[objectstack-fleet]

Requested by Steve · project thread
Fixes #26527
Clause-②: no
Metadata-protocol: none

维护者速读

• 改了什么:契约复核的 ④ 设计判断现在也挂在收窄面(Clause-②: no (narrowing)),模板、简报与规则文本同步改成三面;申报面照旧无 ④。
• 为什么改:你裁了 A(评论 6442046166):只在运行时代码里收窄已发布接受集的 PR,复核也要对照 main 上的 spec 问「这次收窄是在消费端砍掉了一个已声明的能力,还是把 spec 自己的拒绝落实了」。
• 风险与回滚:只动复核流程的文本与模板,不动任何运行时;回滚即 revert 这一个 commit。
• 先看:scripts/pm/record-recognisers.mjs 的 REVIEW_FACES narrowing 行(fullReview: true)与简报的 narrowing 专用 ④ 段;contract-review.md L22/L30。
• 两分钟验证:node scripts/pm/record-recognisers.mjs --brief-template --face narrowing 应打出 ④ 段(「the narrowing itself, read against the published spec on main」);--face declaration 的围栏内无 ④。
• 需要你做的:批准本 PR(GOVERNED_APPROVERS);席位已请总监契约复审,批准后席位落地;在此之前保持 draft。

What changes

The maintainer's ruling on #26527 item 3 (letter A, ruling record 6442046166) hangs item ④, the design judgment, on the narrowing face of the contract review as well. A PR that narrows a published accept set in runtime code alone, touching neither packages/spec/src/** nor a governed path, now gets the same ④ the schema and rules faces get, in the narrowing's own shape: no hits files, so the judgment is of the narrowing itself, read against the published spec on main — does it take a declared capability away at the consumer, or make the spec's own refusal real?

• scripts/pm/record-recognisers.mjs
  • REVIEW_FACES.narrowing gains fullReview: true; recordCarriesDesignJudgment is unchanged and now answers true for that face, so the record prints ### ④ Design judgment on --face narrowing (the declaration face, below tier, still prints three headings).
  • The --template notes name three faces and the narrowing scenario (the narrowing itself against the published spec on main); the declaration face alone is said to carry no ④.
  • The --brief-template on the narrowing face carries a narrowing-specific ④ instruction (read against the published spec on main; a capability declared and not delivered is an implementation gap, implemented or retired, never narrowed at the consumer) and a scope line saying ④ judges the narrowing over the whole-diff floor, lockfiles, generated output, changesets and prose not reviewed. The schema and rules faces keep the per-hits-file shape unchanged.
  • Docblocks on the face table, the record builder and the brief builder say the same.
• .claude/skills/pm-dispatch/references/contract-review.md
  • L30: 「④ 设计判断只欠 schema 与 governed 规则文本两面」 → 「④ 设计判断欠 schema、governed 规则文本、收窄三面」 (114 B → 118 B, same line).
  • L22: the record-shape parenthetical 「(schema/governed 两面加 ④)」 → 「(申报面外三面加 ④)」 (115 B → 111 B, same line). The card named L30 only; L22 states the same face set a second time and would have contradicted L30, so it moves with it — declared in the report as a within-file widening.
  • 55 lines before and after; no new line; every edited line under the 120-byte budget.

Why

Faces resolve schema → rules → narrowing → declaration, so a narrowing PR that also touches a spec or governed path already carried ④; only the runtime-only narrowing fell to a face without it. That is exactly the face where the pm-dispatch principle 「声明而未兑现是实现缺口,补实现或退役,⛔ 不在消费端收窄」 is most easily broken (spec untouched, runtime narrowed first), and the review asked only whether the declaration line was spelled right. The ruling closes that gap without a new face or a new gate.

Tests

Forced test edits, all in the self-test of scripts/pm/record-recognisers.mjs (pnpm check:pm-record-recognisers):

• Flipped: 「the narrowing face carries NO ④」 → carries ④ (fourth heading, fenced on --face narrowing); 「exactly the two path-read faces are full-review faces」 → the three at-tier faces (schema,rules,narrowing); 「the narrowing brief carries no ④」 → carries ④ in the narrowing's own shape (spec scenario, no hits-ONLY rule, no angle brackets, no model identifier).
• Reworded fragments: the notes pin now reads 「on the schema, rules and narrowing faces」 and 「the declaration face alone carries no ④」; the brief face pin covers narrowing too.
• Added: the notes name the narrowing scenario (template battery floor 28 → 29); the hits faces keep the per-file ④ shape and never the narrowing's (brief battery floor 17 → 18).
• Unchanged and still green: the declaration face carries no ④ in record or brief.

