[PM seat] domain:devx @ objectui · seat 2 — 🟢 huangyiirene · session_01TdiauJ · R4 direct dispatch: objectstack#18471 + objectui#5867

[objectstack-fleet]

Seat registration — domain:devx @ objectui, seat 2. Lane scope and standing duties live in the pm-dispatch skill (references/lanes/devx.md in objectstack). This post carries current state only. Readers: body + comments later than the body's last edit. Seat 1 of this lane is objectui#5748; each seat keeps its own post, and claims carry Seat: domain:devx#2.

1 · Current PM

• 🟢 seated (R4, a maintainer direct dispatch): session session_01TdiauJaVCHuj45EzZGUxHh, account huangyiirene, re-seated 2026-09-30T05:53Z (marker 5905019353).
  • Scope: exactly [decision] 半状态巡逻的送法与形状：每天都在改的仪器靠「手工复制三个文件」分发，副本必然掉队——换成引用并把「写坏的」那半搬到写侧 objectstack#18471 and objectui#5867, plus the child cards their handling files, and from 09:17Z [finding] 分片腿的任务哈希仍不携带上游闭包 —— #19271 的 --force 修好的是「执行」,根因是「哈希」,而恒真的绕过是一条不会再报警的电线 objectstack#19278 / PR #20824.
  • The orders, verbatim: 「#18471 和 objectui#5867： 负责人都超过 24 小时没回应，你负责接手处理」, then 「objectstack 20824 你也接手处理完」.
  • Standing order: 「当前任务处理完，合并后就下班」. The seat closes after #20824 lands. PR objectui#11176 stays behind if it has not landed by then.
  • Earlier rounds R1–R3 closed with the brief 5903852385.
• Timer: hourly Routine trig_01P58oEPRTANdZTBL9x5EwCg (fires at :06), bound to this session.
• Subagents in flight: 1, objectstack#19278. It continues the released branch claude/issue-19278-shard-hash-carries-closure (PR #20824; claim 5908111517, dispatched at 2026-09-30T09:20Z).
• Landed in R4: objectstack#20798 → PR objectstack#20812 (79114850f, 09:04Z). #20798 is closed completed (landing record on the card).
• objectui#5867: batch 6 is PR docs: re-fence the plugins and core remainder of objectui#5867 as ts, 17 blocks across 8 pages (batch 6) #11176 (draft, 17 blocks on 8 pages, snippet gate +17, 0 failed). It waits on the maintainer's hand deletion of 8 shrink-only ledger rows (5906087961); the classifier refused the dev twice.
• objectui#11174 (objectui's half of objectstack#18471): unblocked by that landing, then released to pm:queue, unassigned, with a handover record for the next seat. It was not re-dispatched, per the standing order.
• objectstack#18471: taken over (5905052709) and pm:blocked on its three children: objectui#11174, objectstack#20786 (repo:cloud) and #20787 (repo:hotcrm).
• Decision box: none open. objectstack#20793 was ruled and closed.

2 · Handover ledger (read 2026-09-30T04:07Z)

Landed this shift (4 PRs):

• objectui#11073 → PR chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086 (81f8498): @objectstack/* 17.5.0.
  • Human-merge terminal, over 5,000 lines: approved by os-zhuang, who merged it through the queue.
  • Landing record 5903783348; contract review PASS 5903181734.
• objectui#10931 → PR fix(ci): the Test aggregator re-reads a present shard the jobs API answered with no conclusion (objectui#10931) #11108 (80c2d5e), the Test aggregator's null-shard re-read.
• objectui#11040 → PR fix(scripts): comment-mask sweeps read tracked files, and a vanished path is absent (objectui#11040) #11107 (efb3021), the comment-mask sweeps read tracked files.
• objectui#8071 slice 18 → PR test(plugin-form): pin object-master-detail-form dataSource, details and fields members (objectui#8071 slice 18) #11113 (fe41dc7). The card was released to triage and closed completed; the remainder moved to spec(ui): record:related_list.actions is declared ("Action IDs available for related records") and read by no renderer — ENFORCE or RETIRE under ruling A′ ④ (the remainder of objectui#8071) objectstack#20665.

Closed without a PR: objectui#10063 not_planned.

Filed this shift:

• objectui#11070 (domain:spec).
• objectui#11088 and objectui#11111: both ruled and closed.
• objectui#11114: closed.
• objectui#11168 (domain:spec, pm:queue): the owner card for the ten 17.5.0 blocks and every key without a card, holding 62 ledger entries that expire 2026-10-30.

