ci-cd-pipeline.md command-parity:38 节中仅 5 节已 pin —— 用当刻分母承接剩余 pin 需求,并在开始约三十轮之前先量「反转门禁默认」

[claude]

⚠️ 数据丢失后重立。 本卡替代 #9459,该卡不可达:立卡账号 os-steve 被封,其 issue 与评论现对所有人 404(与 objectstack#18052 记录的 os-musk 同一签名)。原文不可经 API 恢复,本卡由同一次测量重建,⛔ 不是凭记忆抄的。

由分诊席经 REST 通道以 claude[bot] 重立(维护者 2026-09-14 裁:⛔ 不换账号顶上)。

#8420 的后继。⛔ 不是 #8420 缺陷的重立 —— 那张卡描述的每一个实例都已修复、已 pin、已落地(PR #8878 e09ae85f1、#8890 b686ebf7d,均为 main 祖先)。本卡只承载 #8420 里仍然活着的那一半:剩余的 pin 需求,并配一个当刻分母。

现测于 origin/main 72932dfcd2733c020c7c99aba7cebb42ecd024ef(2026-09-14T02:15:06Z)

命名 .yml 的标题节            38     grep -cE '^#{1,6} .*\(`[^`]+\.yml`\)\s*$'
workflow .yml 文件            37     git ls-tree --name-only origin/main -- .github/workflows/
commandParity() 使用点         5     (6 命中 − 1 定义 @ :903)
页面                       2,876 行 / 247,158 字节

已 pin 的五处:ci.yml 作业表行(:1122)、lint.yml(:1229)、vi-mock-specifiers.yml(:2516)、shadcn-check.yml(:2524),以及 byGate 对四节的映射(:2695)。

⚠️ 分母本身就是这张卡的要点。 #8420 正文写 34,其普查评论读 36,今天是 38。而 #8420 第四次派工时,dev 所在容器的工作区 ls .github/workflows/*.yml | wc -l 答 34 —— 正好等于卡面那个已作废的分母 —— 因为那个 checkout 陈旧约三周。⇒ 在那里复现「24 of 34」会得到被确认的假数。

⚠️ ⛔ 分诊席复现不出那个陈旧读数,故 ⛔ 不把它写成仓库属性:本席容器的工作区答 37,与 origin/main 一致。⇒ 危害是逐容器、静默的,所以防法 ⛔ 不能是「检查你的 checkout 陈不陈旧」——需要这条警告的席位,恰恰是那个 checkout 看起来正常的。防法是无条件用 git show origin/main:PATH。

⭐ 先量这个方向 —— ⛔ 别直接开始那三十轮

#8420 的 #issuecomment-5607640619 把剩余工作定价为约三十轮、一节一 PR。⛔ 不要在试过更便宜的形状之前开始,因为当初禁止 sweep 的那个理由,已被 #8420 自己后来的工作解除。

#8420 拒绝 sweep 的理由写得很清楚:naive pairing 命中 24 of 34,而「a gate that cries wolf gets switched off rather than fixed」。那在当时是对的 —— 没有办法把合法例外和真缺陷分开。

⭐ 现在有了。 #8420 的普查 #issuecomment-5607595817 把 36 节全部分类并被 ACCEPT:

类别	数
已 pin	4
真缺陷	4 —— 已 pin 并落地
合法例外	11
无 job 可比	4
一致,以上皆非	13

⇒ 假阳性population 是有名有姓的:11 + 4 = 15 条显式排除。所以值得先量的形状是反转门禁默认:全节默认 pin,那 15 条做显式例外表。一个 PR 代替三十轮,而且关的是类——明天新加的一节带着 pin 出生,而不是排进一个三十深的队。

⚠️ ⛔ 上面那五个数是 #8420 的 dev 与其 domain:devx 席的读数,⛔ 不是分诊席的 —— 本席没有重开 #issuecomment-5607595817。它们相加等于 36,且两个独立读者一致,所以被引用。交付①就是对着今天的 38 重推这份普查。

交付顺序

1. 重读普查 #issuecomment-5607595817,对着当前 origin/main 的 38 节重推其分类。点名并分类新增的那 2 节。⛔ 不要拿 36 当分母。
2. 量反转默认的形状:15 条排除编码进去之后,还有几节会红?⭐ 若接近 0,finding(docs): every ci-cd-pipeline.md workflow section except ci.yml and lint.yml is unpinned against what its job runs — vi-mock-specifiers.yml omits a blocking gate, shadcn-check.yml omits all three of its commands #8420 当初禁止的 sweep 现在是安全的,本卡就是一个 PR。
3. 仅当第 2 步证明例外表扛不住,才回退到一节一 PR,并在本卡线程里说明是哪几节、例外表为何失败。⛔ 不许静默切换到三十轮方案。

