docs(pm-dispatch): stop the review checklist from enumerating read-order tiers

[os-litant]

Fixes #13021

The second site of the stale three-tier read order. The parenthetical on the
"does the report contain an API read git could have answered?" checklist item quoted
git → REST → MCP/GraphQL — three tiers, with the zero-quota payload tier missing.

Shape chosen: (b) — keep the citation, stop enumerating tiers

The card offered two legal shapes. I took (b), the one the grading preferred, and it
reads clean. Two reasons, and the second is the one that decided it:

1. The citation was already correct; only the enumeration was wrong. The line cites
「dev 契约的读序」 — the dev contract, .claude/agents/os-dev.md. That is the right
authority for this checklist item, because the item grades a dev's report against the
dev's contract. Shape (a) would have re-pointed it at platform-readings.md, which is
the PM's own reference file — a worse citation for a dev-side obligation, adopted only to
make an enumeration copyable. Fixing the enumeration by moving the pointer would have
traded an accurate citation for a convenient one.

2. The enumeration was load-bearing for nothing. The item's criterion names the
offending reads explicitly on the next line (MCP list_issues/search_issues-class
GraphQL reads, or re-running the dispatched dedupe reading), and the justification it
needs from the read order is exactly one property: git goes first. The tail of the
ladder does zero work here. A pointer that lists no tiers cannot go stale again, which is
the whole failure mode this card exists to close — and it is the smaller maintenance
surface, since nothing mechanically holds a quoted ladder equal to its source.

No tier mechanics were copied into this file. The replacement reuses the house spelling
for the property — platform-readings.md states the same idea as git 先行 — so the
wording is the corpus's, not this seat's invention. The dev contract backs the claim in
its own words at .claude/agents/os-dev.md:58: 「卡与评论先走 git 与它」.

Before / after, with byte counts

Before (line 33 = 105 bytes, line 34 = 116 bytes):

- **报告 `tests`/证据里有 git 本可回答的 API 读吗?**(dev 契约的读序是 git → REST →
  MCP/GraphQL)判据:MCP `list_issues`/`search_issues` 一类 GraphQL 读,或重跑派发词已下发的去重读数

After (line 33 = 97 bytes, line 34 = 104 bytes):

- **报告 `tests`/证据里有 git 本可回答的 API 读吗?**(dev 契约的读序 git 先行)
  判据:MCP `list_issues`/`search_issues` 一类 GraphQL 读,或重跑派发词已下发的去重读数

Lines 35–36 of the bullet are byte-identical — the diff is 2 insertions / 2 deletions.

Ratchet arithmetic

reading	before	after	budget
file lines	84	84	ceiling 84, headroom 0 → net 0
line 33 bytes	105	97	cap 120
line 34 bytes	116	104	cap 120
file's widest line	120 (line 10)	120 (line 10)	cap 120, untouched

Both edited lines were re-measured at dev time rather than taken from the handover; the
handover's 105 B for line 33 held. The line-length reading is taken positively, using
the gate's own code rather than my arithmetic against a recalled cap:

gate MAX_LINE_BYTES = 120
edited line 33 = 97 bytes
edited line 34 = 104 bytes
file widest line = line 10 @ 120 bytes
scanLineLengths offenders = {"offenders":[],"exempt":{}}

Cut ledger: none, and none was owed. The edit spends no lines and no bytes — it is
net −20 bytes across two lines, adds zero lines, and moves no prose across a line
boundary, so there was nothing to pay for. No re-wrap was needed or performed.

Gates

Union derived mechanically, not recalled: node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (9 families matched; re-derived after the final commit and
unchanged). Run under the shared verify lock at head e9e88fc, exits captured before
any pipe. The four extras beyond the derived nine are the dispatch-named
check:pm-governed-prose and check:skill-frame-freshness, plus check:nul-bytes and
the standalone required-set patrol.

gate	exit	verdict line
check:pm-skill-ratchet	0	✓ ... review-checklist.md is 84 lines (ceiling 84; headroom 0). and ✓ ... widest table row is 0 bytes (pin 0; headroom 0).; ✓ check-skill-line-ratchet self-test: 111 cases pass.
check:pm-skill-id-lint	0	✓ check-skill-id-lint: 23 file(s) clean (pattern /#[0-9]{3,}/g).
check:pm-governed-prose	0	✓ check-governed-prose: 2 instruction surface(s) name all 5 registered governed surfaces ... and claim no others.
check:pm-governed-merges	0	✓ check-governed-merges --self-test: 206 assertions (...)
check:skill-frame-sync	0	✓ check-skill-frame-sync: 4 copies of the decision frame are structurally isomorphic across 3 files
check:skill-frame-freshness	0	✓ check-skill-frame-freshness: the decision frame in this tree is current with origin/main (fetched just now).
check:doc-authoring	0	✓ doc authoring guard: 48 published skill files clean — no internal issue-id references.
check:agent-test-spelling	0	✓ check-agent-test-spelling: 0 violations — 396 file(s) ...
check:required-contexts	0	✓ check-required-contexts: 6 required context name(s) pinned across 2 workflow(s) ...
node scripts/check-required-contexts.mjs	0	same verdict, run standalone as the patrol workflow runs it
check:doc-formula-expressions	0	✓ check:doc-formula-expressions: 22 record-scoped formula example(s) across 425 files / 1453 TS blocks judged clean by @objectstack/formula.
check:nul-bytes	0	check-nul-bytes: OK (scanned 7232 text file(s) ... no raw ASCII control bytes).

check:doc-formula-expressions first came back exit 1 with PREREQUISITE NOT MET — the workspace package @objectstack/formula is not built and its own line saying "Nothing was
measured". That is a NOT MEASURED reading, not a red gate: @objectstack/formula and
@objectstack/lint were built, and the gate then ran and passed. The green above is the
post-build run.

Scope

One file, two lines. No changeset — the surface is .claude/** internal agent tooling,
published nowhere; skip-changeset is applied. Governed surface, so this is draft
only: no ready flip, no reviewers, no auto-merge, no queue.

The card's closing section — that nothing mechanically holds a quoted policy equal to its
source, and a third site could appear the same way — is a design question and is
deliberately not addressed here.

Session: https://claude.ai/code/session_01MnijPVVDakqK2J335JoJtq

Generated by Claude Code

---

Generated by Claude Code