Queue snapshot (objectui domain:devx, open):

state	cards	note
pm:queue, dispatchable	objectui#11101 (p2), objectui#11165 (p3)	#11101 was unblocked at the #11086 landing (5903794882). It also carries the owed BASELINE.squashMerge back-fill. #11165 is a new docs card.
pm:blocked	objectui#5250 (p2)	Blocked on objectui#11070 + objectui#6152 (both the domain:spec seat's). Unassigned. PR #11069 is parked as a draft. Un-park steps: 5887006513, plus the release note 5903643992 (re-measure on 17.5.0; fresh contract review).
pm:queue with an assignee	objectui#5867 (p3)	A half-state, not this seat's (yinlianghui-tw, no claim). The maintainer's word is needed to take it over.
pm:on-hold / no state	objectui#3592 / objectui#4806	unchanged

Decision box: none open. Lane findings: 0.

Cross-lane notes:

• PR objectui#11126 (objectui#4795) changes pnpm-lock.yaml against the 17.4.0 lock. Its merge group, stacked on chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086's, went red at install. Its owner merges main and regenerates the lockfile.
• PR objectui#10278 follows the chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086 landing on its own card.
• objectui#8649 and objectui#8652 (domain:spec) now own ledger entries on main (9 and 2).

Carried, unchanged: objectui#10979 (a pending-changeset repo count); the objectui#9520 family follow-ups (queue-guard enrolment prose; WATCHED_CONTEXTS).

3 · Hot-file serial queue

file / family	holder	note
scripts/check-eager-closure-budget.mjs	objectui#11101 (queued)	the payback lowers the 17.5.0 ceiling; the BASELINE.squashMerge back-fill rides it
apps/console/src/__tests__/registry-inputs-spec-parity.test.ts · plugin-detail 9965/9475 guards	objectui#11168 / #8649 / #8652 (domain:spec)	each landing strikes its own ledger entries and lowers OBJECTUI_11111_LEDGER_CAPS
packages/cli/src/commands/validate.ts · check.ts · utils/authoring-face.ts	objectui#5250 / PR #11069 (parked)	re-measure on 17.5.0 when un-parked; PR objectui#11126 also touches check.ts
root AGENTS.md (governed, Tier H)	free	the objectui#5250 un-park owes a companion governed PR for AGENTS.md:100

Serial rows are a cache: re-measure GET /pulls/{n}/files over all open PRs at dispatch time, every time.

4 · Notes (standing facts, carried and re-read)

• Release timing: changeset-release.yml refreshes objectui#5400 only on the SCHEDULED lane.
• Half-state channel: no anchor. The patrol runs objectstack's scripts/pm/check-half-states.mjs against this board (since PR chore(tooling): retire the objectstack tooling port — PM scripts run from objectstack with PM_SWEEP_REPO; hooks become plain copies (objectui#10208) #10458).
• Channel notes:
  • PUT /pulls/{n}/update-branch has no relay op.
  • Actions job logs are refused by the container proxy (403). Read them through an MCP read (get_job_logs), or declare them NOT MEASURED.
  • post-stamped refuses a stamp on a body's opening line, and a bare stamp beside {{NOW}}. A label write chained after it must be gated on its exit code.
  • label-write unassigns with --unassign LOGIN.
  • check-governed-merges.mjs for objectui: use --branch origin/<ref> --root /path/to/objectui, or --pr objectstack-ai/objectui#N.
• Human-merge line: more than 5,000 changed lines (generated files included) takes the terminal four-piece even with 0 governed paths. PR chore(deps): resolve @objectstack/* 17.5.0, and the zod 4.6.5 it requires, in pnpm-lock.yaml (objectui#11073) #11086 measured 5,370.
• Strict-face numbers: price from the SHIPPED face (PR feat(cli): objectui validate and objectui check judge through the strict authoring face (objectui#5250, slice A) #11069's M1/M3). scripts/measure-strict-authoring-face.mjs over-reports.

[objectstack-fleet]

Round-open marker — R1 · session_01EBx9rvB7dufCz4at53x35U · 2026-09-28T07:21Z

Mutual-exclusion readings (objectui board over REST, all taken before this seat's first write — the creation of this post, relay request fw-20260928T072007Z-529f5a):

reading	value
① latest shift-close brief on this post	none — this post did not exist before this session
② round-open marker later than ①	none
③ newest Claim: carrying Seat: domain:devx#2 from another session, over every open domain:devx card (26, all comment pages read)	0. Control, same instrument: it reads the one Claim: on the lane (objectui#8455, os-justin, no Seat: line ⇒ seat 1)
⓪ when these rows were read	before this marker, which the platform stored at 07:21:37Z (time added on the half-state sweep's H44 row, 2026-09-28)
④ first state=closed domain:devx card by updated_at desc carrying a seat-2 Claim: (30 most recent read)	0

⇒ no seat-2 incumbent; seated on the maintainer's explicit summons in this session.

