[objectstack-fleet]

Fixes #24444

Clause-②: no

What

Two lines of .claude/skills/pm-dispatch/references/decision-analysis.md, nothing else (1 file, +2 / −2):

• line 9 (the director's pre-batch increment): 「推荐/回退/口径」 → 「推荐/口径」, the card's wording verbatim.
• line 26 (the recommendation item of the card-face template): 「推荐 + 回退 + 置信缺口:荐一项、给回退项、明说看不见什么;⛔ 不沿用他席推荐。」 → 「推荐 + 置信缺口:只荐一,余各一句不取理由;盲区明说,⛔ 不因此另荐;⛔ 不沿用他席推荐。」

The four facets, the 自检 line and the 裁后执行 paragraph are untouched; SKILL.md is untouched.

Why

The maintainer, 2026-10-19: 「你的决裁总是有一个回退合理吗？是完全不同的开发方向。最终结论应该只有一个吧？是 skills 的问题吗」, then 「同意」 on the card's change. Ruling of record: director ledger comment 6259199455 (on #12708).

Route A: compressed in place, line count unchanged

The card's line-26 text measures 150 bytes; the ratchet's MAX_LINE_BYTES is 120 (scripts/pm/check-skill-line-ratchet.mjs), and the file sits at 50 lines against its 50-line ceiling (headroom 0). So the ruled wording cannot land verbatim on one line. Route A (compress the wording, keep every clause, add no line) loses no clause, so route B (fold the verbatim text into two lines and raise the ceiling 50 → 51 under the ruling) was not taken and the CEILINGS row is unchanged.

Clause survival, ruled text → landed fragment:

Ruled clause (card text)	Landed fragment	Change
item name 推荐 + 置信缺口	推荐 + 置信缺口	verbatim
① 只荐一项	只荐一	measure word 项 dropped
② 其余选项各一句不取理由	余各一句不取理由	其余选项 → 余 (the remaining options, set against 一)
③ 看不见什么明说	盲区明说	看不见什么 → 盲区 (blind spots)
④ ⛔ 不因此给第二推荐	⛔ 不因此另荐	给第二推荐 → 另荐 (recommend another); 因此 and ⛔ kept
⑤ ⛔ 不沿用他席推荐	⛔ 不沿用他席推荐	verbatim, byte-identical to the old line's tail

Nothing is added: no new clause, no new term, each clause keeps its own separator and its own ⛔. A shared-⛔ variant that kept 只荐一项 whole (…盲区明说,⛔ 不因此另荐,不沿用他席推荐。, 119 B) was weighed and refused, because it lets 因此 read across clause ⑤.

Line budget

	before	after	ceiling
file lines	50	50	50
line 9 bytes	119	112	120
line 26 bytes	110	120	120

Control: with the card's verbatim text placed at line 26, the gate's own scanLineLengths reports [{"line":26,"bytes":150}]; on the landed tree it reports [].

Premise checks on origin/main f4755a68a

• The card's line numbers 13 and 30 are off: the two strings sit at lines 9 and 26.
• git grep -n "回退" f4755a68a -- .claude/skills/pm-dispatch/ .claude/agents/ gives 15 hits. Only these two are decision-analysis uses; the rest (os-dev wording ×4, SKILL.md line 249's ?? fallback, dispatch-runbook.md, execution-duties.md, lanes/triage.md, platform-readings.md ×3, reading-discipline.md) are untouched.
• No script, test or doc pins the old wording: git grep for 回退/口径, 给回退项 and 推荐 + 回退 finds only the two lines themselves.

Verification, all at head ddf97945b

The 17 families node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derives for this diff, each exit code captured before any pipe:

Command	Exit
node scripts/check-closing-keyword-parity.mjs	0
node scripts/check-closing-keyword-parity.mjs --self-test	0
node scripts/check-comment-mask-corpus.mjs	0
node scripts/pm/check-harness-current.mjs --self-test	0
pnpm --filter @objectstack/lint run check:doc-formula-expressions	0 (first run 3, PREREQUISITE NOT MET: @objectstack/formula and @objectstack/lint unbuilt in the fresh worktree; built under the verify lock, rerun 0)
pnpm check:agent-test-spelling	0
pnpm check:cross-package-test-inputs	0
pnpm check:doc-authoring	0
pnpm check:driver-memory-census	0
pnpm check:gitlink-declared	0
pnpm check:nul-bytes	0
pnpm check:pm-governed-merges	0
pnpm check:pm-skill-id-lint	0
pnpm check:pm-skill-ratchet	0, decision-analysis.md is 50 lines (ceiling 50; headroom 0), no line-length offender
pnpm check:refd-timer-probe	0
pnpm check:skill-frame-sync	0
pnpm check:watch-hint-literal	0