Self-test at 78af1243: 219 cases pass, exit 0 (217 before the change). Mutation leg (one shot, scripts/ablation-replace.mjs, anchor hit exactly once, trap-restored, restored blob byte-identical to HEAD 462459281f4b): fullReview: true → false on the narrowing row ⇒ exit 1, 5 of 219 cases fail (the three flipped pins and the two added ones). No build or dist is involved; the script runs from source.

Acceptance notes

• The card's acceptance line 「--template --paths packages/objectql/src/x.ts with a narrowing face prints ④」 is not reachable through --paths alone: reviewFace reads the narrowing face off the Clause-②: line, which a path list cannot hand it. Measured at 78af1243: --template --paths packages/objectql/src/x.ts (no changeset in the list) prints the declaration face, exit 0; --template --paths .changeset/x.md packages/objectql/src/x.ts is refused, exit 1, telling the caller to pass --face narrowing (pinned before this PR). The face that prints ④ is --template --face narrowing, and it does.
• No other reader of fullReview exists outside this file (git grep fullReview at 78af1243: record-recognisers.mjs only); the queue guard and dispatch-gates do not read it.
• The pm-dispatch SKILL.md rule the brief now echoes (「⛔ 不在消费端收窄」) is untouched.

Verification

• node scripts/pm/record-recognisers.mjs --self-test — 219 cases pass, exit 0.
• pnpm check:pm-skill-ratchet — exit 0 (self-test and scan); pnpm check:pm-skill-id-lint — exit 0, 104 files clean.
• Derived gate list (node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 78af1243): 38 commands; results in the os-dev-report comment on pm-dispatch(contract-review): item ④ names the rules face as three surfaces and the brief names the accepted scenario for an ADR only — reword to the governed register, add per-surface scenarios, decide ④ on the narrowing face #26527 and the details block below.

Derived gate sweep at `78af1243` — 38 commands, every exit 0, captured before any pipe; `--ran` reconciliation: 38 derived, 38 run, 0 NOT-MEASURED, 0 UNRUN

node scripts/check-closing-keyword-parity.mjs :: exit 0
node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0
node scripts/check-comment-mask-corpus.mjs :: exit 0
node scripts/check-declaration-mirrors.mjs :: exit 0
node scripts/check-declaration-mirrors.mjs --self-test :: exit 0
node scripts/check-self-test-wired.mjs :: exit 0
node scripts/check-self-test-wired.mjs --self-test :: exit 0
node scripts/check-self-test-workflow-commands.mjs :: exit 0
node scripts/check-self-test-workflow-commands.mjs --self-test :: exit 0
node scripts/check-whole-set-label-write.mjs :: exit 0
node scripts/check-whole-set-label-write.mjs --self-test :: exit 0
node scripts/ci/plan-test-core-shards.mjs --self-test :: exit 0
node scripts/pm/bare-root-worklist.mjs --self-test :: exit 0
node scripts/pm/check-harness-current.mjs --self-test :: exit 0
pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0
pnpm check:agent-test-spelling :: exit 0
pnpm check:bash32-floor :: exit 0
pnpm check:ci-filter-parity :: exit 0
pnpm check:cli-command-ids :: exit 0
pnpm check:cross-package-test-inputs :: exit 0
pnpm check:doc-authoring :: exit 0
pnpm check:driver-memory-census :: exit 0
pnpm check:entry-guard :: exit 0
pnpm check:gitlink-declared :: exit 0
pnpm check:nul-bytes :: exit 0
pnpm check:parse-guard :: exit 0
pnpm check:pm-dispatch-gates :: exit 0
pnpm check:pm-governed-merges :: exit 0
pnpm check:pm-record-recognisers :: exit 0
pnpm check:pm-skill-id-lint :: exit 0
pnpm check:pm-skill-ratchet :: exit 0
pnpm check:pnpm-filter-targets :: exit 0
pnpm check:pnpm-flag-spellings :: exit 0
pnpm check:ratchet-remedy-authority :: exit 0
pnpm check:refd-timer-probe :: exit 0
pnpm check:scripts-symbol-anchors :: exit 0
pnpm check:skill-frame-sync :: exit 0
pnpm check:watch-hint-literal :: exit 0

Notes: pnpm check:pm-dispatch-gates ran its fast tier (1935 cases pass; 7 slow sections deferred, as the script does on every local run) in 5m30s; @objectstack/lint check:doc-formula-expressions needed @objectstack/formula and @objectstack/lint built first (through scripts/pm/os-verify-lock.sh, VERDICT command-exit 0, 99 s). Not in the derived list and not run locally: repo-wide pnpm lint (CI owns it) and the CI families outside the derivation; the PR check-runs are the record for those.

