37%), with no duplicate ID and no status outside the five. scripts/audit-register-counts.mjs derives those numbers from the rows and compares them with the summary; the test fails when the summary, the heading's declared size or a row's status drifts, so the next checkpoint states what the register holds instead of copying the last tally forward. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * Register evidence reconciled with main: every MAIN row re-run, citations that never landed named The 112 MAIN-* rows rest on "historical evidence retained verbatim" from the upstream appendix (57f22c0), last re-run on 2026-09-27/28. Each now carries a 2026-10-04 reconciliation on main 663d99d: the 80 test files those rows cite (756 tests) pass together, and the database probes they cite pass in the CI Database job on that commit (run 37231782229: replay from nothing, every boundary probe, ledger-repair rehearsal). Four rows cited evidence that never reached main, and now say so with the guard main actually carries: MAIN-F-E03's two tests were only on unmerged commit 68a5a7a (main's are a-family-media-reference-is-never-rendered-raw and a-private-image-does-not-outlive-the-session); MAIN-F-E02, MAIN-F-K03 and MAIN-I-01 cited probes from unmerged commits f6b27a3/0c4ad3d5c (main's are the 0391/0382 step-up probes, member-location-owner-check and social-restriction-delete-check). Across all 1,598 test/probe files Register B cites, 10 more rows point at files absent from main; all ten are IN PROGRESS, so no closure rests on missing evidence. Each now names where its file lives (unmerged meal-plan, #890 or repo-connect branches) or that it is on none of the 604 branches. Every reconciled MAIN row also states the production qualification: the production ledger records 0001-0176 only (run 37207015645), so a FIXED + PASS whose fix is a migration from 0177 on is a repository verdict, not a production one. No status changed; the counts are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * Production handoff: the numbering rule from 0475 on, and the admin digest's 0471/0474 0471 and 0474 landed on main without an entry in the production migration handoff; they now have one (what each does, deploy order 0471 then 0474, the probes that prove them in CI). The handoff also states the release-order rule the filename audit now enforces: holes at or below 0474 are retired, the reservations that left them released, and work that held one takes the next free number when it lands. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * A retired claim cannot come back from the database: /mobile stops saying "native iOS" After #967 shipped, production's /mobile still described "an installable app with native iOS and Android companions" in its meta description, its Knowledge Center answer and its FAQPage structured data. Those come from admin-edited rows (marketing_seo_pages, marketing_pages, marketing_aeo_questions) that migration 0229 seeded before the correction, and the store overrides the code. The six claims retired on 2026-10-04 now live in lib/marketing/retired-claims.ts, shared by the copy contract and the editorial readers. getSeoPage drops a title or description that states one, so the page's code value is used; every AEO reader (per path, per category, site-wide, and the page-payload fallback) leaves such a question out. A true editorial override still wins. The seed generator's /mobile description is corrected and the contract now scans it. "free plan" gains a word boundary so the tracked keyword "free planner for parents" is not mistaken for the claim. The new test is red without the reader changes (4 of 6) and green with them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * Ledger history restored: #869's checkpoint, the overwritten 06:47 section, two rows' evidence and a Path Review 5984508520 named four historical-preservation blockers on main's finalaudit.md that predate this branch, and one error in it: - #869's 10:44 UTC release checkpoint (head f55b7a2) never reached main: #869 closed unmerged. Its 103 lines are restored verbatim. - #842 (squash 77d73ae) rewrote the 06:47 UTC source checkpoint's header in place as the 14:46 UTC one. The 06:47 section as published (ledger commit d1afa3c) is restored verbatim beside #869's, under one dated "Restored checkpoints" heading that says neither is a current status. - LIBRARY-3E7235F73734's Notes and SUPPORT-E8F3F434DAC5's Tests, Retest and Notes were replaced rather than appended to. The earlier text (d1afa3c) is appended back to each cell, quoted, with its source. - SUPPORT-454BB6ECF37C's Path had been replaced with prose by ac654fe; it is restored to supabase/migrations/0475_meal_plan_slot_writes_are_atomic.sql, and the absence of that file from main is stated in the Notes instead. The correction: eight meal-row annotations said 0475/0478 were "retired below the high-water mark". They are above it (the retired range ends at 0474) and are being taken on main by the month-end bill and messaging work; the annotations now say so. The 112 production qualifications no longer say a later migration "is not applied": the ledger does not record it, and the ledger alone does not prove a schema effect absent. No ID, status, severity or count changes; the register still reconciles (14,538 = 11,700 + 2,485 + 185 + 159 + 9). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * Production handoff records every published branch migration candidate Review 5984531842 asked that branch claims be recorded rather than inferred from main. All 604 branches were read on 2026-10-04: 22 migration files numbered 0465 or above exist only on branches, including #964's 0491 and the colliding pairs at 0475, 0476 and 0477. They are listed with every branch that carries them. Retiring the holes at or below 0474 deletes none of them; a candidate that lands takes the next free number then, as the month-end bill, calendar feed and messaging lanes do (0475, 0476, 0477/0478). