docs(pm-dispatch): three measured container readings into the platform table (419 to 422) - #17317

Fixes #17308

Two measured container facts from the triage seat's handover, landed as terse rows in
.claude/skills/pm-dispatch/references/platform-readings.md — deduplicated against what the
table already says, which is why three lines land rather than the seven the readings offered.

The rows, verbatim, with the section each landed in

## API 配额 — inserted directly under the gate-open channel row (「门开会话实测通道宽于标签…」),
which is where its neighbours live:

- 标签事件流 `GET /issues/{n}/events` 归门态:200 带 labeled/unlabeled 名与时刻,三容器两通一 403。

## 读数陷阱 — inserted inside trap ④'s shallow-checkout family, between the trap row
(「④:浅检出上的历史读数不可信…」) and its remedy row (「同族 branch -r --contains 零输出;先 --deepen 再判…」),
so the family's trap-then-remedy ordering survives:

- ④ 里最静默的是窗口式 `git log --grep`:回 0 结果、exit 0、无警告,读作该提交从不存在。

## 读数陷阱 — inserted directly under the check-governed-merges row
(「check-governed-merges 浅克隆上拒答而非少报…补救 git fetch --shallow-since=。」), the row it extends:

- ⛔ 不越过该拒答自行枚举:短清单读作合规;加深日期取窗口起点之前,不猜深度。

Assumption C measured rather than assumed: the file has exactly six ## sections and no
核验 / 树读数 section — the git/tree readings live in ## 读数陷阱, so that is where two of the
three landed. Widest row: 117 bytes / 101 display columns, against the file's widest existing
bullet (114 columns) and the ratchet's 120-byte line cap. No table rows.

去重计数 — 候选 7 / 落地 3 / 已有 4 / 拒收 0

⚠️ A dispatch premise was falsified on the tree before any line was written. The dispatch (and
the claim comment's premise check) recorded that the file holds ZERO lines for shallow,
history-horizon or --shallow-since. It does not: :261/:262 carry the shallow-checkout trap
family (spelled 浅检出) and :381 carries the check-governed-merges refusal together with the
literal git fetch --shallow-since= remedy (spelled 浅克隆). The zero-hit list was an ASCII grep
against a table written in Chinese. Fact 2 therefore lands as a two-line residue appended to rows
that already exist, not as the fresh pair the grading assumed — which is the seat's own instruction
(「a duplicate row is a 拒收」) applied to the half it had measured as absent.

Fact 1's residue is smaller for the same reason: rows 3 and 4 above already carry the
authenticated quota and the per-seat probe rule, so the landed row states only what is missing —
the endpoint spelling, what a 200 carries, and the measured two-through-one-403 spread.

Ratchet arithmetic

• CEILINGS row for platform-readings.md: 419 → 422, the landed count, headroom 0, same
  convention as every entry above it. ⛔ No other ceiling moves.
• A ninth ruledRaises record on the cross-file-move declaration: delta: 3,
  date: '2026-09-10', quoting the standing one-file exception from pm-dispatch SKILL.md
  verbatim and untranslated — the same words the eighth record quotes, quoted again rather than
  cross-referenced, because a record quoting no ruling is RED and each record stands alone.
• was: 314 does not move. Records now sum 34 + 3 + 26 + 7 + 7 + 17 + 3 = 97, so the move's
  own raise is 422 − 314 − 97 = +11, unchanged, against a net source decrease of 20. The gate
  prints exactly that:

✓ check-skill-line-ratchet: cross-file move into .claude/skills/pm-dispatch/references/platform-readings.md: +11 (314→422, less 97 lines of ordinary ruled raise) against a net source decrease of 20 (...); authorised by #14685 item 5 (comment 5520452691).
✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/platform-readings.md is 422 lines (ceiling 422; headroom 0).

• Nothing was paid in place; no re-wrap funded anything (re-wrap funding is refused in any case).
  The accounting comment beside the ceiling names each landed line and each deduplicated candidate.