dispatch-gates --ran with the exit-coded record: 17 derived, 17 run, 0 NOT-MEASURED, 0 UNRUN (a derived zero). Also run: pnpm check:pm-settings-deny-roster exit 0 (its roster lives under .claude). node scripts/pm/check-governed-merges.mjs --branch answers GOVERNED, Tier S (exit 3, the governed code). Text-only change: no tests, no ablation. The type-check lanes, test:repo and pnpm lint belong to CI.

Acceptance notes

• Tier S (.claude/** only). This PR stays a draft; ready and queue belong to the owning seat after an isolated contract-tier review.
• No changeset: .claude/** publishes nothing; skip-changeset is declared with this PR's creation.

维护者速读(草稿)

• 改了什么: 决策分析写法删掉「回退项」。总监批前增量从「推荐/回退/口径」改为「推荐/口径」;卡面推荐条改为「推荐 + 置信缺口」:只荐一个方案,其余方案各用一句话说明为何不取,看不见的地方明说,但不因此再给第二个推荐。只动这两行,文件仍是 50 行。
• 为什么改: 您指出「最终结论应该只有一个」。旧模板要求每次附一个回退项,实际常是另一条完全不同的开发方向,等于把决定又推回给您。
• 风险与代价(含回滚): 纯文本,不发布,不影响产品运行。第 26 行为了压进每行 120 字节的上限做了措辞压缩(「其余选项」→「余」、「看不见什么」→「盲区」、「给第二推荐」→「另荐」、「只荐一项」→「只荐一」),五个分句全部保留,但不是卡上原文逐字。回滚 = revert 本 PR 的一个提交。
• 席位意见:
• 你要做的: 无。本 PR 属 Tier S,由席位隔离复核通过后入队落地。

---

Generated by Claude Code

[objectstack-fleet]

维护者速读(终稿,事后记录)· PR #24520 · domain:skills seat 1 · os-justin · session_01HEdP5tPhqQfzwRmKDxwn85 · 2026-10-19T23:03Z

• 改了什么: 决策分析写法删掉「回退项」。总监批前增量从「推荐/回退/口径」改为「推荐/口径」;卡面推荐条改为「推荐 + 置信缺口」:只荐一个方案,其余方案各一句为何不取,看不见的地方明说,不因此再给第二个推荐。只动两行,文件仍 50 行。
• 为什么改: 您 2026-10-19 的话:「最终结论应该只有一个吧」,随后对卡上改法「同意」。旧模板要求附回退项,实际常是另一条开发方向,等于把决定推回给您。
• 风险与代价(含回滚): 纯文本,不发布。第 26 行为压进每行 120 字节上限做了措辞压缩:「只荐一项」→「只荐一」、「其余选项」→「余」、「看不见什么」→「盲区」、「给第二推荐」→「另荐」;五个分句与两个 ⛔ 全在,但不是卡上原文逐字。回滚 = revert 一个提交。
• 席位意见: 接受路线 A。卡上原文 150 字节装不进 120 字节的行,文件又已顶满 50 行;压缩版五分句逐条可对照,不丢义、不添义。「只荐一」读起来略省,但在行预算下是最忠实的一版;若您更想要原文逐字,一句话即可换路线 B(折两行、上限 +1)。
• 你要做的: 无。已由 os-zhuang 批准并经队列合并。

