docs(pm-dispatch): grading reads the checklist ITEM, not the area — the Path: line carries 定义项 | 清单项 id/缺项 | 路步, the card inherits the item's priority, and 「有人在等」 is not a grading test

[os-steve]

Fixes #19155
Clause-②: no

Lands the maintainer's grading ruling (quoted verbatim on the card body, 2026-09-19: 「我们是一个开发工具,应该考虑北极星有没有这个需求,有人在等吗 是个伪命题」) and the filer's two-layer test (comment 5737998283 — the spec) into the charter's grading line, in place, at net 0 on the 813-line ceiling: one line rewritten, two lines added, two second pointers retired. Every touched line is under 120 bytes (117 / 113 / 112 B). Diff: 1 file, +3 / −3. Head 9f87053 on origin/main 8e368dc (which carries PR #19449's landed lines; its neighbours are untouched, quoted below).

Sanitizer note: the file writes each grammar slot inside angle brackets (the same convention the old line used). This body spells them as placeholder words — DEFN, ITEM-ID, P-STEP — because GitHub strips short angle-bracket fragments from bodies, backticks included.

The line that stood, the lines that land

	before (origin/main 8e368dc)	after (head 9f87053)
grading line SKILL.md :363 (110 B)	「定级按 docs/NORTH-STAR.md「优先级」第 1、2 条;正文首行 Path: P-n | AREA | none」 — the second slot is an AREA (清单区), so a card that merely touches an area read as 「清单上的能力断了」	:363 (117 B) 「定级按北极星「优先级」1、2 条;首行 Path: DEFN | ITEM-ID/缺项 | P-STEP | none」 — definition item, then the checklist ITEM that would fail (its area.slug id) or 缺项, then the path step
definition layer — nothing	no line said where a 定义项 comes from, what an instrument card's is, or that a card outside the definition closes	:364 (113 B, new) 「定义项取北极星「做出来的是什么」或仪器卡的车队决定;无则关;none 非分诊缺口」
ledger layer + the negative rule — nothing	no line said that only an item that would FAIL counts as broken, that the card inherits the item's own priority, what a missing item does, or that 「有人在等」 is not a test (0 occurrences on main)	:365 (112 B, new) 「清单项会 fail 才算断,继承其 priority;缺项先补账本;⛔「有人在等」不是定级判据」
分诊缺口 rule (:464 → :466)	「…无级/缺 Path: 轮报记分诊缺口」	unchanged — a MISSING line is still the round-report gap; :364 says a none slot is a written answer, not that gap

The 「有人在等」 sentence is written as the negative rule (⛔ … 不是定级判据), not quoted as narrative: the file's register is one executable sentence per rule, and check:pm-skill-id-lint's corpus rule (no ruling dates, no quotations, no issue numbers) holds — the ruling itself stays on the card and in this PR.

Paid for by retiring two second pointers — net 0

retired (all SKILL.md line numbers in this table are on 8e368dc)	what it was	the detail still lives at
:588 「失报与停摆细则(梯度、身份三元组、验收动作)见 references/dispatch-runbook.md」	pointer, 4th of four to the same file	dispatch-runbook.md :164 「## 停摆复位梯度」, :173 「## 通知重放去重」, :179 「## 直接验收兜底」; the 收集 section keeps the rules themselves at :583 (梯度 / unreliable), :584 (身份), :585–:587 (直接验收兜底's three conditions); the file stays pointed at from :384, :532 (the general 「构造细则与条款原文见」) and :729
:698 「检测读数、盲区与恢复 playbook 细则见 references/platform-readings.md」	pointer, 3rd of three to the same file	platform-readings.md :450 「## 断粮检测与跨墙恢复(5 小时用量墙)」 — the same title as the SKILL.md section; the section keeps its six rules (:699–:704); the file stays pointed at from :155 and :160 (「实测事实表在 …,做对应操作的那一刻查阅」)

Not retired: :306 「收尾四步、僵尸回收与领地防撞细则见 references/seat-post-protocol.md」 is the only signpost in SKILL.md to that file's 「Epic 车道收尾与僵尸回收」 section (僵尸 occurs nowhere else in the file), so retiring it would orphan the detail, not a duplicate.