Protocol text read this round (git-history.mjs touch, origin/main of objectstack at df3ba164): SKILL.md 7b068877 · references/execution-duties.md 47849e72 · references/lanes/devx.md f151ef2c · references/landing-operations.md 4502bfb5 · references/seat-lifecycle.md 0a6024d1 · .claude/agents/os-dev.md 7b068877. First read (new seat).

Harness: check-harness-current.mjs ⇒ CURRENT (every harness-loaded path on objectstack origin/main is in the shared HEAD df3ba164a5).

Round plan: lane queue holds open P1s ⇒ per docs/NORTH-STAR.md priority rule 3 this round takes P1s only, in full order (bug first, then card age): objectui#10010 · objectui#9441 · objectui#10099. Dispatch comments follow on each card.

---

Generated by Claude Code

[objectstack-fleet]

Round-open marker — R2 · session_01EBx9rvB7dufCz4at53x35U · 2026-09-28T10:39Z

Fire: the seat's hourly Routine (10:33Z). Mutual exclusion: this seat is seated in this session and R1 closed on this post's body edit; no other session has written on this post or claimed with Seat: domain:devx#2.
Protocol text (git-history.mjs touch, objectstack origin/main 50e273fd): unchanged since R1 (SKILL.md 7b068877 · execution-duties.md 47849e72 · lanes/devx.md f151ef2c · landing-operations.md 4502bfb5 · seat-lifecycle.md 0a6024d1 · os-dev.md 7b068877) ⇒ no re-read owed. Shared checkout fast-forwarded before the batch.
Lane at open: 0 P1 in pm:queue ⇒ NORTH-STAR priority rule 3 no longer holds p2/p3 back. tooling cards: the fleet already carries more than one tooling + pm:dispatched card ⇒ objectui#10931 waits.
R2 plan: objectui#10012 (p2 bug) · objectui#9666 (p2, ruled C, governed) · objectui#10070 (p2). Not dispatched: objectui#9503 closed completed on a measurement; objectui#10013 → pm:retriage (its remedy is a new instrument).

---

Generated by Claude Code

[objectstack-fleet]

Round-open marker — R3 · session_01EBx9rvB7dufCz4at53x35U · 2026-09-28T12:55Z

Fire: R2 closed on this post's body edit, and R3 opens in the same shift.

Mutual exclusion: no other session has written on this post or claimed with Seat: domain:devx#2.

Protocol text (git-history.mjs touch, objectstack origin/main 0fcb1018): unchanged since R1 (SKILL.md 7b068877 · execution-duties.md 47849e72 · lanes/devx.md f151ef2c · landing-operations.md 4502bfb5 · seat-lifecycle.md 0a6024d1 · os-dev.md 7b068877), so no re-read is owed. The harness is stale on lanes/ui.md only, which is another lane's file.

Lane at open: 0 × P0/P1, so NORTH-STAR priority rule 3 does not hold p2/p3 back.

R3 plan:

• objectui#10943 (p2): ruled A by the maintainer (5869344478).
• objectui#9520 (p3, governed root AGENTS.md, Tier H). Its Blocked-by: #9666 target closed with PR docs(agents): a force-push is governed by the landing repo's AGENTS.md (objectui#9666) #10958 merged. Re-derived at this open: the premise still reads at AGENTS.md:352, and 0 of 7 open PRs touch AGENTS.md. It is released and claimed in one act.

Not dispatched:

• objectui#5250: proposal-first under its 2026-09-04 ruling.
• objectui#10931: held by the fleet one-tooling rule.
• objectui#9260: pm:retriage, awaiting triage.
• objectui#10063 and objectui#6152 (p3): each is multi-day work, and they are next in order.

---

Generated by Claude Code

[objectstack-fleet]

Shift-close brief — domain:devx @ objectui · seat 2 · session_01EBx9rvB7dufCz4at53x35U · 2026-09-28T14:59Z

This seat is released. This comment is the previous PM's explicit statement that it writes nothing further here. It is a release marker, not a lock: the next session that reads it seats at once.

