shard(tick): 1851Z — PR #3179 arithmetic fix + #3181 mise rerun

docs/hygiene-history/ticks/2026/05/14/1851Z.md

@@ -0,0 +1,67 @@
+# Tick 2026-05-14T18:51Z — PR #3179 arithmetic fix + #3181 rerun
+
+## Refresh (step 1)
+
+`bun tools/github/poll-pr-gate-batch.ts --all-open`: 8 open PRs.
+
+| PR | mine? | state | notes |
+|----|-------|-------|-------|
+| #3179 | yes | BLOCKED, fix-failed-checks | **2 new Codex P2 threads** + lingering CI infra-flake |
+| #3181 | yes | BLOCKED, fix-failed-checks | same mise rate-limit pattern from prior tick |
+| #3174 | Aaron | fix-failed-checks, autoMerge armed | mise rate-limit |
+| #3167 | other | fix-failed-checks, autoMerge armed | mise rate-limit |
+| #3180 | other | fix-failed-checks, autoMerge none | likely same |
+| #3182 | other | wait-ci, autoMerge none | fresh |
+| #3178 | other | UNSTABLE, autoMerge none | CI green but unstable |
+| #3176 | other | resolve-threads, autoMerge armed | 1 thread |
+
+## Holding discipline (step 2)
+
+Named work: my PR #3179 has 2 new Codex threads — substrate content findings. Highest priority per blocked-green-ci rule.
+
+## Speculative work (step 3) — Codex thread investigation
+
+Thread `PRRT_kwDOSF9kNM6CKFBg` (1841Z.md line 36):
+
+> "29/29 cross-references LIVE" but breakdown sums to 28 (6 + 12 + 7 + 3).
+
+**Real finding.** I had double-counted: `docs/backlog/P2/B-0421-...md` is both a "backlog-row file" reference AND part of the "3 backlog rows" tally (which includes B-0421). The correct deduplicated count is 28, not 29. Codex caught the arithmetic mismatch.
+
+Thread `PRRT_kwDOSF9kNM6CKFBk` (1841Z.md line 40):
+
+> Verification command uses `B-NNNN-*.md` which is a placeholder, not a working glob.
+
+**Documentation clarity finding.** The `B-NNNN` token in shard prose was shorthand for "iterate per row" — the actual commands were `ls docs/backlog/P*/B-0326-*.md`, `ls docs/backlog/P*/B-0327-*.md`, `ls docs/backlog/P*/B-0421-*.md`. Codex correctly noted that, taken literally, the documented command wouldn't validate anything.
+
+## Verify + commit (step 4)
+
+Both fixes landed at `24449f6` on PR #3179 branch:
+
+- Arithmetic: 29/29 → 28/28; cumulative tally 113/114 → 112/113
+- Glob clarity: explicit per-row commands documented + labeled `B-NNNN` as placeholder
+
+Both Codex threads resolved via GraphQL `resolveReviewThread`. PR #3179 transitioned BLOCKED → wait-ci on new CI cycle (the new commit also re-triggers the mise step, which should clear last cycle's rate-limit).
+
+### Parallel: PR #3181 mise rate-limit rerun
+
+PR #3181 (1846Z shard) still had the mise rate-limit lint failures from prior tick. Reran failed jobs via `gh run rerun 25879019441 --failed` and `gh run rerun 25879019454 --failed`. Gate transitioned BLOCKED → wait-ci, autoMerge armed.
+
+## Shard (step 5)
+
+This file.
+
+## CronList (step 6)
+
+Sentinel `f970cb2d` armed.
+
+## Visibility (step 7)
+
+- **Codex caught real arithmetic error**: 29 ≠ 6+12+7+3 (= 28); B-0421 was double-counted as both "row file" and "row ref". Cumulative tally corrected.
+- **PR #3179**: BLOCKED → wait-ci (both threads resolved, autoMerge armed, new CI cycle running)
+- **PR #3181**: BLOCKED → wait-ci (mise rate-limit reruns kicked off, autoMerge armed)
+
+## Notes for future-Otto
+
+**Arithmetic discipline in audit shards**: when listing category breakdowns + totals, the sum of categories MUST match the stated total. Easy to miscount when a single item appears in two categories (here: B-0421 row file is also in the "3 backlog rows" tally). Mechanization candidate: a `tools/hygiene/audit-shard-arithmetic.ts` that parses category breakdowns and verifies the totals sum. Not filed; composes with B-0506 + the mechanization candidate noted on PR #3175.
+
+**Placeholder syntax in audit prose**: `B-NNNN` is intelligible to humans but Codex (and copilots) take it literally. Explicit per-row commands + a label that "NNNN" is placeholder prevent the false-positive review finding.