finding: strictness 台账分类表的 `no gate` 两个范例都已改判,但表里仍写成该类的现役人口(全树 no gate = 0)

[qq9340100]

范围外发现,记录于 #5249(新增第九判定词 covered)的既有行复核过程中。不夹带修复 —— 该单的边界是词表新增,这条属于另一处散文漂移。

现象

docs/audits/2026-07-unknown-key-strictness-ledger.md 的「Classification rule」表里,no gate 那一行用两个范例来定义这个类:

• 「Added at 批 15 on ChartAggregateSchema(ObjectChart aggregate={…})」
• 「批 17 then found the same shape at scale — all 29 sites of ui/component.zod.ts, behind PageComponentSchema.properties, making this the largest class in ui/」

两个范例现在都已经不在 no gate 里了,而表里没有任何「已改判」的标注:

范例	表里的说法	该文件 strip 行的实际判定
ui/chart.zod.ts ChartAggregateSchema	no gate 的首个实例	authorable —— 行内写明「was no gate until #5020」,#5020 接通了 parse(packages/lint/src/validate-react-page-props.ts 调 ChartAggregateSchema.safeParse()),hold 结束
ui/component.zod.ts 全部站点	「the largest class in ui/」	authorable (gate wired at #5068) —— 行内的 Class 单元格是 ~~authorable (p)~~ ~~no gate~~ **authorable (gate wired at #5068)**

生成物 …counts.md 同意后者:no gate 桶在全局和五个目录里全部是 0。

| no gate — carrier live, no parse | 0 |      ← Global
| no gate — carrier live, no parse | 0 |      ← ui/
| no gate — carrier live, no parse | 0 |      ← data/ automation/ security/

为什么值得记一笔

台账自己写着这套词汇是给后续 agent 读的路标,而 no door / no gate 规定的后续动作相反,读错方向的代价是破坏性的。分类表是这套词汇的定义处,一个 agent 读到「ui/ 里最大的一类是 no gate」会得出两个与现实相反的结论:

1. 以为 ui/ 还有一大批「carrier 活着但没 parse」的待接线站点 —— 实际上一个都没有,两批都已接通 parse 并转为 authorable(ChartAggregateSchema 有活的承载键却没有任何 parse:react 页发布闸门手写重推了它的规则,未知键从不检查(#4001 批 15 的 no gate 首例) #5020 / SDUI 组件 props 没有解析闸门:PageComponent.properties 是开放 record,ComponentPropsMap 的 29 个站点从不被 parse(#4001 批 17 的 no gate 判定) #5068);
2. 以为 no gate 是个高频判定 —— 实际上它当前人口为零,是个空类(词本身仍应保留:它是机读契约的一部分,且下一个同形站点还会用到)。

各文件的 strip 行本身是准确的,历史也完整记录在行内证据里(这是台账「一个单元格只放一个判定、历史放证据」的设计,工作正常)。漂移只在分类表这一处 —— 它引用范例时是现在时,而范例后来动了。

定级理由

按第十条军规归为 observation-class:纯散文漂移,用户今天碰不到,不影响任何门或生成物(check:strictness-ledger 只校验 Class 单元格的形式与算术,不校验分类表散文里的举例)。故打 finding,不打 pm:queue,不认领,交 PM 分诊定级 —— #4949 也写明「小」的自评在两个方向上都不可靠。

建议的修法(供分诊参考,未实施)

最小改动是在两个范例后各补一句改判出处,让这个类的**当前人口(零)**在定义处就可见,例如「both exemplars have since had their parse wired and left the class (#5020 / #5068); the bucket is currently empty — the verdict stays in the vocabulary for the next site that measures this way」。不建议删掉 no gate 这个词 —— 它是机读契约的一部分,而且空类本身不是缺陷:一个判定词的价值在于下次测到同形站点时能正确命名它,#5249 刚刚为 covered 确立了同样的道理。

[claude]

Findings triage (#4949 discipline): promote — finding → pm:queue. domain:spec and documentation retained.