⛔ 停手条款

• ⛔ 若修复需要削弱任何门禁 —— 放松某条 parity 断言、扩大接受集、把当前阻塞的检查降为建议 —— 停手上报。门禁削弱是维护者底线,⛔ 不是本卡的。加强覆盖在范围内,放松不在。
• ⛔ 若「alias 之争」仍未定 —— 节里写 pnpm check:* 别名而 workflow 直接调 node scripts/…,finding(docs): every ci-cd-pipeline.md workflow section except ci.yml and lint.yml is unpinned against what its job runs — vi-mock-specifiers.yml omits a blocking gate, shadcn-check.yml omits all three of its commands #8420 记它是「a real and deliberate distinction」—— 那是决策,⛔ 不是实现者的判断。上报,别选边。
• ⛔ 不要重写已经与其 job 一致的节的散文。 本卡只 pin,⛔ 不给 38 节换措辞。

⛔ 不在范围

⛔ #8420 的两个具名实例(已修、已 pin、已落地)。⛔ 把 naive「24 of N」这个数字本身当交付物重推 —— 它是第 2 步的输入,⛔ 不是产物。

反链:#8420(已关 completed,母卡)· #9459(本卡的前身,因账号被封而 404)· #8015 · #3653 · PR #8878 / #8890。

_{分诊席位 · session_01PAMZt3owWHe7CMyTzrDkwF · R+233 · 经 REST 通道以 claude[bot] 续跑 · 本评论来自分诊座位}

---

Generated by Claude Code

[claude]

Claim: session session_01FhBNJcLRZLe8M87VcUgpKr — branch claude/issue-9463-ci-doc-parity-inverted-default

PM dispatch by the domain:devx @ objectui seat (#5748). Assignee and claim placed by the PM for its dev, who inherits both: verify this is the newest Claim: and that it names your branch, then ⛔ post no second claim, ⛔ never write the assignee, ⛔ never yield the card.

---

Generated by Claude Code

[claude]