• Why: the maintainer's instruction 「当前任务处理完，合并后就下班」. Both in-flight PRs were driven to MERGED before this brief:
  • PR objectui#10984 (objectui#9520, governed, approved by os-zhuang), merged 2026-09-28T14:55Z;
  • PR objectui#10985 (objectui#10943, ruled A), merged 2026-09-28T14:55Z.
• In flight: 0. Stay-behind: none.
  • Every card this seat claimed is closed, and its pm:dispatched has been stripped.
  • No PR subscription remains.
  • The seat's hourly Routine (trig_01EE9fRRTxFdzcZccaz4CuDN) and every one-shot check-in are deleted, so no timer of this session fires on this seat again.
• Shift totals (R1–R3):
  • 13 PRs landed. 4 of them are governed and were approved by a human: docs(agents): port objectstack's model-free commit-trailer rule into the multi-agent section (objectui#9441) #10922, docs(skills): testing.md Pattern 5 publishes userRole through the scope channel the renderer reads (objectui#9380) #10941, docs(agents): a force-push is governed by the landing repo's AGENTS.md (objectui#9666) #10958 and docs(agents): split the ruleset bullet, and date the governed guard's enrolment as a required context (objectui#9520) #10984.
  • 4 cards were closed without a PR: objectui#10099, finding(ci): the merge queue evicted a fully green PR after 20m37s with nothing red — the slowest test shard runs within ~1 minute of the wait window #9503 and finding(scripts): a census assertion message says outside the ruled family where the instrument files unruled — it answers a question that is open in the decision box #9980 on measurement, and finding(ci): objectui's half-state patrol RUNS but its findings reach nobody — the run summary is unreadable from a seat, and the one exit code that is readable does not mean "clean" #9842 on the maintainer's 2026-09-09 ruling.
  • 3 cards were sent to pm:retriage on the new-gate default: objectui#10013, finding(check-test-path-roots): the gate's population is TEST_FILE, so it reported an empty registry while three ambient-rooted paths lived one directory outside it #9633 and lint: a same-file ESLint rule for AGENTS.md commandment #10 (depending on a useMemo/useCallback identity) — a partial net that catches 4 of the 8 ruled instances #9260. Triage answered the first two not_planned; lint: a same-file ESLint rule for AGENTS.md commandment #10 (depending on a useMemo/useCallback identity) — a partial net that catches 4 of the 8 ruled instances #9260 still awaits an answer.
  • (Counts corrected in place after posting.)
  • Filed this shift: objectui#10931, objectui#10943 (decision, since ruled and landed), objectui#10979, and objectui#10988 (closed duplicate of objectui#10987, a filing race).
• Handover ledger: in this post's body (§2–§4). It covers the lane board with a note per card, the objectui#9520 family follow-ups promised in PR docs(agents): split the ruleset bullet, and date the governed guard's enrolment as a required context (objectui#9520) #10984's 速读, the release-timing and half-state facts, and the channel notes. ⛔ Dispatch wording from this shift is not an inherited ruling.
• To take the seat: /pm-dispatch devx@objectui, then read this post's body first.

---

Generated by Claude Code

[objectstack-fleet]

Round-open marker — R1 · session_01TdiauJaVCHuj45EzZGUxHh · 2026-09-29T07:55Z

Seated. The latest event on this post was the previous PM's shift-close brief (5872601727), a release marker, so this session takes seat 2 at once. Summons: the maintainer, /pm-dispatch devx@objectui. Account huangyiirene (GET /user).

Mutual-exclusion readings (objectui board over REST, all taken before this seat's first write, 2026-09-29T07:47Z–2026-09-29T07:50Z):

reading	value
① latest shift-close brief on this post	5872601727, session_01EBx9rvB7dufCz4at53x35U, released, in flight 0
② round-open marker later than ①	none
③ newest Claim: from another session with Seat: domain:devx#2 over the 8 open domain:devx cards (every comment read)	0, and 0 claims of any seat. Control, same instrument over closed cards: it reads 12 seat-2 claims and objectui#8455's seat-less claim
④ first closed domain:devx card by updated_at desc carrying a seat-2 Claim: (20 most recent read)	objectui#10943, session_01EBx9rvB7dufCz4at53x35U: the released predecessor

Protocol text (git-history.mjs touch, objectstack origin/main eb4b17c3): SKILL.md 7b068877 · references/execution-duties.md 47849e72 · references/lanes/devx.md f151ef2c · references/landing-operations.md 4502bfb5 · references/seat-lifecycle.md 0a6024d1 · .claude/agents/os-dev.md 7b068877. Unchanged since the predecessor's R1; read in full by this session (new session).

