docs(governed-guard): replace the retired sha pin with the ruled approval-record predicate

AGENTS.md

@@ -493,7 +493,7 @@ ls <dir>/* | wc -l                                   # 两个数不等 ⇒ 工
 - **CI 全绿、已 review 都不构成例外。** 这类文件是后续每一次 dispatch 读的操作规程,绿灯说明不了它该不该成为规程。
 - **发现自己已经挂上了怎么办**:把 PR 转回 **draft** 是唯一能可靠退出合并队列的动作 —— 只调 `disable_pr_auto_merge` 会摘掉 auto-merge 但**不取消队列成员资格**,两个都要做。⚠️ 只回收**你自己**挂上的:本仓多 agent 共用同一 GitHub 身份,不是你设置的状态就属于别的 actor —— 去问、去报告,别替他回退。
 
-**本仓的机械兜底只有一件,而且它现在只报告、不拦截 —— 别读成一道拦得住的门,也别再读成「什么都没有」。** 本仓仍然没有 CODEOWNERS(核实:仓内不存在该文件),受管面上也没有钩子;但 `.github/workflows/governed-surface-guard.yml`(check 名 `Governed Surface Queue Guard`,判定逻辑在 `scripts/check-governed-queue-guard.mjs`)**是活的**:`pull_request` 腿是早期告警、**故意 exit 0**(受管 PR 停在 draft 正是健康终态,所以**绿不等于不受管**),`merge_group` 腿才是会拒绝的那条 —— 它要求 `GOVERNED_APPROVERS`(`os-zhuang` / `hotlong`)的 APPROVED review **钉在当前 head**。⚠️ 但它**尚未**是 required context:ruleset 开关只有维护者能翻(#6596,`pm:awaiting-maintainer`),**在翻转之前,那条拒绝腿只报告、不阻止队列**。事后一侧:`../objectstack` 的 report-only 合并后审计(`scripts/pm/check-governed-merges.mjs`)自 objectstack#9619 起**已覆盖本仓**(四个受管仓一次扫完),它把受管面的合并列出来,但同样不阻止任何事。⇒ 违规不再完全静默,但**仍然没有任何东西会替你拦下它**,这条规则的效力主要还是在于你读到了它并照做。**⛔ 别再把本段当成兜底工具的完整清单** —— 覆盖面以脚本自己的 `GOVERNED_SURFACES` 为准(它随树变化,本段不会);⚠️ 该清单曾与上面的受管面清单**并不一致**(脚本的集合含已发布 `skills/**`),这一分歧**已裁**:维护者第 5 场决裁批 #7 采 **Option A**(#6866 评论 5469339478)—— 已发布 `skills/**` **受管**,守卫的读法才是裁定的那个;该裁决**已随本段上方的清单落地**(#6866):上面五项已含 `skills/**`,与脚本的 `GOVERNED_SURFACES` 一致,曾经那条「仓根 `skills/**` 不受管、自行入队」的豁免**已作废**,⛔ 别再照它行事。
+**本仓的机械兜底只有一件,而且它现在只报告、不拦截 —— 别读成一道拦得住的门,也别再读成「什么都没有」。** 本仓仍然没有 CODEOWNERS(核实:仓内不存在该文件),受管面上也没有钩子;但 `.github/workflows/governed-surface-guard.yml`(check 名 `Governed Surface Queue Guard`,判定逻辑在 `scripts/check-governed-queue-guard.mjs`)**是活的**:`pull_request` 腿是早期告警、**故意 exit 0**(受管 PR 停在 draft 正是健康终态,所以**绿不等于不受管**),`merge_group` 腿才是会拒绝的那条 —— 它要求 `GOVERNED_APPROVERS`(`os-zhuang` / `hotlong`)里某个账号的一条 latest-decisive APPROVED review,**留在哪个 commit 上都算**;DISMISSED 与被顶掉的批准(同一 reviewer 后续给了 CHANGES_REQUESTED)不算,该集合之外账号的 APPROVED 不算,review 列表为空或读不到则 fail closed —— 判定读的是**有没有一条人工批准记录**,不问它是对哪些字节给的(维护者 2026-09-04 裁,逐字未译:「你的门禁有问题，只需要有人工批准记录就行，不需要卡最新的提交。」;sha pin 是**退休**不是放宽,守卫里已没有任何判定读 `commit_id`)。⚠️ 已接受的代价:批准之后的 push 不再被这道门重审,一个已批准的受管 PR 可以带着批准者没读过的字节落地 —— 维护者接受这一点,而这道拒绝先印的补救仍是转回 draft、交人类合并。⚠️ 但它**尚未**是 required context:ruleset 开关只有维护者能翻(#6596,`pm:awaiting-maintainer`),**在翻转之前,那条拒绝腿只报告、不阻止队列**。事后一侧:`../objectstack` 的 report-only 合并后审计(`scripts/pm/check-governed-merges.mjs`)自 objectstack#9619 起**已覆盖本仓**(四个受管仓一次扫完),它把受管面的合并列出来,但同样不阻止任何事。⇒ 违规不再完全静默,但**仍然没有任何东西会替你拦下它**,这条规则的效力主要还是在于你读到了它并照做。**⛔ 别再把本段当成兜底工具的完整清单** —— 覆盖面以脚本自己的 `GOVERNED_SURFACES` 为准(它随树变化,本段不会);⚠️ 该清单曾与上面的受管面清单**并不一致**(脚本的集合含已发布 `skills/**`),这一分歧**已裁**:维护者第 5 场决裁批 #7 采 **Option A**(#6866 评论 5469339478)—— 已发布 `skills/**` **受管**,守卫的读法才是裁定的那个;该裁决**已随本段上方的清单落地**(#6866):上面五项已含 `skills/**`,与脚本的 `GOVERNED_SURFACES` 一致,曾经那条「仓根 `skills/**` 不受管、自行入队」的豁免**已作废**,⛔ 别再照它行事。
 