席位核对

• 形态: draft、base main、首行 Fixes #24444、行首 Clause-②: no、assignee os-justin、skip-changeset 随创建同笔;1 文件 +2/−2;正文闭合关键词只有首行一处,另一个卡号 [PM seat] director(项目总监) — ⏳ vacant · 第 47 场(os-bill)已收班 · 第 43 场续场 5(os-zhuang)、第 44 场(os-project-manager)与第 46 场(huangyiirene)无收班简报 · 台账在评论 #12708 旁无动词。
• 前提实测(席位,origin/main f4755a68a): 两段原文在第 9、26 行(卡写 13、30,已过时);git grep "回退" 在 .claude/skills/pm-dispatch/ 与 .claude/agents/ 下 15 命中,决策分析用语只此两行。分支 tip 上实测:第 9 行 112 字节、第 26 行 120 字节、文件 50 行,与报告一致。
• 门禁 check-runs(head ddf97945b,2026-10-19T11:46Z): 38 条,26 success、12 skipped、0 failed、0 in_progress;check-expected-skips --pr 24520:OK,12 skipped 全在名单。dev 本地 17 条派生门禁全 0,--ran 17/17。check-governed-merges --pr 24520:GOVERNED,Tier S,4 行。
• 接受的偏差: 路线 A 的措辞压缩(见速读);doc-formula-expressions 首跑 exit 3 为前置未建,补建后 0。
• 范围外去向: dev 的中继 pr_create 读回行不列 labels 的观察 —— 平台事实,记座位贴 Notes,不立卡。
• 落地条款与实际落地: Tier S。本席派出的隔离 CONTRACT_REVIEW_TIER 复核子代理在 2026-10-19T11:50Z 前后撞上账号周额度墙(HTTP 429,resets 23:00Z),未落任何记录;本席会话同时停摆。其间 os-zhuang(GOVERNED_APPROVERS 账户)APPROVED 本 head(review 5573069631),翻 ready、挂 auto-merge、入队,队列于 2026-10-19T14:45Z 合并为 5876b5474。按 SKILL.md 〈路径分叉〉②,授权账户的批准与合并即审核记录;本记录为事后补录,⛔ 不是第二次放行。

ACCEPT

---

Generated by Claude Code

[objectstack-fleet]

Contract review — PR #24520 (card #24444), after-the-fact audit record rendered 2026-10-19T23:27Z

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: ddf97945bf8f22609b3986e4052794d044987a96
Local-runs: none
Scope: .claude/skills/pm-dispatch/references/decision-analysis.md + scripts/pm/check-skill-line-ratchet.mjs (the line-byte budget and the CEILINGS row the deviation rests on), .claude/skills/pm-dispatch/SKILL.md and .claude/skills/pm-dispatch/references/lanes/director.md (the two pointers at the template), scripts/pm/check-half-states.mjs, scripts/pm/check-prior-rulings.mjs and scripts/pm/post-stamped.mjs (the readers of decision-analysis comments)

Read: the card body and its three comments (claim, os-dev-report, landing), the director's ledger comment 6259199455 on #12708, the PR body, its one-file list, the diff against main at the head (base f4755a6), the 40 check-runs on the head, and the governed file at base and head via git show. Not read: the dispatch order, the seat's own conclusions, the PR's reviews. The PR has since merged as merge commit 5876b54; this record is the audit of that landing, asked for by the maintainer (「两张 PR 你可以重新复审」).

① Derived judgments

• D1 Public surface: none. .claude/** ships in no package and reaches no exports map; no schema, type or generated artefact moves. Right.
• D2 Governed rule, the ruling: the maintainer's words 「你的决裁总是有一个回退合理吗？是完全不同的开发方向。最终结论应该只有一个吧？是 skills 的问题吗」 then 「同意」, recorded on ledger 6259199455 as "a decision analysis recommends exactly one option; no 「回退项」". After the diff the template asks for exactly one recommendation, one sentence per other option on why it is not taken, the confidence gap stated and ⛔ never turned into a second recommendation, ⛔ no borrowed recommendation. The accept-set of a decision analysis narrows from "recommend plus fallback" to "recommend one". Right, and the whole of what the ruling orders.
• D3 Line 9: 「推荐/回退/口径」 → 「推荐/口径」, the card's text verbatim, 119 → 112 bytes. Right.
• D4 Line 26: the card's 「推荐 + 置信缺口:只荐一项;其余选项各一句不取理由;看不见什么明说,⛔ 不因此给第二推荐;⛔ 不沿用他席推荐。」 measures 150 bytes; the landed 「推荐 + 置信缺口:只荐一,余各一句不取理由;盲区明说,⛔ 不因此另荐;⛔ 不沿用他席推荐。」 measures 120. Clause by clause: 只荐一 = 只荐一项 (measure word dropped); 余各一句不取理由 = 其余选项各一句不取理由 (余 reads as the remaining options, set against 一); 盲区明说 = 看不见什么明说; ⛔ 不因此另荐 = ⛔ 不因此给第二推荐; ⛔ 不沿用他席推荐 verbatim. Nothing added, no clause lost, each ⛔ on its own clause. Right. One narrative inaccuracy with no effect on the rule: the PR body says each clause keeps its own separator, but the separator between clauses ① and ② is 「,」 where the card wrote 「;」; the resulting grouping (recommendation pair ; gap pair ; independence) reads soundly.
• D5 Nothing else in the file implies a fallback: the 骨架 line names 选项 / 推荐 singly, line 28 already forbids 分期半量 as a recommendation, the 自检 line and the 裁后执行 paragraph are untouched. Right.
• D6 Nothing else in the repo carries the retired term: at base, 「推荐/回退/口径」, 「给回退项」, 「推荐 + 回退」 and 「回退项」 occur only on the two changed lines; at main after the merge, nowhere. Of the 15 回退 hits under .claude/skills/pm-dispatch/ and .claude/agents/ at base, the 13 that remain are shell, ??, promote and rewrite fallbacks, not the decision-analysis term; a sweep for fallback in other words (备选, 次选, 退路, fallback, 第二推荐, B′) finds only REST and MCP fallbacks. Right that the diff stops at two lines.
• D7 Pointers and readers: SKILL.md line 231 and lanes/director.md line 53 point at the template without naming its items (director.md's 「四件」 is the 受管草稿批 four of line 12, unchanged). The machine readers of decision-analysis comments read the facets block (check-half-states H62), the Prior-rulings paste line (check-prior-rulings) and a frame plus options plus a recommendation letter (post-stamped decisionAnalysisSignals); none reads or requires a 回退 clause, so their accept-sets are unchanged. Right.
• D8 Ratchet: 50 lines before and after against the 50 ceiling, CEILINGS row untouched, which is the route the ratchet header prefers (compress in place rather than raise); MAX_LINE_BYTES is 120 and the widest line is 120; the file's MAX_TABLE_ROW_BYTES pin of 0 holds (no table row). Gate verdict: Lint · PM dispatch-gates success. Right.
• D9 Check-runs on the head: 40, all completed; 28 success, 12 skipped, none failed. Governed Surface Queue Guard success, Check Changeset success, every Lint and Type Check lane success. The gate verdicts are green.