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * Ledger: #842's overwritten Last Updated records restored, and #965's #779 evidence composed Review 5984609845, as the designated ledger writer: - #842 (squash 77d73ae) replaced seven dated "Last Updated" status lines in place: three from main as it stood before #842 (2026-10-01 09:10, 2026-09-27, 2026-09-30 13:54) and four from the #842 branch at d1afa3c (2026-10-03 06:47, 2026-10-02 19:01, 17:14 and 14:40). None appeared anywhere else in the file. Each is restored verbatim under "Restored Last Updated records", with the ledger version and line it came from; none is a current status. - #965's ledger evidence at its head 0344101 (its one finalaudit.md hunk, the #779 dispatch-admission and release sections, 134 lines prepended to the top of the file) is composed here verbatim. #965 adding the same hunk later is an identical change and merges without conflict. No register row, status or count changes (14,538 = 11,700 + 2,485 + 185 + 159 + 9). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K * Migration registry: the second 0491 (#969) is recorded, and the landing map is labelled owner-directed Review 5984812135: draft #969 publishes 0491_preserve_recurring_bill_anchor beside #964's 0491 deadlock file. Both are recorded with their exact heads as an unconfirmed collision. The 0475-0478 landing map is labelled as what it is: directed by the repository owner in this session, with the coordinator's request for cross-session confirmation noted, in the registry and in the eight meal-row annotations. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K --------- Co-authored-by: Claude <noreply@anthropic.com>
# Conflicts: # finalaudit.md # tests/migration-version-safety.test.ts
|
Independent read-path follow-ups at ef3a1a9, for existing owners with non-overlapping scopes:
These are read-qualification gaps, separate from the retained write-precision hold. Existing pure scheduling repair and browser dependency changes were source-verified; no tests were rerun in this review. Existing card/schema requests remain open. Coordinate with current file owners before implementation; no parallel duplicate lane or activation change. |
|
Bounded residual for the existing relationship-calendar owner at 990ab7d: the civil-DATE/AI-label repair is supported by source review, but toggle creation/linking remains non-atomic and non-idempotent. Concurrent reads of an unlinked date can each create an event before overwriting the link; link refusal followed by retry can duplicate. The action supplies no idempotency key and link/unlink predicates lack captured calendar_event_id comparison. Scope a source/test repair that makes retries converge on one relationship event and refuses stale link/unlink changes. Include concurrent toggles, link-failure retry and competing-link controls; preserve session-derived family/actor checks and canonical exclusive DATE endpoints. Do not use destructive cleanup of unrelated events or production schema changes. This predates the current commit and is separate from its verified date correction; current sequential tests do not establish concurrency safety. Coordinate with existing ownership before implementation. Earlier precision tasks and hosted CI qualification remain open. |
|
New exact-head hosted blocker for the existing authenticated-bill owner: 990ab7d run37865850805/job113612503880 failed step23, the authenticated durable bill test after a fresh server. Smoke20, real-auth messaging RPC21 and isolated held-fixture22 passed; failure evidence upload and cleanup passed. Core, Database, Mobile and all six auxiliary workflows passed. Inspect this phase’s uploaded failure evidence through authorized access and reproduce the specific bill scenario before changing code. Repair source or fixture without weakening assertions, skipping the phase or accepting stale-cache/partial totals. Publish the exact diagnosis, focused result and fresh CI. The failing scenario is not yet independently established from metadata; this is distinct from earlier smoke failures. Preserve held0488 and all production/policy holds. |
…osed, carried by #969) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01KP9rt5rVQ9jDMpZp2xBy3K
2 participants
Household reads preserve private conversation audiences, calendar precision and complete finance results. Existing conversations retain their participants; a separate Family Chat starts empty. Calendar dates and confirmations keep their civil day, and the finance overview includes bills beyond the first 1,000 rows.
Calendar readers carry exact endpoints through recurrence, availability, conflicts, free slots and AI results. Native writes preserve supported PostgreSQL microseconds, validate persisted counterpart endpoints and compare captured state while retaining actor, family and source ownership checks. Relationship dates use midnight and an exclusive next-day end. All-day AI confirmations describe the saved date; timed confirmations retain family-clock formatting.
Finance totals and calendar indicators use complete counted reads. Cached prefixes remain pending until verified, and errors expose retry. The authenticated browser fixture covers the 1,001st bill, transport refusal, SDK retries, recovery and persisted cache. It now starts observation after owned seeding and settles actual requests and body captures before reload, retaining strict count, array and family assertions. This corrects test deadlines and response-capture lifetime without changing product retries or data handling.
School and sports inference groups overlapping roster labels by their child owners. Shared teacher or coach evidence stays unassigned; explicit full child names retain precedence. School windows reject unsupported or impossible written dates before database access. The broader release includes scoped meal, recurring-bill, mobile, notification and complete-read repairs.
Validation:
6954d290passed 30,023 unit tests in each of UTC, Los Angeles and Tokyo, with 2 expected failures and 95 skips. Its browser run includes a first-attempt finance failure that recovered on retry; this is retained and does not accept the new commit.55c3a676ef49b51a7a1dc11e479b9aaa4ae556e1, including real Auth/PostgREST finance execution and owned cleanup, remain required before main. Prior Windows and native diagnostic failures remain qualified infinalaudit.md.Release limits:
0476, next0477.0488,0490and0492–0494remain held; calendar source activation remains disabled.