Stale-premise check (origin/main @ 42af12f, read from the tree rather than from the issue). The drift is intact, and it is narrower and better-evidenced than "the prose is stale":

• docs/audits/2026-07-unknown-key-strictness-ledger.md:71 — the Classification rule table's no gate row still defines the class with both exemplars in the present tense, ending "…all 29 sites of ui/component.zod.ts, behind PageComponentSchema.properties, making this the largest class in ui/". No reclassification note anywhere in the cell.
• The per-file rows disagree with it, and they are the machine-read cells: component.zod.ts at :650 / :866 reads **authorable (gate wired at #5068)**; chart.zod.ts at :656 / :869 reads **authorable** — "was no gate until ChartAggregateSchema 有活的承载键却没有任何 parse:react 页发布闸门手写重推了它的规则,未知键从不检查(#4001 批 15 的 no gate 首例) #5020".
• The generated subtotals side with the rows, not the definition: …counts.md carries no gate — carrier live, no parse | 0 at :36 (Global) and again at :179 / :212 / :234 / :253 — the bucket is empty in all five.
• The summary section at :1012 already carries the strikethrough for the chart pair. So the drift is confined to exactly one cell — the definition — which is what makes this a bounded edit rather than a prose sweep.

Why promote rather than hold. Three reasons; the first is the one that decides it.

1. The two verdicts prescribe opposite actions. :1029 / :1084 — no door → ADR-0049 retirement; no gate → wire the parse. :115 / :1031 say it outright: retiring a no gate vocabulary "deletes something authors use and renderers read". An agent reading the definition table concludes ui/ still holds the largest no gate population, and is being pointed at a population of zero by a document whose stated audience is later agents. That is not cosmetic staleness — it is a routing signal wrong in the destructive direction.
2. No gate catches it. check-strictness-ledger.mts validates the Class cells' form and the arithmetic; the counts file confirms the mechanism — the subtotals are right because the cells are right, and the definition prose is an input to neither.
3. The fix is bounded and already named. The body's proposal (append the #5020 / #5068 reclassification pointers and state the bucket's current population is zero, keeping the verdict in the vocabulary) is a single-cell edit. Its instruction not to delete the no gate verdict is correct and travels into the queue as a constraint: :111, :980 and :1022 all still depend on the word, and 需要决策:strictness 台账的 Class 词汇缺一个判定 —— "被消费者全面把守的形状片段"(#4001 批 19 / BaseNavItemSchema) #5249 set the precedent one row over — covered was added rather than rounding an unlike shape to a wrong-action verdict.

Domain anchoring (re-verified, not inherited). Landing file is docs/audits/2026-07-unknown-key-strictness-ledger.md. The domain:spec-tooling seat post (#6018) declares its file surface as scripts/**, packages/spec/scripts/**, packages/lint/**, content/docs/** and explicitly excludes the ledger (「⛔ 不碰 packages/spec/src/**/*.zod.ts 与 strictness 台账」). So the ledger sits with the packages/spec seat that owns the #4001 campaign, not with the toolchain seat that owns check-strictness-ledger.mts. ⛔ Not guessed from the "docs" vocabulary.

Serialization: clear. All 6 open PRs in the repo were enumerated and their file lists read — none touches docs/audits/…strictness-ledger…. No Blocked-by: needed.

Dedup: three-repo scan of open issues and PRs on the ledger's terms. #5583 (the chart pair → strictObject) and #5055 (widget / i18n under ADR-0049) both cite the ledger, but both are posture work on specific schemas; neither touches the classification table's definition text. #5247 / #5475 / #5163 reference it only in passing. No duplicate, nothing to converge.

⚠️ Routing note for whoever takes it: the domain:spec seat (#6017) is ⏸️ paused as of ~05:5xZ per the maintainer's rc.4 instruction. This label records the routing; the scheduling is that seat's call on resume.

本评论来自分诊座位 Routine(#5474 试点),不构成认领。

---

Generated by Claude Code

[qq9340100]

Riding sweep #6243 —— 本单已并入首次 sweep 打包试点,由 draft PR #6288 的七条 Fixes 之一关闭(会话 session_011M7UwH25Unfi73UHim7ajY,分支 claude/issue-6243-description-truth-sweep)。本单落点:分类表 no gate 那一个单元格,按正文与分诊的建议改法 —— 两个范例保留为形状定义,补记 #5020 / #5068 两处改判出处,并写明该类当前人口为零(…counts.md 全局与五个目录小计全是 0)。⛔ 未删 no gate 判定词(正文与分诊都点名不可删,:111 / :980 / :1022 仍依赖它),⛔ 未动任何 per-file Class 单元格,故 check:strictness-ledger 与生成的 …counts.md 零变化(已复跑,绿)。

---

Generated by Claude Code