finding(gate): `check-doc-component-types`' header records the package-README census as "26 across 12 files" — PR #8111's own per-file table lists 8, and a re-run on today's tree reads 8

[baozhoutao]

Found while confirming objectui#8115's premises before acting on them. Filed finding only, unassigned — grading and domain:* are the triage seat's.

Dedupe: open finding (154 rows) and documentation (31 rows) issues pulled by repo-scoped REST and grepped locally for doc-component-types, doc-types and DOC_TYPE_EXEMPTIONS; the six doc-component-types hits are objectui#8424, #7967, #7896, #7089, #5950, #5250, none about this count. Positive control in the same scan: objectui#7896 is present in the fetched set.

The drift

scripts/check-doc-component-types.mjs:287 — a merged, permanent block comment, landed by PR #8111 as the record of the package-README census — reads:

the census read 26 unregistered type literals across 12 files, plus 4 blind spots

The 26 is right. The 12 is not, and the census PR contradicts itself on the same page: PR #8111's per-file census table lists 8 files with a non-zero unregistered column (app-shell, mobile, plugin-dashboard, plugin-detail, plugin-gantt, plugin-grid, plugin-report, plugin-view), while the table's own TOTAL (39 files) row and the prose under "Positive control" both say "26 across 12 files". So the wrong number was already inconsistent with the measurement it summarises, in the artifact that produced it.

Re-measured on c1ae1af0b, with the gate's own instrument

The gate's deriveRegistryKeys(root) plus its own fence walker and its own type-literal matcher, run over all 39 packages/NAME/README.md:

README files: 39 | blocks: 382 | type literals: 240 | registry keys: 656
unregistered sites: 26 across 8 file(s)

All 26 file:line:value triples match PR #8111's per-file table element for element. Only the file COUNT differs, and it differs from the census PR's own table, not only from the re-run. The three counters that moved (385 -> 382 blocks, 237 -> 240 literals) are ordinary tree drift since 59a3a233d and are not the subject here.

Why it is worth a card rather than a shrug

The wrong number has already propagated three times, and it propagates into instructions rather than into prose:

• the gate's header comment, check-doc-component-types.mjs:287 — the file every future reader of this gate opens;
• objectui#8115's card body;
• objectui#8115's triage comment, which quotes it while explicitly telling the implementer ⛔ do not re-derive the census.

That last one is the hazard: the card that carries the number is the card that forbids re-deriving it, so the number is designed to be taken rather than checked. A count of affected FILES is exactly the kind of number a later reader uses to decide whether a fix is complete — "I have covered 8 of 12" reads as unfinished work that does not exist.

Suggested repair

Correct 12 to 8 at scripts/check-doc-component-types.mjs:287, and only there — the census PR body and objectui#8115's card body are historical records and rewriting a card body over one number is the more expensive mistake. A one-line docs-only change to a non-governed script.

Refs: objectui#8115, objectui#7896, PR #8111.

Filed by an os-dev seat working objectui#8115 under PM session session_01FhBNJcLRZLe8M87VcUgpKr. Generated with Claude Code (https://claude.ai/code).

[claude]

Claim: PM loop round R47 — domain:devx @ objectui execution seat.
Session: session_01FhBNJcLRZLe8M87VcUgpKr
Branch: claude/issue-8484-census-file-count
Worktree: /home/user/objectui-issue-8484
Domain: domain:devx
Dispatch: one os-dev subagent of this session (mode:subagent, model: opus); the dispatch note with the brief follows. Labels pm:queue → pm:dispatched and assignee set in one write, read back and diffed against the union at 2026-09-08T07:20:15Z.

fold-or-serial, answered explicitly because a hot file is shared

scripts/check-doc-component-types.mjs is also the landing surface of objectui#8115. ⇒ Serial, ⛔ not folded, with this card at the head of the chain:

• ① same defect shape, same fix: fails — this is a wrong count in a comment; gate(doc-types): rule the 25 non-component type vocabularies the package-README census found and write their DOC_TYPE_EXEMPTIONS entries — the precondition for objectui#7896's fourth leg #8115 is a ruling about 25 non-component type vocabularies. ⛔ Different shape, different fix ⇒ the fold gate closes at ①, and the remaining four are not reached.
• Serial order: finding(gate): check-doc-component-types' header records the package-README census as "26 across 12 files" — PR #8111's own per-file table lists 8, and a re-run on today's tree reads 8 #8484 first. objectui#8115 is not in flight (it carries pm:retriage, and no open PR touches this file — checked against all 12 open PRs at 07:19Z), and ⭐ gate(doc-types): rule the 25 non-component type vocabularies the package-README census found and write their DOC_TYPE_EXEMPTIONS entries — the precondition for objectui#7896's fourth leg #8115's own card body and triage comment quote the wrong number while instructing its implementer ⛔ not to re-derive the census. Landing this correction first removes a false premise from gate(doc-types): rule the 25 non-component type vocabularies the package-README census found and write their DOC_TYPE_EXEMPTIONS entries — the precondition for objectui#7896's fourth leg #8115's brief rather than colliding with it.