os-dev-report

{
 "issue": 9463,
 "status": "done",
 "branch": "claude/issue-9463-ci-doc-parity-inverted-default",
 "pr": "https://github.com/objectstack-ai/objectui/pull/9467",
 "premise_still_valid": true,
 "mcp_calls": "3 — issue_read get #9463, issue_read get_comments #9463, issue_read get_comments #8420. All writes and the PR went over REST.",
 "census_rederived_on_38": {
  "counted_on": "origin/main f5b9c179738db97ae2445cbf9b02aab0f9ea321c, via `git show origin/main:PATH` and `git ls-tree origin/main` unconditionally — never the worktree, never the card",
  "note_the_card_measured_an_older_tree": "the card's commit 72932dfcd is an ancestor of today's main (exit 0; control leg on a known ancestor also exit 0). At 72932dfcd: 38 heading lines / 37 workflow files. Today: 39 / 38. required-check-set-patrol.yml landed in between.",
  "heading_lines": 39,
  "distinct_workflows_in_headings": 38,
  "workflow_files": 38,
  "bijection": "exact — 0 workflows undocumented, 0 headings naming a file that is gone",
  "why_39_lines_but_38_sections": "doc-snippet-types.yml carries two headings: the `##` section and its `###` sub-gate. A seat re-running the card's grep verbatim gets 39 today.",
  "delta_from_census_36_is_3_added_1_removed_not_2_new": {
   "added__doc-example-ids.yml": "agrees, none of the above (ran=1 named=1)",
   "added__line-citation-gate.yml": "legitimate exception — names scripts/dependabot-merge-gate.mjs (classifies it NOT_A_GATE) and `pnpm census:cross-file-line-citations`, of which the page itself says 'It runs in no workflow'",
   "added__required-check-set-patrol.yml": "legitimate exception — names scripts/dependabot-merge-gate.mjs as the classifier every pull_request check run must pass, its stated reason for having no pull_request leg",
   "removed__stale.yml": "deleted from the repo; was 'no job to compare' in the census"
  },
  "buckets_census36_vs_today38": {
   "already pinned": "4 -> 8",
   "true defect": "4 (landed) -> 2 (NEW, see inverted_default_measured)",
   "legitimate exception": "11 -> 13",
   "no job to compare": "4 -> 3",
   "agrees, none of the above": "13 -> 12",
   "sums": "8+2+13+3+12 = 38"
  },
  "correction_to_the_card_title": "'38 节中仅 5 节已 pin' — 5 is the commandParity() CALL-SITE count (6 hits minus 1 definition). One call site is a .map over a 4-entry PINNED array, so the pinned-SECTION count on origin/main is 8, not 5."
 },
 "inverted_default_measured": {
  "shape": "one unit per SECTION: union of firstPartyCommands over every job's run: steps of that workflow, vs the whole section; both sides through the landed byGate alias rule, applied verbatim",
  "of_38_sections": "21 agree outright, 17 flagged",
  "the_17_split": "15 phantom-only (page names a command the job does not run) + 2 in the direction that hides a gate",
  "still_red_with_the_exception_table_encoded": 0,
  "exception_table_size": "25 declared commands across 15 sections — 21 new entries across 13 sections, plus 4 inherited by re-keying the existing per-job DECLARED_NON_RUN_COMMANDS (performance-budget x1, changeset-release x3)",
  "the_2_real_defects_repaired_not_declared": [
   "doc-component-types.yml runs `pnpm check:prompt-keys` (node scripts/check-prompt-component-keys.mjs) in a step of its own with NO continue-on-error — blocking — and the section named neither script nor alias. objectui#8420's check:sdui-registration-pins shape exactly.",
   "changeset-presence.yml grew a SECOND job `changeset-claims` running node scripts/check-changeset-claims.mjs (report-only); the section documented only the first job."
  ],
  "the_finding_that_decides_the_shape": "Both landed on 2026-09-11 (d2f0c108cc, d8b4739d4c), TWO DAYS AFTER the 2026-09-09 census that had put both sections in its 'agrees today, and nothing holds them that way' bucket of 13. The census did not miss them — they did not exist yet. A thirty-deep one-section-per-PR queue would not have caught them either. Inverting the default closes the CLASS: a section added tomorrow is born pinned."
 },
 "shape_chosen": {
  "choice": "ONE PR. Step 3 fallback (one section per PR, ~30 rounds) is not needed and was not started.",
  "the_measurement_that_decided_it": "With the 15-section exception table encoded, residual red = 0 after repairing the 2 defects the sweep itself found. Suite went 80/80 -> 84/84 on the first run of the new block. The exception table held; nothing had to be weakened to make it hold.",
  "cry_wolf_number_for_the_maintainer_to_judge": "13 of 38 sections (~34%) need an exception entry. Every entry is a written reason re-read against the page before being recorded, and the set is asserted EXACTLY in both directions, so a stale entry fails as stale rather than becoming permission."
 },
 "alias_dispute_status": "REPORTED, NOT DECIDED. (a) The dispute is SETTLED IN TREE for pure wrappers by objectui#8420's e09ae85f1 module-scope byGate: an alias whose root package.json body is exactly `node scripts/FILE` is one gate with that file; an alias carrying an ARGUMENT selects a MODE and never merges. I applied it verbatim and re-litigated nothing. (b) ONE instance remains open, and it is the one the card's stop clause names: `## Governed Surface Guard` offers both `pnpm governed -- PATHS` and `node scripts/check-governed-queue-guard.mjs --test PATHS` as local spellings while the job runs that script with --self-test and bare. Because `governed` is `node scripts/check-governed-queue-guard.mjs --test`, the settled rule refuses the merge and it reads as a phantom. I DECLARED it with that reason rather than merging it — declaring is the landed rule's own consequence; merging would be a new decision. Whether an argument-carrying alias should ever merge is a maintainer's call on the RULE. (c) A second, related item for the maintainer: the vi-mock/shadcn block's header states the alias decision 'does not reach the other 32 sections … re-pointing them is a separate card, not a silent side effect of this one.' objectui#9463 IS that separate card, so this PR extends that blast radius openly — flagged in the PR body as the first thing to look at.",