② Semver level

• .claude/** publishes nothing from any released package; the PR carries skip-changeset, no changeset file is in the diff, and Check Changeset is green. Correct: that label is for exactly this class of diff.
• Clause-②: no on the PR body, no (widening) or (narrowing) arm. Correct: no accept-set of a published surface moves.

③ Boundary flags

• F1 (claim) The card and the ledger cite lines 13 and 30; the strings sit at lines 9 and 26 at base, and the dev edited those. Answered: the text identity governs, the numbers were stale; no action.
• F2 (claim and landing comment) The card's acceptance says draft until the maintainer's hand merge; the claim planned the SKILL.md 路径分叉 route (isolated review, then the seat queues); what happened is an authorised approval and merge by os-zhuang, an account in GOVERNED_APPROVERS, with the queue guard green on the head, while the seat's dispatched at-tier review never posted (account limit). Answered: the landing matches the card's own acceptance line and SKILL.md's 「授权批准 ⇒ 席位落地」; this record is the at-tier review, rendered after the fact at the maintainer's request. It does not adjudicate landing conduct (AGENTS.md feat: Comprehensive CRM example demonstrating all ObjectStack protocol features #14: the post-merge audit does, no seat judges it) and finds nothing in its inputs to file.
• F3 (deviations 1) Route A, compressed line 26 instead of the card's verbatim text. Answered, accepted: the ruling is the maintainer's words and the card's five clauses, not a byte string; the verbatim string cannot land on one line under the binding 120-byte rule, and the card's one-file surface excludes route B; every clause survives (D4); the dev disclosed it in the PR body, in the 维护者速读 draft and in the report.
• F4 (deviations 2) The 119-byte shared-⛔ variant was refused. Answered: right refusal; a shared ⛔ would let 因此 govern clause ⑤ and change its meaning.
• F5 (deviations 3) The doc-formula-expressions gate needed a lock-held build before it could measure. Answered: immaterial; the check-run conclusions on the head are the gate verdicts and the Lint lanes are green.
• F6 (deviations 4) Commit trailers are the model-free pair, not the harness's model-named trailer. Answered: correct per AGENTS.md (the pre-push hook refuses a model identifier in the pair); the head commit carries Claude-Session and Co-authored-by only, and no model identifier appears in the PR title or body.
• F7 open_questions: none.
• F8 (out_of_scope_findings) The relay's pr_create read-back names assignees but never labels. Answered: the label landed (the PR's labels read documentation, size/xs, skip-changeset by REST); the gap is relay tooling outside this card and is passed to the seat as a candidate tooling card, no action owed by this PR.

Implemented-by: claude/issue-24444-one-recommendation
Reviewed-by: session_01HEdP5tPhqQfzwRmKDxwn85

VERDICT: PASS

---

Generated by Claude Code