Why two lines, not the claim's one — a measured correction

The claim's proposed rule sentence (「命中清单区 ≠ 能力断了;写不出会 fail 的清单项 / 断的路步 ⇒ none ⇒ p3 或不做;定级继承该项 priority(断)或 P2(能跑但错);⛔「有人在等」「拉动」不是判据」) measures 212 bytes against the 120-byte cap, so it is not one line. The tightest two-line spelling (grammar + one rule) dropped the definition layer's closure (「无则关」 — the maintainer's actual sentence 「北极星有没有这个需求」) and the none-is-an-answer clause, so the delta is three lines and the payment is two retirements, both of the pattern PR #19449 used (a second pointer to a file that stays pointed at). Clauses NOT restated because the pointed-at North Star already carries them: 能跑但出错 ⇒ P2 and 不在路上、不在清单上 ⇒ p3 或不做 (优先级 rule 2), and a 缺项's own grade (a path step named ⇒ 路断了 ⇒ P0/P1; none ⇒ 不在路上 ⇒ p3 — the P-STEP slot plus rule 2).

Path: specimen — a grading this shift, before and after

#19343's first line (graded p3 by this seat today): 「Path: none | fact table (.claude/skills/pm-dispatch/references/platform-readings.md, 队列 section — the DEQUEUE direction) | family: … | graded by domain:skills#2」 — the second slot holds a file / area description; no item id could be named because none exists. Under the new grammar the same card reads 「Path: 席位对 dequeue 事件的读法(仪器卡的车队决定) | none | none | none | graded by …」 — slot 1 is the fleet decision (North Star 「仪器为车队服务」), slots 2–3 are none as written answers, and the grade follows from rule 2 (no item fails, no step breaks), not from the area hit. The triage seat's #19143 grade (comment 5748750091 on this card) rested on five area word-counts and named no item — the failure shape the new slot 2 makes unwritable.

Acceptance greps (control on origin/main 8e368dc, reading on head 9f87053)

word	control (main)	head	where
清单项	0	2	:363, :365
定义项	0	2	:363, :364
缺项	0	2	:363, :365
有人在等	0	1	:365, as the negative rule
Path:	2	2	:363, :466
清单区	1	0	—

Ceilings before / after (node scripts/pm/check-skill-line-ratchet.mjs, green before and after)

file	before	after	ceiling
.claude/skills/pm-dispatch/SKILL.md	813	813	813 (headroom 0)

No ceiling raised, no cross-file move. Twins: references/core-rules.md carries no Path: grammar line and references/lanes/triage.md (7 lines) carries no grading line — measured by grep over the whole pm-dispatch/ tree and os-dev.md (Path: · 清单区 · 定级 · 有人在等 · 清单项): every 定级 hit outside SKILL.md is a lane-charter sentence about WHO grades, none about the test — so no twin changes.

Mechanism assumptions — each verified

1. :363 reads 清单区, 「有人在等」 0× on main — verified on 8e368dc (counts above).
2. PR docs(pm-dispatch): land the maintainer's filing/merging directive — items 1 2 3 4 5 8 — as in-place charter deltas, net 0 per ratcheted file #19449 MERGED and on origin/main — merged: true, merge_commit_sha be488ce, git merge-base --is-ancestor be488ce origin/main exit 0 (exit 0 is self-proving on the shallow clone). Its landed lines, before and after this edit, byte-identical: :351 「(a) 须今天可达,(c) 须具名生产者;观察、休眠、零拉动 ⛔ 不立卡、不进汇总卡、无配额」, :357 「决策箱勤务:…Prior rulings on this card: 行;存量卡低频回填」, :358 「线程有 Ruling:/RULED 的卡再入箱,…pm:retriage 回立卡者」, :368–:370 (now :370–:372, the three merge-at-first-grading lines), :458 (now :460) 「自在飞派发升级出的卡首行一句:…」.
3. The three gates are in the derived union and green — check:pm-skill-ratchet, check:pm-skill-id-lint, check:pm-governed-prose all in the 21-command union below, exit 0; the 813 ceiling is untouched.
4. check-governed-merges.mjs --pr N → GOVERNED (exit 3), Tier S — runs after this PR opens; its reading is in the report comment on skills(pm-dispatch): grading reads the checklist ITEM, not the area — Path: carries the item id, the card inherits the item's P, and 「有人在等」 is not a test #19155. Tier: dispatch-gates --tier --repo objectstack-ai/objectstack .claude/skills/pm-dispatch/SKILL.md → the path-derived mandate (clause ①), MANDATORY_TIER_GLOBS.

