docs(pm-dispatch): an enumeration that cannot say it was truncated may not report clean

[os-litant]

Fixes#12817

What was wrong

The two-tier triage inventory section (references/dispatch-runbook.md, maintainer ruling
2026-08-20) tells each hourly read-method what to enumerate — sweep 析取, 解锁扫描反演,
健康指标 — but never required any of them to check that it had finished. A list read that
returns fewer items than totalCount proceeded silently, and the scan's own output could not
distinguish a complete reading from a partial one.

Measured 2026-08-27: an enumeration ran at perPage=50 against a totalCount of 51. One card
was never examined. Nothing printed distinguished "read all of them" from "read all but one" —
the caller saw a list and proceeded.

This is the same collapse class the tree-staleness cards name: an incomplete reading and a
complete reading print the same thing.

What changed

Three sentences appended to the paragraph that already specifies the hourly layer's three
read-methods — the completeness predicate lands where the enumerations are defined, not as a
second procedure, and appending costs no separator line:

• every enumeration compares its returned count against totalCount;
• a shortfall means the round 没读完 — it reports sweep INCOMPLETE naming the gap, and
  ⛔ may never be reported clean, the standard check-governed-merges.mjs already applies to
  unaudited repos;
• pagination 只治当次 — it cannot fix the next list to outgrow the page size, which is why the
  rule is one predicate and ⛔ not a second code path.

Out of scope and untouched, per the 2026-08-24 ruling recorded in platform-readings.md: where
Blocked-by: lives, and the body-only read direction. No issue numbers were written into the
protocol text — the entry carries a date, per the file's own 行文纪律.

Ratchet

The file sits at its ceiling with zero headroom, so the rule is paid for by deletion in the same
file. No ceiling was raised.

file	ceiling	before	after	headroom
.claude/skills/pm-dispatch/references/dispatch-runbook.md	274	274	274	0

Verdict line, from the union run on the final commit:

check-skill-line-ratchet: .claude/skills/pm-dispatch/references/dispatch-runbook.md is 274 lines (ceiling 274; headroom 0).

Cut ledger — net 0 (+3 rule, -3 deleted content)

⚠️ The first commit on this branch (b4160a5) funded the rule by re-wrapping orphan
continuation lines with no prose deleted. That is banned outright by the ratchet's governing
clause, quoted verbatim (maintainer 2026-08-17, SKILL.md):

⛔ re-wrap(折行合并)不得用作筹行 —— 棘轮治理的是内容体量,行数只是机读代理,新增以删减付账;密度优化仅随净减内容的 PR 顺带

Content volume net-increased while the line count stayed flat — the exact shape the clause
forbids. Commit ac9ceed reverts every re-wrap: those five paragraphs are byte-identical to
origin/main again, and the rule was compressed from six lines to three. The funding is now
real deletion, each cut with a surviving home:

cut	lines	surviving home
the 落卡分析模板 + 业务写法六项 pointer bullet	2	references/decision-analysis.md carries the template, the six writing requirements, the 四棱块 shape and both ruling dates (中文 2026-08-19 at its line 17; 业务角度 2026-08-20 verbatim at its line 3); SKILL.md already routes readers there. The bullet was a routing stub, which the file's own header excludes — it holds 细则增量, not signposts to other files.
the 惰性判定 back-pointer	1	references/seat-post-protocol.md states the same lazy/standing split from the other end — 「惰性半边在接管冲突时评估一次 … 常设半边是分诊轮的座位贴活性巡查,同一判据」 — and points back to this file for the details the deleted line merely re-announced.
new rule text	-3	appended to the existing paragraph (no separator line needed)

The resulting diff against origin/main is 4 insertions / 4 deletions in one file, with no
re-wrapped line anywhere in it.

Serialization with PR #13106

PR #13106 also edits this file and is accepted ahead of this one; it lands first. Its cuts are
四件补遗 items ② and ④, the pnpm --workspace-concurrency sentence, and the rejection-envelope
clause. This PR double-claims none of them — verified against its branch: it keeps the
落卡分析模板 pointer bullet as unchanged context, and never touches the 惰性判定 line in the
座位贴活性巡查 section. The two cut sets are disjoint, and the file is 274/274 under either
order, so the net-0 arithmetic holds whichever lands first.

At the time of writing #13106 had not merged, so this branch is merged up to origin/main
d48929e instead. Its new bullet lands in the same section this PR deletes a bullet from, so a
textual merge may need a hand at landing time.

Gates

Derived with node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (no paths
passed; the script took the change set from the merge base itself — 1 path). All eight ran as
one && chain on the final commit e215a64, exit 0 — each verdict quoted from the gate's
own output, never from a bare shell status.

gate	verdict line
check:pm-skill-ratchet	dispatch-runbook.md is 274 lines (ceiling 274; headroom 0)
check:pm-skill-id-lint	23 file(s) clean (pattern /#[0-9]{3,}/g)
check:agent-test-spelling	0 violations — 398 file(s)
check:doc-authoring	392 files clean — no bare metadata literals
check:doc-formula-expressions	22 record-scoped formula example(s) across 425 files / 1453 TS blocks judged clean
check:pm-governed-merges	--self-test: 206 assertions
check:skill-frame-sync	4 copies of the decision frame are structurally isomorphic across 3 files
check:nul-bytes	OK (scanned 7285 text file(s) ... no raw ASCII control bytes)

@objectstack/formula and @objectstack/lint are built first in the same chain —
check:doc-formula-expressions otherwise exits PREREQUISITE NOT MET, which is the gate
declining to measure, not a red.

Changeset

None. The diff is .claude/ protocol text only and publishes nothing, so the PR carries
skip-changeset.

Generated by Claude Code