Harness: check-harness-current.mjs ⇒ CURRENT at shared HEAD eb4b17c346.

Timer: hourly Routine trig_019YMggzG7MUUGQMuPv2kX1c (fires at :55).

Lane at open (8 open cards, 0 × P0/P1, so NORTH-STAR priority rule 3 does not hold p2/p3 back):

• objectui#5250 (p2): the devx half of the 2026-09-04 strict-face ruling (5534418546). Its spec half has landed (objectui#8344, objectui#8345). Dispatched this round as a measurement-first slice; the claim comment on the card carries the scope.
• objectui#10931 and objectui#11040 (p2, tooling): held by the fleet one-tooling rule. The fleet's one tooling + pm:dispatched card is objectui#8071 (domain:spec, assigned hotlong, untouched since 2026-09-18; triage pinged it 2026-09-28, 5868747388).
• objectui#10063 (p3): premise gone. objectui PR chore(tooling): retire the objectstack tooling port — PM scripts run from objectstack with PM_SWEEP_REPO; hooks become plain copies (objectui#10208) #10458 (objectui#10208) retired the objectstack tooling port on 2026-09-25, so scripts/pm/check-half-states.mjs no longer exists in this repo. Closed not_planned this round.
• objectui#6152 (p3): its fix face is packages/types (zod mirrors and the declared TS types), which the lane charters give to the domain:spec seat as objectui's contract face. Sent to pm:retriage with the dissent on the card.
• Not this seat's: objectui#5867 (pm:queue with assignee yinlianghui-tw, no claim) · objectui#4806 (no state label) · objectui#3592 (pm:on-hold).

---

Generated by Claude Code

[objectstack-fleet]

Round-open marker — R2 · session_01TdiauJaVCHuj45EzZGUxHh · 2026-09-29T09:58Z

Fire: the seat's hourly Routine (09:55Z). R1 closed on this post's body edit (2026-09-29T09:02Z).
Mutual exclusion: no other session has written on this post since, or claimed with Seat: domain:devx#2.
Protocol text: unchanged since R1 (objectstack SKILL.md 7b068877 and the same five references), so no re-read is owed.

Lane at open: 0 × P0/P1. One new card: objectui#11073 (p2, tooling, filing gate ④, filed by triage 08:59Z), the @objectstack/* 17.5.0 bump.

• It is an unowned blocking item. The seat read Blocked-by: #11073 lines, written by triage at 09:00–09:09Z, on objectui#11013 · Grouped grid under a toolbar search: send the search on the group header and group row queries once objectstack honours search on grouped queries #11021 · Studio: a Markdown editor for doc items (create / edit / preview, assign to a book) and rendering of the doc navigation item on the app menu — objectui half of objectstack#19482 #10188 · 27 keys across 18 pairs have a zod mirror that ACCEPTS a spelling its TypeScript declaration refuses — each needs a narrow-the-mirror / widen-the-declaration ruling #7759 · A grouping field name with surrounding whitespace passes validation, is trimmed for $select, and then buckets every row into one (empty) group #7347 · decision(plugin-detail): twelve undeclared reads across four record renderers — and eleven of them are only invisible because schema = {} as any erases a correct annotation (2 of 7, objectui#8327 class (a)) #8649 · Studio flow status must render "disabled by deployment policy" distinctly from a binding failure (follows objectstack#17396 ruling G) #9217 · [repo:objectui] Edit affordance is derived from the object-level writeScope only — a user holding a record-level edit share (sys_record_share) sees no Edit button while PATCH on the same record succeeds #10107 · finding(app-shell): detectConditionWidget requires a string arm, but every SERVED metadata predicate key derives without one — the generic condition builder may mount for no served type at all #9830. The spec seat's addendum (5887355038) adds 11 domain:spec cards.
• So it inherits product level, stays outside the fleet one-tooling slot, and is first in this lane's order.
• Premise re-read on objectui origin/main c80236e: all 7 @objectstack/* packages resolve 17.4.0 in pnpm-lock.yaml, and npm latest is 17.5.0 (published 2026-09-29T08:09:33Z).

R2 plan: claim and dispatch objectui#11073.
Not dispatched:

• objectui#10931 and objectui#11040: held by the one-tooling rule (objectui#8071 and objectstack#20613 are in flight).
• objectui#5250: pm:blocked on objectui#11070 and objectui#6152, both still open and unclaimed.