 ### 服务纪律(本仓库与 `../objectstack` 多 agent 并行开发)
 

content/docs/guide/ci-cd-pipeline.md

@@ -40,7 +40,7 @@ one has its own section below.
 | `shell-escape-residue.yml` | Shell Escape Residue Scan | Push / PR to `main`, `develop` — **no path filter**; merge-queue builds; manual | **Yes** — when a fenced block in `AGENTS.md`, `CLAUDE.md`, `skills/**` or `content/docs/**` carries the enumerated machine-produced shell escape, or a scan root fails to resolve |
 | `readme-exports.yml` | README Export Check | Push / PR to `main`, `develop` — **no path filter**; merge-queue builds; manual | **Yes** — when a `packages/**/README.md` imports a name from its own package that the package does not export, or the scan's population collapses |
 | `docs-route-eager-closure.yml` | Docs Route Eager Closure Check | Push / PR to `main`, `develop` — **no path filter**; merge-queue builds; manual | **Yes** — when a package named in `apps/site/app/components/registerCatalogBlocks.ts` is not already reachable from the docs route's module graph (exit 1), or when the gate's own gauge cannot be trusted (exit 2) |
-| `governed-surface-guard.yml` | Governed Surface Queue Guard | PR to `main`, `develop` (incl. `ready_for_review`) — **no path filter**; merge-queue builds | **Yes on a queue build only** — a governed-surface diff with no authorized approval pinned to the PR's current head is refused there; on the pull request itself it is deliberately green and prints an early warning |
+| `governed-surface-guard.yml` | Governed Surface Queue Guard | PR to `main`, `develop` (incl. `ready_for_review`) — **no path filter**; merge-queue builds | **Yes on a queue build only** — a governed-surface diff with no authorized approval record (on any commit) is refused there; on the pull request itself it is deliberately green and prints an early warning |
 | `performance-budget.yml` | Bundle Analysis | Push / PR touching `packages/**`, `apps/console/**`, `pnpm-lock.yaml` | **Yes** — the console entry gzip budget |
 | `live-e2e.yml` | Live E2E (informational) | PR to `main`, `develop` (code paths); nightly cron `30 6 * * *`; manual | No — informational lane, `continue-on-error` |
 | `labeler.yml` | Auto Label PRs | PR `opened`, `synchronize`, `reopened` | No |
@@ -1251,10 +1251,17 @@ case forever — and a permanently red check is one everybody learns to ignore.
 build** the same finding is a refusal: that is a state a governed pull request should never be in at
 all, so red there is red on the anomaly.
 
-**What clears the queue leg** is an `APPROVED` review by an account in `GOVERNED_APPROVERS` whose
-`commit_id` equals the pull request's *current* head sha. The sha pin is what makes the approval an
-approval *of something*: a push after the approval goes stale and reopens the refusal, so a
-clearance cannot outlive the bytes it was given for. Dismissed and superseded approvals never count.
+**What clears the queue leg** is a latest-decisive `APPROVED` review by an account in
+`GOVERNED_APPROVERS`, on whichever commit it was left. Dismissed and superseded approvals (a later
+`CHANGES_REQUESTED` by the same reviewer) never count; an `APPROVED` review by an account outside
+that set never counts; an empty or unreadable review list fails closed. The predicate is the
+existence of a human approval record, and says nothing about which bytes it was given for — the
+maintainer ruled the sha pin out on 2026-09-04, quoted verbatim and untranslated because rewriting a
+ruling is rewriting the ruling: 「你的门禁有问题，只需要有人工批准记录就行，不需要卡最新的提交。」
+The pin is retired, not softened: no predicate reads `commit_id`, there is no stale bucket, and the
+pull-request head read that existed only to feed the pin is gone with it. **The accepted cost**,
+stated out loud rather than left to be discovered: a push after an approval is no longer re-reviewed
+by this gate, so an approved governed pull request can land carrying bytes its approver never read.
 The remedy the refusal prints **first** is not approval at all — convert the pull request back to a
 draft and leave the merge to the maintainer.