Gates

Derived with node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack
(39 commands; change set read from the merge base by the script itself, not from a hand-written diff).
38 of 39 run green; the 39th is still running and is reported below rather than guessed at.

governed-surface predicate: 1 of 2 path(s) hit the register (5 surfaces, repo-agnostic).
  ⛔  GOVERNED — a human merge is the review record for this PR (#9495 regime).
      No seat flips it ready, enqueues it, or arms auto-merge (AGENTS.md Prime Directive #14).
      One hit governs the whole PR — 「混合 diff 一条命中即整 PR 分叉」; proportion is not a question.
      .claude/** ×1 — the agent instruction tree (skills, agents, hooks, settings)
        - .claude/skills/pm-dispatch/references/platform-readings.md
  paths not on the register: scripts/pm/check-skill-line-ratchet.mjs

⚠️ Reading worth carrying: the dispatch stated scripts/pm/** is on the governed register. The
predicate says it is not — only .claude/** hit. The PR is governed either way (one hit governs
the whole diff), so nothing about the terminal changes; the wrong half of the premise is recorded
here rather than repeated.

pnpm --filter @objectstack/lint run check:doc-formula-expressions first exited 3 =
PREREQUISITE NOT MET (@objectstack/formula and @objectstack/lint not built) — NOT MEASURED,
not a finding. Built both under the shared verify lock
(VERDICT command-exit 0 · held the lock 2s · waited 0s) and re-ran: exit 0.

No build and no package tests are owed: the diff touches no package. No changeset is owed either
— nothing published moves (.claude/** and scripts/pm/** are both outside every package's
files[]), so skip-changeset applies.

维护者速读(草稿)

改了什么 — 平台事实表 platform-readings.md 加三行:① 标签事件流 GET /issues/{n}/events
在门开会话可读(回 200 带 labeled/unlabeled 的名与时刻,三个容器实测两通一 403);② 浅检出上
窗口式 git log --grep 回 0 结果、exit 0、无警告,读作该提交从不存在;③ 遇到
check-governed-merges 的拒答,正确处置是按窗口起点之前的日期加深,而不是自己去枚举。
配套把该文件的行数上限从 419 抬到 422(落地行数),并在棘轮脚本里记一条引常设裁决的 ruledRaises。

为什么改 — 这两个事实这一班各自造成过一次错误读数。其一:有席位把「REST 403」当成整个机群的
属性继承下来,于是放弃了本可用的第二条回读通道;其二:一次调查据窗口式 git log --grep 的零结果
差点公开发布「该提交从不存在」的结论,而那个零只是提交落在浅检出的地平线以下。事实表就是给这类
读数兜底的地方 —— 写下来,下一个席位才 grep 得到。

风险与代价(含回滚) — 风险很低:改的是 agent 指令树里的只读事实表,不进任何发布包,不改运行时
行为。代价是三行预算(419 → 422),按常设裁决走,免决策卡。回滚 = revert 本 PR 一个提交,行数上限
随之回到 419,棘轮与其余门禁自动恢复绿。

席位意见 —

你要做的 — 治理面 PR,合并是人工的:请直接读上面三行中文原文是否准确、是否值得占三行预算,认可
就合并。⛔ 不需要席位翻转 ready、不需要入队、不需要挂 auto-merge。

Acceptance notes

• Scope held. Two files touched, exactly the surface the claim named: the readings table plus
  its CEILINGS row and one ruledRaises record. Nothing in SKILL.md, core-rules.md,
  lanes/hotcrm.md (sibling dev's surface) or skills/**, scripts/check-*.mjs,
  scripts/pm/dispatch-gates.mjs, docs, packages/** (the other sibling's).
• Mechanism assumption A held: the ratchet's ruledRaises path accepts the raise and its
  --self-test stays green (157 cases) with the ninth record in place.
• Mechanism assumption B held: no card number appears in either edited file's prose;
  check:pm-skill-id-lint reports 27 files clean. The ruling is cited by its wording, as the
  eight records before it do.
• Mechanism assumption C measured, and it decided placement — see the section list above.
• noted, not filed: :312–:315 (PR-body footer eaten with its rule line) and :321–:323
  (create-side appends a footer only when the sent tail is not already that block) are two
  readings of the same write path that a reader must reconcile by hand; both are hedged by
  :316/:328 (「⛔ 不由任一条推其余,写后必回读」), so this is a legibility observation, not a
  defect, and it is outside this card. 承接者:无 — no queued card and no in-flight PR touches
  the footer block; it would sit unowned if filed.
• noted, not filed: the dispatch's zero-hit premise list and the claim comment's
  Premise checks: line were both drawn from an ASCII grep over a Chinese table. That is a PM-side
  measurement habit, not a repo defect, and it has no file to be fixed in. 承接者:无.

Generated by Claude Code