The checklist ledger, measured on 8e368dc: docs/qa/platform-checklist/areas/*.json — 15 files, 264 items, every id of the shape area.slug (264 / 264), field name priority on every item (P0 = 20, P1 = 153, P2 = 91), every item status: active (no planned state — which is why the definition layer comes first).

Verification (head 9f87053)

Derived union — node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no paths; change set 1 path vs merge base 8e368dc, committed 1 / working tree 0 / untracked 0) → 21 commands, each exit captured before any pipe, reconciled with --ran: 「✓ dispatch-gates --ran: 21 derived famil(ies) accounted for — 21 run, 0 NOT-MEASURED (a DERIVED zero — all 21 recorded an exit code and none of them is 3)」.

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/pm/check-governed-queue-guard.mjs --self-test :: exit 0
node scripts/pm/check-harness-current.mjs --self-test :: exit 0
pnpm check:agent-test-spelling :: 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:gitlink-declared :: exit 0
pnpm check:nul-bytes :: exit 0
pnpm check:pm-expected-skips :: exit 0
pnpm check:pm-governed-merges :: exit 0
pnpm check:pm-governed-prose :: exit 0
pnpm check:pm-half-states :: exit 0
pnpm check:pm-skill-id-lint :: exit 0
pnpm check:pm-skill-ratchet :: exit 0
pnpm check:refd-timer-probe :: exit 0
pnpm check:skill-frame-sync :: exit 0
pnpm check:watch-hint-literal :: exit 0
pnpm --filter @objectstack/lint run check:doc-formula-expressions :: exit 0

check:doc-formula-expressions first exited 3 (PREREQUISITE NOT MET — @objectstack/formula unbuilt in the fresh worktree; not a measurement); @objectstack/lint... (lint plus its upstream closure, spec included) was built under bash scripts/pm/os-verify-lock.sh (VERDICT command-exit 0, held the lock 175 s, waited 0 s) and the gate re-ran to exit 0. The ratchet's own line: 「✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/SKILL.md is 813 lines (ceiling 813; headroom 0)」. check:skill-frame-sync and check:pm-half-states (named on the card) are in the union above. pnpm check:pm-dispatch-gates (not in the derived union) runs detached with its exit captured to a log; its final exit is in the report comment on #19155. No package is touched, so ① (dependency closure build) and ② (package tests) are empty by the local-verification rule; the closure build above served one gate's prerequisite only.

四维分析 — 语法槽序与每处措辞的取舍

• 实际业务需求 — 每一槽都对应本班实测到的失效:分诊席在 [finding] a runtime-created dataset reaches ZERO author-time rules — the type declares allowRuntimeCreate: true while nothing declares it in runtimeTypes and TYPE_TO_STACK_KEY has no row #19143 上以五个清单区的词频命中作为「清单上的能力断了」的论据、未点名任何清单项(评论 5748750091);本席自己今天的 Path: 行(removed_from_merge_queue fires on a SUCCESSFUL merge too — 3 of 5 on this board today were merges, so the event is not a failure signal #19343、本卡)第二槽填的是文件/区域描述。账本实测:264 项、每项一个 area.slug id 与自己的 priority(P0 20 / P1 153 / P2 91),第二槽有真实总体可填;全部 264 项 status: active、没有 planned 态,所以定义项必须排在账本前 —— 先查账本会把每张能力缺口卡关掉(立卡者更正的原话)。「有人在等」写成否定规则,是维护者原句的操作形式。
• 项目长远合理性 — 槽序照抄北极星自己的顺序:定义行(「做出来的是什么」)→ 那条路 → 账本(docs/qa/platform-checklist/);章程不复制账本、不带数字(北极星:本页不带任何数字),级直接继承账本项的 priority 字段而不是再推导一遍。已在北极星里的句子(能跑但出错 ⇒ P2;不在路上不在清单上 ⇒ p3 或不做)不重抄,指针指向即可。
• 防 AI 写错 — area.slug 形状的清单项 id 是能机械核对的 token,清单区名不是;none 作为每槽的书面答案,让「没答」与「不适用」可区分(缺 Path: 行仍是分诊缺口,:466 未动)。否定规则限定为「定级判据」:不限定的「拉动不是判据」会与 :351「零拉动 ⛔ 不立卡」(立卡门槛)和 :756「推荐序按拉动定向」(方案推荐序)直接冲突,正是 :46 禁止的两条细则冲突 —— 所以席位读法里的「拉动」二字没有落,「有人在等」即拉动之问在定级层的原话,已覆盖其实质。
• 创业阶段不扩散需求 — 不加文件、不抬上限、不加门禁、不为仪器卡另起语法:仪器卡按北极星「仪器为车队服务」原句,把会不一样的车队决定填进第一槽;缺项卡保留但先回账本补项(checklist-author 的输入),不按「有人在等」排程。