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic. (not touched)
• 2. Every name I added uses the casing for its kind.
• 3. No formatter ran; pnpm lint is CI's run (repo-wide scan).
• 4. One issue, one branch off an up-to-date main, one pull request.
• 5. A change that publishes carries a changeset — nothing here publishes (.claude/** + scripts/pm/**), so skip-changeset; both declaration lines are filled in above.
• 6. packages/spec not touched.
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English (the 维护者速读 block is the maintainer's Chinese read, as the dev contract asks).
• The Before Submitting list in CONTRIBUTING.md is done.

---

Generated by Claude Code

[objectstack-fleet]

席位核对

ACCEPT · domain:skills#1 · session_01ADHkmSrRBdu9MdC6fc33Uu · head 78af1243 · 2026-10-27T13:15Z

1. 形态 — Tier S (.claude/** hit) plus scripts/pm/record-recognisers.mjs: REVIEW_FACES.narrowing fullReview: true; --template notes and the narrowing brief carry ④ in the narrowing's own shape; declaration face still without ④; contract-review.md L22 and L30 same-line edits (111 B / 118 B), 55 lines unchanged.
2. 前提实测 — seat ran the head: --template --face narrowing prints ### ④ Design judgment; --self-test 219 cases pass; the narrowing ④ text matches the ruling (6442046166, letter A) and SKILL.md's 「⛔ 不在消费端收窄」.
3. 门禁 check-runs — dev ran the 38 derived families on the head, all exit 0 (--ran 38/38, 0 NOT-MEASURED); mutation leg (narrowing back to false) reds 5/219. CI on the head is read again before landing.
4. 接受的偏差 — ① L22 edited beyond the claimed L30: the same rule's second statement of the two faces, left alone it would contradict L30 (seat picks the dev's option A, keep). ② The delta's acceptance command was the seat's error: --paths cannot reach the narrowing face (it is read off the Clause-②: line); --face narrowing is the reachable spelling and prints ④.
5. 范围外去向 — none. Skills-docs: none. Contract review at CONTRACT_REVIEW_TIER: requested from the director seat.
6. 落地条款 — governed (.claude/**): stays draft until a GOVERNED_APPROVERS approval and a PASS contract review on the head; then the seat readies, arms and the queue lands it.

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 78af1243ae304c489f3f80867c94f007ca13361e
Local-runs: none
Scope: .claude/skills/pm-dispatch/references/contract-review.md + scripts/pm/record-recognisers.mjs, scripts/pm/check-skill-line-ratchet.mjs, .claude/skills/pm-dispatch/SKILL.md

Director seat (summon #52, ledger objectstack#12708), summoned by the domain:skills#1 seat. Read-only at the head above: the diff, card #26527 and its rulings (6442046166 — this seat's, batch #412 item 1 — and 6442757569, summon #51's batch #413 item 3, both letter A), the seat's ACCEPT 6442835214, the head tree (fetched, read with git show / git grep), the head's check runs. dispatch-gates --tier: .claude/** ⇒ the rules face, CONTRACT_REVIEW_TIER; record-recognisers.mjs is the gate that enforces the rule, so it is in scope as one change with it. Served at CONTRACT_REVIEW_TIER. This seat wrote the ruling the PR implements and is a reader of the template it changes; the review is of the diff against the ruling's words, named so the record is read with that in view. The seat records; it does not approve, ready, queue or merge.