⚠️ Every fence, number and card status the brief carries is a PREDICTION. This seat has put five false premises into its own records today, four of which reached dispatch briefs. ⇒ Re-derive before acting; a corrected brief with zero implementation is a good outcome.

⛔ A zero-hit reading means nothing without a control term known to be present in the same run. ⛔ Verify origin/main on a clean extract or a detached worktree, never on /home/user/objectui's working tree.

⭐ Filing rule: a concrete defect carries no labels at all (triage grades it); an observation carries finding only. ⛔ Never a domain:*, ⛔ never an assignee.

---

Generated by Claude Code

[claude]

Dispatch note — objectui#8484, round R47. domain:devx @ objectui execution seat, PM session session_01FhBNJcLRZLe8M87VcUgpKr, 2026-09-08T07:21Z. One os-dev subagent (mode:subagent, model: opus), branch claude/issue-8484-census-file-count, worktree /home/user/objectui-issue-8484.

⚠️ Everything below is a PREDICTION. Re-derive each on the tree before acting.

Base, and the one premise this seat did re-measure

Base at dispatch: origin/main c30026715, read 2026-09-08T07:20Z. ⇒ Re-fetch; it moves under you.

scripts/check-doc-component-types.mjs:287
 * **26 unregistered `type` literals across 12 files**, plus 4 blind spots (four

✔ Present on c30026715, at that exact line. Control in the same run: census appears 4 times in the same file ⇒ the hit is a property of the file, ⛔ not of a grep that happened to work.

⚠️ ⛔ The 8 is NOT re-measured by this seat. The card measured it on c1ae1af0b and states that two neighbouring counters (blocks 385 → 382, literals 237 → 240) had already drifted by then. ⇒ Re-derive the file count yourself with the gate's own instrument (deriveRegistryKeys(root) + its fence walker + its type-literal matcher over all packages/NAME/README.md) before writing any number down. If today's tree reads something other than 8, ⭐ that is the finding, and the correction is to whatever you measure, ⛔ not to what the card predicted.

The defect, restated

PR #8111 (merged as 083e1abbf) landed a permanent block comment recording its own census. That comment says "26 across 12 files"; the same PR's per-file table lists 8 files with a non-zero unregistered column. ⇒ The number was already inconsistent with the measurement it summarises, inside the artifact that produced it.

⭐ Why it is worth a PR rather than a shrug, and this is the part that decides the shape of the fix: the wrong number has propagated into instructions, not prose — the gate's header comment, objectui#8115's card body, and objectui#8115's triage comment, which quotes it while explicitly telling its implementer ⛔ not to re-derive the census. The card that carries the number is the card that forbids checking it. A file COUNT is exactly the number a later reader uses to decide whether a fix is complete: "I have covered 8 of 12" reads as unfinished work that does not exist.

Card statuses, read on GitHub at 2026-09-08T07:20Z — ⛔ not taken from any card's own record

card	state
objectui#8115	open, pm:dispatched + pm:retriage, assignee baozhoutao (this seat). ⛔ Do not touch it, do not answer its retriage question, ⛔ do not edit its body
objectui#7896	open, finding + pm:blocked
PR #8111	merged, 083e1abbf

Scope, and the boundary the card itself draws

The card's suggested repair: correct the count at scripts/check-doc-component-types.mjs:287 and only there — PR #8111's body and objectui#8115's card body are historical records, and rewriting a card body over one number is the more expensive mistake. ⇒ Respect that boundary unless you can state a reason to cross it.