维护者速读(草稿)

改了什么 — 把您 2026-09-19 对定级的裁定(「我们是一个开发工具,应该考虑北极星有没有这个需求,有人在等吗 是个伪命题」)落进 PM 章程的定级行:Path: 行从「路步 | 清单区 | none」改成「定义项 | 清单项 id 或缺项 | 路步 | none」——先问北极星「做出来的是什么」里落在哪一项(仪器卡则填它会改变的车队决定),落不到就关;再问账本里哪一个清单项会 fail(写它的 id),级继承该项自己的 priority,只命中某个清单区不算断;账本里没有对应项的能力缺口先回账本补项;「有人在等」不再是定级判据。SKILL.md 净 0 行(813 / 813),两条新行以退掉两条重复指针付账,细则文件本身未动。

为什么改 — 章程原来的定级行把「清单区」写成判据,分诊按区命中就判「清单上的能力断了」,本班一批 8 张卡 8 张定 P0–P2、零关闭,再拿「有人在等吗」当平手裁决 —— 与您的裁定相反;两仓约 965 张卡的每一次首触定级都读这一行。

风险与代价(含回滚) — 纯文本改动,零代码、零工作流;派生门禁 21 / 21 绿。代价是退掉的两条指针(指向 dispatch-runbook.md 的第四条、指向 platform-readings.md 的第三条),细节仍在那两个文件里、SKILL.md 里各自还有其它指针;若您认为不该退,恢复是一行 revert,但要同时退另一行才过棘轮。北极星 docs/NORTH-STAR.md 本 PR 未动:「清单上的能力断了 = 指得出会 fail 的清单项」这一句(以及断了的 P2 项继承 P2 而不是您第 2 条写的 P0/P1)由席位在聊天里向您提出,不在本 PR 里。

席位意见 —(留空,由席位定稿)

你要做的 — 本 PR 走 Tier S 席内复核落地,不需要您点击。可选一字:是否同意把「清单上的能力断了 = 指得出会 fail 的清单项,级随该项的 priority」写进北极星第 2 条(A 同意,席位另开 docs PR / B 不写,章程这行足够)。

Acceptance notes

• noted, not filed: North Star 优先级 rule 2 says 「清单上的能力断了 ⇒ P0/P1」 while 91 of the ledger's 264 items carry priority: P2; the charter now inherits the item's own field, so a broken P2 item grades P2. The North Star text is the maintainer's; the card offers him the clause in chat. carrier: the skills seat (chat), then a docs-only PR if he says yes.
• noted, not filed: the seat's own Path: lines written this shift under the area shape (removed_from_merge_queue fires on a SUCCESSFUL merge too — 3 of 5 on this board today were merges, so the event is not a failure signal #19343, skills(pm-dispatch): grading reads the checklist ITEM, not the area — Path: carries the item id, the card inherits the item's P, and 「有人在等」 is not a test #19155's grading comment, and the first batch of 16) are specimens, not errors to fix here; re-reading them under the new grammar is the seat's next grading round. carrier: the skills seat.
• noted, not filed: dispatch-gates.mjs refuses --commands together with --ran (「--ran renders a verdict on stdout; --commands and --json render the derivation. Pass one.」) — correct behaviour, but os-dev.md :137–:138 reads as if the --ran reconciliation is the same invocation with a file added; one word (「单独」) would save a dev a refused call. 承接者:无 (no PR heads for os-dev.md :137).
• noted, not filed: PR docs(pm-dispatch): land the maintainer's filing/merging directive — items 1 2 3 4 5 8 — as in-place charter deltas, net 0 per ratcheted file #19449's Acceptance notes already record that os-dev.md :48 keeps 「未演练漂移」 while SKILL.md :351 says 「休眠」; unchanged here. 承接者:无.