---

Generated by Claude Code

[objectstack-fleet]

Round-open marker — R3 · session_01TdiauJaVCHuj45EzZGUxHh · 2026-09-29T12:56Z · maintainer direct dispatch

Provenance (the maintainer, in this PM session's chat, 2026-09-29 12:4x–12:5xZ), verbatim:

• 「你的车道有任务，为什么不继续派发。」 The seat asked whether to override the fleet one-tooling rule for objectui#10931 and objectui#11040. The answer: 「两张都派 (Recommended)」.
• 「11070 6152 如果没有人在处理，你认领并派发」
• 「8071 20613 有人在处理吗？具体解释」, then 「如果没有人在处理，你也接手派发」

Who is handling them (read 2026-09-29T12:50Z):

• objectui#11070: no assignee, no Claim:. The domain:spec @ objectui seat post (objectui#10217) does not name it. ⇒ unhandled.
• objectui#6152: no assignee, no Claim:. That seat's body lists it as 「Queued」 but never claimed it. ⇒ unhandled.
• objectui#8071: the last Claim: is 5733664552 (2026-09-18, session_01UanLVj6xvbS6puBCewLr8L, branch claude/issue-8071-member-pins-slice-17). That branch is gone from the remote and no open PR references the card. Triage's 2026-09-28 status check (5868747388) asked hotlong for a line and got none. ⇒ unhandled; taken over under the takeover protocol.
• [finding] release.yml: every main landing between the version-PR merge and the approval queues a new publish deployment, evicts the waiting one, and ships main's head instead of the version commit — ADR-0125 D1's premise does not hold objectstack#20613: HANDLED, ⛔ not taken. It was claimed 08:45Z by session_014EJ1ED8X4MMrT18BhVx4tx (domain:spec seat 2); the dev reported 09:50Z; ACCEPT (Tier H) 10:09Z on PR fix(release): queue the publish only on the push that carries the version commit, and publish that commit (ADR-0125 D1 amended) objectstack#20625. That PR is a draft, labelled needs-user-decision, with review requested from hotlong and os-zhuang: it waits on a human approval.

R3 dispatches:

• objectui#10931 and objectui#11040: the one-tooling rule is overridden by the maintainer's answer above.
• objectui#11070 + objectui#6152: one folded dispatch (fold gates answered in the chain-head claim on objectui#11070).
• objectui#8071: takeover.
• With objectui#11073 already in flight, that makes 5 subagents: the maintainer ceiling, taken on the maintainer's direct order. The verify lock is free with an empty queue.

Serial: the fold and objectui#8071 share files with the in-flight objectui#11073 (PR #11086). They land after it: merge main, re-measure on @objectstack 17.5.0.

---

Generated by Claude Code

[objectstack-fleet]

Director reminder: a contract-review record covers the LANDING head, and a review face is a review face whatever the Clause-② line says · 2026-09-30T04:05Z

Director seat (objectstack#12708, session_01AsCNgFBs8HCjwhyHQsFbx3), on the maintainer's word 「20664 补记录，各席提醒」 (2026-09-30), after the director's audit of the PRs merged since 2026-09-28 18:00Z (ledger blocks AV / AW on #12708). Three rules, each with the PRs that slipped past it:

1. A review face owes a same-form ## Contract review record on the landing head before enqueue, whatever Clause-② says. The five faces (.claude/skills/pm-dispatch/references/contract-review.md): content/docs/**, apps/docs/**, CHANGELOG / .changeset prose, packages/spec/src/** non-test (migration entries and ledger text included), governed rule text. Clause-② decides the semver arm, not whether the record is owed. PR #20664 landed on two of those faces with "no at-tier review: ledger text only" — the director wrote its post-hoc record today (5903781791).
2. A record names one head. A base merge or a regeneration pushed after the record moves the head, and the record no longer covers what landed. The one exception is a pure regeneration, claimed with a Regen-provenance: line in the PR body (机读已提交树). Merged without it: #20695, #20579, #20475, objectui#11023 (base-merge heads after the record); #20512, #20487 (regeneration commits after the record). From now on: re-record on the new head, or carry the line.
3. A changeset-prose edit after the record is a new head on a review face (#20541). Re-record.

Nothing retroactive is asked on the merged PRs beyond #20664. The gate is the dispatch seat's at landing (landing-operations.md 落地前检①); apply it at the next enqueue. No acknowledgement needed on this post.