① Derived judgments

• The ruling is implemented as worded. Ruling A: REVIEW_FACES.narrowing gets fullReview: true; contract-review.md L30 reworded so ④ is owed on the schema, governed-rules and narrowing faces; the self-test pins flip; the face's ④ asks, against the spec on main, whether a runtime-only narrowing retired a declared capability at the consumer or made the spec's own refusal real. Head: the face row flips (REVIEW_FACES line 169); L30 「④ 设计判断欠 schema、governed 规则文本、收窄三面」; the template notes and the narrowing brief carry that exact question. recordCarriesDesignJudgment(face, tier) is unchanged (tier === CONTRACT_REVIEW_TIER && (face === null || face.fullReview)), so the flag alone moves the record, the template and the brief — one reader, no second list. git grep fullReview on the head: this file only.
• Declaration face untouched. TIER_DEFAULT still prints three headings in record and brief; pinned both ways (the kept 「nor the declaration face」 cases). Correct — the ruling reached the narrowing face only.
• L22 moved with L30. 「(schema/governed 两面加 ④)」 → 「(申报面外三面加 ④)」: the same rule's second statement; left alone it would have contradicted L30 in the same file. Within the card's rule, declared in the body and the seat's ACCEPT; the right call. 55 lines before and after; L22 111 B, L30 118 B under the 120-byte cap.
• Narrowing brief shape is the ruling's. No hits on that face, so ④ is of the narrowing itself over the whole-diff floor, with the hits-ONLY rule withheld and the spec and cited code READ, not reviewed. The schema and rules briefs keep the per-hits-file ④ and are pinned never to carry the narrowing's wording. Consistent with contract-review.md L31–32 (逐 hits 文件判 for the path faces).
• Pins: three flipped, two added, two batteries raised by one (28 → 29, 17 → 18), each named in the file's own ledger comment; the body's one-shot mutation (narrowing back to false ⇒ 5 of 219 fail) shows the pins are load-bearing. Read from the diff and the body, not re-run here.
• Acceptance command corrected honestly. The card's 「--template --paths packages/objectql/src/x.ts prints ④」 is unreachable by design: the narrowing face is read off the Clause-②: line, which a path list cannot carry (a pre-existing, pinned refusal tells the caller to pass --face narrowing). The reachable spelling, --template --face narrowing, prints ④ — pinned. The card's intent is met; the wording was the filer's slip.

② Semver level

.claude/** and scripts/pm/** publish nothing; skip-changeset; Clause-②: no, Metadata-protocol: none in the body. No level owed. Correct.

③ Boundary flags

• Governed path (.claude/**, Tier S): no approval yet; draft; reviewers os-zhuang, hotlong requested through the relay. This seat's account is not a GOVERNED_APPROVERS member; this record is not an approval. Under AGENTS.md Tier S, a PASS record at CONTRACT_REVIEW_TIER on the head plus green checks is the landing condition — this is that record.
• Check runs on 78af1243 at this record: 26 success (incl. Lint · PM dispatch-gates — home of check:pm-record-recognisers — Lint · repo & tooling gates, Lint · scoped · tooling self-tests, Governed Surface Queue Guard, the PR-body guards, Test Core (2/8)), 20 expected skips, 1 in_progress (Test Core (1/8), no scripts/pm surface in it). Nothing red.
• Two ruling records exist for the same item (6442046166 by this seat at 2026-10-27T12:37Z; 6442757569 by summon [WIP] Fix issues in action run step #51 at 2026-10-27T13:12Z), both letter A with the same execution; the PR cites the first. No conflict; the duplicate goes on this seat's ledger, not on the PR.
• The seat's ACCEPT item 4 ② (acceptance command) and the dev's note agree with the reading above.

④ Design judgment

• contract-review.md L22 / L30 against the ruling and the scenario it is accepted against (a PR that narrows a published accept set in packages/objectql alone, Clause-②: no (narrowing)): the reviewer at tier now owes a fourth section that reads the spec on main and asks whether the narrowing retired a declared capability — exactly the check SKILL.md 〈升级与决策〉 「声明而未兑现是实现缺口 … ⛔ 不在消费端收窄」 wanted and no face performed. The sentence names three faces and keeps 「spec、skills、ADR 皆已在内 ⛔ 不另立面」, so no new face is added and the register stays the rules face's definition. Holds; nothing to break.
• record-recognisers.mjs as the gate: one boolean on the face table drives record, template and brief through the one existing reader; the narrowing brief's ④ names the question, the floor and the read-only rule, and the pins refuse a silent revert. The brief text stays free of angle brackets and model identifiers (pinned). Holds.

Implemented-by: claude/project-thread-2u3auo
Reviewed-by: session_01ADHkmQr732toX135tyo5zs

VERDICT: PASS

---

Generated by Claude Code