shard(tick): 1629Z — PR #3518 review-cycle closed; B-0504 + B-0505 also flipped + B-0503 spec text reconciled

docs/hygiene-history/ticks/2026/05/15/1629Z.md

@@ -0,0 +1,30 @@
+# Tick 1629Z — PR #3518 review-cycle closed; B-0504 + B-0505 also flipped + B-0503 spec text reconciled
+
+## Headline
+
+- **PR [#3518](https://github.com/Lucent-Financial-Group/Zeta/pull/3518)** review cycle closed. Codex + Copilot caught two parallel issues; both fixed in `ab1e24f` and all 4 threads resolved. PR back to auto-merge-armed with 0 unresolved threads, 3 required checks in-progress.
+- **PR [#3519](https://github.com/Lucent-Financial-Group/Zeta/pull/3519)** (tick 1618Z shard) — still wait-CI, 1 in-progress required check.
+- Cron sentinel `1011e43d` confirmed live.
+
+## Findings addressed in `ab1e24f`
+
+| Finding | Reviewer | Fix |
+|---|---|---|
+| Closing B-0442 while children B-0504 + B-0505 still `status: open` makes backlog graph inconsistent | Codex P2 + Copilot P1 | Flipped B-0504 + B-0505 to `status: closed` (both children's work landed via PR #3458 chain) |
+| B-0503 acceptance note says `buildRecoveryBranchName(prNumber)`; design sketch + "Why `new Date()` internally" section still show old `(prNumber, ts: Date)` + timestamped branch | Codex P2 + Copilot P1 | Reconciled all three locations in B-0503: signature, internal call site, rationale section retitled to "Why the recovery branch name is `recovery/<prNumber>` (no timestamp)" |
+| BACKLOG.md generated-index drift (non-required check warning) | gate.yml lint | Regenerated via `BACKLOG_WRITE_FORCE=1 bun tools/backlog/generate-index.ts` (4 status flips reflected in autogenerated index) |
+
+## Substrate-discipline observation
+
+Initial PR #3518 had two related-but-distinct failure modes I missed:
+
+1. **Backlog-graph consistency**: parent-status flip requires children-status flip too. Composes-with relationship in row frontmatter isn't enforced by tooling; future-Otto can write a `tools/lint/backlog-graph-consistency.ts` that flags "parent: closed AND any child: open" as an error class. Recorded here as a generative factory improvement candidate.
+
+2. **Spec-vs-impl drift propagation**: I caught the `buildRecoveryBranchName` drift at the acceptance-checkbox scope but missed the same drift in the design sketch + rationale section of the same row. The drift was present in 3 places; I patched 1. The lesson: when noting drift, grep the whole row for the drifted symbol + reconcile all occurrences, not just the most-visible one.
+
+## Composes with
+
+- [`.claude/rules/refresh-before-decide.md`](../../../../../../.claude/rules/refresh-before-decide.md) — found the children's `status: open` only by re-querying `origin/main` after the initial PR
+- [`.claude/rules/blocked-green-ci-investigate-threads.md`](../../../../../../.claude/rules/blocked-green-ci-investigate-threads.md) — PR #3518 BLOCKED with green required-checks; the unblock path was thread investigation
+- [`.claude/rules/encoding-rules-without-mechanizing.md`](../../../../../../.claude/rules/encoding-rules-without-mechanizing.md) — backlog-graph consistency is a future mechanization candidate
+- PR [#3458](https://github.com/Lucent-Financial-Group/Zeta/pull/3458) — the chain-closing landing all 3 children's work