---

Generated by Claude Code

[os-steve]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 9f87053d8fc28b884a7cdbd1209356fcbadf137e

① Derived judgments

• Published accept set / public surface: none moves — the PM charter (.claude/skills/pm-dispatch/SKILL.md, +3 / −3, net 0 at 813 / 813, the three landed lines 117 / 113 / 112 B under the 120 B cap), one file; no twin (measured: core-rules.md and lanes/triage.md carry no grading-grammar line). check-governed-merges --pr 19459: GOVERNED — Tier S, this record is the landing carrier; queue landing by the seat. Judged correct.
• The fix, read line by line against the maintainer's ruling 「有人在等吗 是个伪命题」 and the filer's two-layer spec (5737998283): :363 「定级按北极星「优先级」1、2 条;首行 Path: <定义项> | <清单项 id/缺项> | P<路步> | none」 — the area slot gone, the definition item first, then the checklist ITEM id or 缺项, then the path step; :364 「定义项取北极星「做出来的是什么」或仪器卡的车队决定;无则关;none 非分诊缺口」 — the definition layer's closure (none ⇒ close) and the clause that keeps an instrument card gradable (the North Star's own 「仪器为车队服务」 question, without which 无则关 would close skills(pm-dispatch): grading reads the checklist ITEM, not the area — Path: carries the item id, the card inherits the item's P, and 「有人在等」 is not a test #19155 itself — a refinement the seat accepts); :365 「清单项会 fail 才算断,继承其 priority;缺项先补账本;⛔「有人在等」不是定级判据」 — the ruling written as the negative rule, scoped to 定级 so it does not collide with :351's 零拉动 filing rule or :756's design ordering (the dev's deviation from the claim's 「拉动」, correct under the two-细则 conflict rule). Two rule lines instead of the claim's one, because the claim's sentence measured 212 B against a 120 B cap — paid by retiring two pure second pointers (old :588, the fourth pointer to dispatch-runbook.md — three remain; old :698, the third to platform-readings.md — two remain), each section keeping its own rules. PR docs(pm-dispatch): land the maintainer's filing/merging directive — items 1 2 3 4 5 8 — as in-place charter deltas, net 0 per ratcheted file #19449's lines (:351, :357–:358, :370–:372, :460) untouched. Judged correct.
• Evidence, seat-run on this head in a detached worktree: check-skill-line-ratchet.mjs exit 0 (813 / 813), check-skill-id-lint.mjs exit 0, check-governed-prose.mjs exit 0; the acceptance greps head vs origin/main: 清单项 2 / 0, 定义项 2 / 0, 缺项 2 / 0, 有人在等 1 / 0 (the negative rule), 清单区 0 / 1, Path: 2 / 2; merge-tree vs origin/main 50bae37: 0 conflicts; CI on 9f87053: 29 check runs, 2 still running, 0 red. The dev's union 21 / 21 exit 0 with --ran reconciled, check:pm-dispatch-gates 1883 cases exit 0, the ledger measured (264 items, area.slug ids, priority on every item: P0 20 / P1 153 / P2 91), the removed_from_merge_queue fires on a SUCCESSFUL merge too — 3 of 5 on this board today were merges, so the event is not a failure signal #19343 Path: specimen re-read under both grammars.

② Semver level

None — nothing published; skip-changeset on the PR is the correct declaration. Judged correct.

③ Boundary flags

Two open_questions, both ruled here by the seat, ⛔ not the decision box: (1) 「拉动」 in the negative rule — A, keep as landed (the ruling's own phrase is 有人在等; an unscoped 拉动 clause would manufacture a conflict with :351 / :756). (2) the North Star clause the card offers the maintainer (「清单上的能力断了 = 指得出会 fail 的清单项,级随该项的 priority」 — rule 2 says P0 / P1 while 91 of 264 ledger items are P2) — A, the seat puts the sentence on the seat post's 待您 list; a docs-only PR on docs/NORTH-STAR.md follows his word, ⛔ never in this PR. All four mechanism assumptions confirmed (:363 as it stood; PR #19449 merged as be488ce and its lines untouched; the three gates green; GOVERNED Tier S). Deviations read and accepted (two lines for one, paid by two pure pointer retirements; 北极星 for the path spelling to fit 120 B; the lock's leaf-only build failing before the closure build; a vanished PID file re-recorded from the transcript). Four noted, not filed items: the North Star rule-2 / ledger-priority tension (carrier: this seat, via 待您), this shift's own Path: lines as specimens of the old shape (carrier: this seat's next grading round — re-read under the new grammar, ⛔ not rewritten), an os-dev.md :137–:138 wording (--ran runs alone) and the 休眠 / 未演练漂移 pair (already recorded).

Implemented-by: claude/issue-19155-grading-reads-the-checklist-item
Reviewed-by: session_017ETYWqMQD4qMtZzAGovWNi

VERDICT: PASS

---

Generated by Claude Code