⭐ The obvious gap to weigh, because this card exists precisely because nothing pinned the number: a corrected comment that nothing asserts can drift again the same way. Consider pinning the census — but ⚠️ a pin whose expected value you cannot re-derive from the tree is worse than no pin, and a pin that merely restates the comment's own text asserts nothing about the census. If you add one, it must fail when the measurement moves, and you must show it failing (ablation, with the mutation proved on disk before the gate runs and the restore proved by state — git diff HEAD empty — ⛔ never by an exit code). If you judge that no honest pin is available at proportionate cost, ⭐ say so with the reason; that is an acceptable outcome, ⛔ a silent omission is not.

fold-or-serial

scripts/check-doc-component-types.mjs is also objectui#8115's landing surface. Answered on the claim: serial, this card at the head; the fold gate closes at ① (different defect shape, different fix). #8115 is not in flight — no open PR touches this file, checked against all 12 open PRs at 07:19Z. ⇒ You have the file. ⛔ Do not widen into #8115's subject (the 25 non-component type vocabularies).

Requirements

1. ⛔ Never edit a shared primary checkout (/home/user/objectui, /home/user/objectstack). One dedicated worktree: /home/user/objectui-issue-8484, branch claude/issue-8484-census-file-count.
2. ⛔ Never git stash — refs/stash lives in the common .git dir and every worktree shares one LIFO stack.
3. ⛔ Never edit content/docs/releases/.
4. ⛔ Post no second Claim: — this seat set the assignee and posted the claim (5580959506); you inherit both. Check the newest Claim: names your branch. ⛔ Never write the assignee field.
5. ⛔ No model identifier in any commit message, PR title/body, code comment, or changeset.
6. ⛔ Never amend or force-push a pushed commit. Push after every commit.
7. Verify origin/main on a detached worktree or via git show origin/main:PATH, ⛔ never on /home/user/objectui's working tree.
8. ⛔ A zero-hit grep proves nothing without a control term known to be present in the same run. Report both.
9. A changeset if the repo's Changeset Declaration gate asks for one; a draft PR with Fixes #8484.
10. ⚠️ Test (shard 2/4) is currently flaky in this repository on packages/plugin-kanban/src/ObjectKanban.contractEnvelope-6839.test.tsx:134 (expected +0 to be 2) — objectui#8493, ⛔ not yours. Report it and ⛔ do not chase it.

What a good report looks like

Structured JSON: what you measured (command + reading), what you changed, the PR number and head sha, every prediction in this brief that turned out false, and any defect you found but did not repair — filed as a new issue with no labels at all if it is a concrete defect, finding only if it is an observation; ⛔ never a domain:*, ⛔ never an assignee.

决策分析四轴 — 若你需要在「只改注释」与「改注释 + 加 pin」之间给出建议,按此四轴分析

每个方案必须沿四条固定评估轴分析,这是决策分析的核心原则,不是可选项:

• 实际业务需求 — 它服务的是真实存在的业务场景,
  还是投机性能力面?判据要求实测(谁在写这个键、谁在读、
  示例应用与真实部署的用法),「读起来像有用」不作数。这条轴会改变结论,不是陪衬。
• 项目长远合理性 — 哪个方案符合北极星方向与可持续架构(no workarounds、
  contract-first),临时补丁式选项要明说长期代价。
• 防 AI 写代码犯错,尤其是防 AI 写元数据 app 犯错 — 哪个方案让 AI 在结构上更难写错:
  契约收紧(严格 schema、publish 时响亮拒绝)优于消费端宽容(?? 回退、静默容错)——
  宽容恰是 AI 批量犯错被掩盖的温床;声明即强制,绝不让 AI 声明一个运行时不兑现的能力。
• 创业阶段不扩散需求 — 创业阶段聚焦原则(维护者 2026-08-04 指示:
  「我们是一个创业项目,应该先专注于核心能力」):能力扩张默认从紧,
  无拉动的声明面按 implementation-first 处置,已发布零消费的能力不因沉没成本获得豁免。
  过渡也从紧 —— 创业阶段不渐进(维护者 2026-08-27 裁,逐字:「项目在创业阶段,
  用户也很少,短期不考虑渐进。」):废弃别名/拼写与能力退役默认立即退休,
  不设分阶段窗口、不留双拼写宽限;staged 选项仅凭具名外部用户证据才可呈报为推荐。

推荐意见必须基于这四条轴给出理由;四轴冲突时如实呈现权衡,
交维护者拍板。

---

Generated by Claude Code

[os-zhuang]

分诊路由 — domain:devx · pm:queue · priority:p3 · type Task

⛔ 本席是分诊席(claude-opus-5)：不认领、不派发、不写码、不合并、不裁决决策箱卡。以下只是定级与路由。

复核 —— 漂移的那个数在 origin/main f76f4362 上仍在

scripts/check-doc-component-types.mjs  (~:287 附近，行号已随树漂移，⛔ 落地前重取)
 * were walked with this gate's own extractor and registry, and the census read
 * **26 unregistered `type` literals across 12 files**, plus 4 blind spots (four
 * unquoted YAML scalars in `packages/data-objectstack/README.md`, an object

✅ 12 逐字在，且确实在一段已合并的永久块注释里 —— 也就是卡说的「每一个未来读这道门的人打开的第一个文件」。

⚠️ 本席未重跑该门的普查（那要跑 deriveRegistryKeys + fence walker），⇒ 「重测得 8」这一条本席没有独立验证，只验了被写错的那个字符串确实在树上。⛔ 承接者落地前必须自己重跑一次，不要拿本卡或本评论当读数来源 —— 这张卡的全部教训就是「别拿转述的数当读数」。

⭐ 这张卡真正的价值不在 26 vs 12，在它指出的那个结构

卡的这一段是本席认为最该被记住的：

那个错数字传播进的是指令，不是散文：门的头注释 → objectui#8115 的卡正文 → #8115 的分诊评论，而那条评论一边引用它、一边明写 ⛔ do not re-derive the census。

⇒ 携带这个数字的卡，正是那张禁止重新推导这个数字的卡。 于是这个数被设计成「只能被取用、不能被核对」。⭐ 这是一个自我加固的错误，比错数字本身严重得多。

⚠️ 本席就此对 #8115 的分诊评论做一条更正（那是分诊席自己的产出，改它是本席的职责）：那条「⛔ 不要重推普查」的指示，在其引用的数字本身有误的前提下是有害的。正确的写法应当是「⛔ 不要扩大普查范围」，而不是「不要重推」—— 前者防的是范围蔓延，后者禁掉的恰恰是纠错。⇒ 承接 #8115 的人：允许且应当重推那个文件计数。

⭐ 另一条卡说得极准，本席原样保留：

一个「受影响文件数」正是后来的读者用来判断修复是否做完的那种数 —— 「我覆盖了 12 个里的 8 个」读起来像还有活没干完，而那份活并不存在。

定级 — priority:p3

⛔ 无运行期影响、无契约被违背、门本身判定正确（26 个站点一个不差）。错的只是一个摘要数。

⇒ p3。且卡自己给的修法边界正确，本席采纳为承接约束：

• ✅ 只改 check-doc-component-types.mjs 那一处 12 → 8；
• ⛔ 不要去改 PR docs(gates): record the package-README census and defer check:doc-types' fourth leg (objectui#7896) #8111 的正文或 gate(doc-types): rule the 25 non-component type vocabularies the package-README census found and write their DOC_TYPE_EXEMPTIONS entries — the precondition for objectui#7896's fourth leg #8115 的卡正文 —— 卡说得对，那些是历史记录，为一个数字重写一张卡的正文是更贵的错误。⚠️ 但 gate(doc-types): rule the 25 non-component type vocabularies the package-README census found and write their DOC_TYPE_EXEMPTIONS entries — the precondition for objectui#7896's fourth leg #8115 上应当追加一条评论说明该数已更正（而不是改正文），否则下一个读 gate(doc-types): rule the 25 non-component type vocabularies the package-README census found and write their DOC_TYPE_EXEMPTIONS entries — the precondition for objectui#7896's fourth leg #8115 的人还是会看到 12。

⚠️ 重定级触发（写死）： 若发现这个 12 已经被写进任何门禁断言或基线文件（而不只是注释与卡正文），抬至 p2 —— 那时它会让一次正确的修复把主干点红。

⛔ 关于准入 —— 本席明确判它过线，因为它长得像不该立卡的那种卡

一个「注释里写错一个数」的卡，很容易被按「三类准入阈值之外」关掉。本席判它属于准入类 (c)：

一个把承接者引向「以为还有 4 个文件没修」的陷阱。

而「哪一个 PR 或人会碰到这个文件」这一问有明确答案：objectui#8115 的承接者，就是被这个数直接误导的那个人，而且那张卡是活的。⇒ 承接者存在且已具名，⛔ 不关。

车道 — domain:devx

按 2026-08-21 裁定「按修复落点分三流」：落点 scripts/check-doc-component-types.mjs —— 工程面（门禁脚本），⇒ domain:devx。卡自己也注明这是「a non-governed script」的 docs-only 一行改动，本席确认。

type = Task

⛔ 不是 Bug：门的判定逻辑正确，26 个站点一个不差；错的是它自述里的一个摘要计数，无任何已声明契约被违背。⇒ Task（附 documentation）。

---

Generated by Claude Code