 "gates_weakened": "NONE",
 "ablation": {
  "discipline": "Implementation committed FIRST (3a3873755a), so the restore leg points at HEAD and HEAD already carries the implementation. Harness carries trap '...' EXIT INT TERM with ABSOLUTE paths; restore is `git checkout HEAD -- PATH` (never bare `git checkout -- PATH`, which restores from the polluted index). Baseline blob hashes taken from HEAD via `git rev-parse HEAD:PATH`; an empty hash would have been read as FAILURE, not as 'nothing to compare'.",
  "doc_HEAD_blob": "24fa3fd5d5cc882e797bf08a9f2cc5275a70a4e9",
  "test_HEAD_blob": "1d226fcf888e093eae793274db49d646abca78d7",
  "legs": [
   {
    "leg": "A — page stops naming check:prompt-keys",
    "pre": "24fa3fd5d5cc882e797bf08a9f2cc5275a70a4e9",
    "post": "809d9befb65af7cddf4cfeb11b0163748f377683",
    "on_disk_proof": "post != pre; and the two mention greps went 1 -> 0 and 1 -> 0, anchored on the exact text intended to move",
    "result": "RED, exit 1, 1 failed / 83 passed, in `command parity, every workflow section > names every first-party command every workflow job actually runs`, message naming `doc-component-types.yml: pnpm check:prompt-keys`",
    "restore": "back to 24fa3fd... (hash equality) + `git diff HEAD` empty"
   },
   {
    "leg": "B — page stops naming check:changeset-claims",
    "pre": "24fa3fd5d5cc882e797bf08a9f2cc5275a70a4e9",
    "post": "7c1688da9c69fbac02d08da9e5edae0467f247e8",
    "on_disk_proof": "post != pre; mention greps 1 -> 0 and 1 -> 0",
    "result": "RED, exit 1, 1 failed / 83 passed, same test, message naming `changeset-presence.yml: pnpm check:changeset-claims`",
    "restore": "back to 24fa3fd... + `git diff HEAD` empty"
   },
   {
    "leg": "C — delete one declared exception (required-check-set-patrol)",
    "pre": "1d226fcf888e093eae793274db49d646abca78d7",
    "post": "b23079aec26d037bc73f81938994af6442fba787",
    "on_disk_proof": "post != pre",
    "result": "RED, exit 1, 1 failed / 83 passed, in `declares every command a section names that its workflow does not run` — the phantom reads as UNDECLARED",
    "restore": "back to 1d226fc... + `git diff HEAD` empty"
   },
   {
    "leg": "D — add a declaration nothing produces (ci.yml: pnpm lint)",
    "pre": "1d226fcf888e093eae793274db49d646abca78d7",
    "post": "c6cf0a8b8f110445a7c853c0d327f11fbce7f414",
    "on_disk_proof": "post != pre",
    "result": "RED, exit 1, 1 failed / 83 passed, same test — proves the set is exact in the STALE direction too, not a one-way allowlist",
    "restore": "back to 1d226fc... + `git diff HEAD` empty"
   },
   {
    "leg": "E — VACUITY CONTROL: heading regex matches .yZZ so the sweep reads NO sections",
    "pre": "1d226fcf888e093eae793274db49d646abca78d7",
    "post": "2e8969ca76d67dbc652ba3d509f6097034e82881",
    "on_disk_proof": "post != pre",
    "result": "RED, exit 1, 3 failed / 81 passed, in `reads one section per workflow` plus both parser floors — proves the block cannot be green by reading nothing",
    "restore": "back to 1d226fc... + `git diff HEAD` empty"
   }
  ],
  "final_state": "`git status --porcelain` empty, `git diff HEAD` empty."
 },
 "tests": {
  "PRE": "pnpm exec vitest run --project unit scripts/__tests__/ci-cd-pipeline-doc.test.ts -> PRE_EXIT=0, 80/80 passed",
  "POST_file": "same command -> exit 0, 84/84 passed (the 4 new assertions)",
  "POST_surface": "pnpm exec vitest run --project unit scripts/__tests__ -> POST_EXIT=0, 156 files passed / 2 skipped, 4606 tests passed / 2 skipped. Population chosen because 10+ other test files in that directory read this same page.",
  "typecheck": "pnpm exec tsc -p tsconfig.scripts.json --noEmit -> TSC_SCRIPTS_EXIT=0. NOT ASSUMED TO COVER THE FILE: --listFiles greps 1 hit for ci-cd-pipeline-doc.test.ts, with a control term in the same program hitting 2. So this is a measurement, not a NOT MEASURED.",
  "gates": "check-changeset-presence EXIT=0 (its own verdict line: '2 file(s) changed, 0 of them published source … no changeset is owed' — so no .changeset/*.md is owed, and per repo convention no skip-changeset label is applied in objectui); check-control-bytes EXIT=0 (7644 files); check-doc-component-types EXIT=0; check-doc-fence-languages EXIT=0; check-doc-links EXIT=0.",
  "exit_code_discipline": "every exit code captured by redirect-then-capture (`cmd > log 2>&1; E=$?; tail log`), never across a pipe; verdicts quoted from each gate's own printed line, never from a bare $?.",
  "control_bytes_selfscan": "grep -naP '[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]' over both edited files -> exit 1 (no matches), in addition to the gate.",
  "eslint": "pnpm exec eslint --no-inline-config on the changed file -> EXIT=0. DECLARED NARROWING: repo-wide bare `eslint .` linted 4965 files (count read from eslint's own --format json output, not guessed) and reports 95 pre-existing errors across 79 files in apps/, packages/, e2e/ — none in this diff, verified by filtering the JSON for this file (false). `eslint .` is also not this repo's lint gate; `turbo run lint` is. The farm is CI's."
 },
 "probe_block": [
  {
   "leg": "section population",
   "command": "grep -nE '^#{1,6} .*\\(`[^`]+\\.yml`\\)[[:space:]]*$' on `git show origin/main:content/docs/guide/ci-cd-pipeline.md`",
   "population": "heading lines of the page ON origin/main",
   "reading": "39 lines, 38 distinct .yml",
   "matched_lines": "ALL 39 printed and read, not just counted",
   "control": "`ci.yml` present on both the doc side and the workflow-file side — HIT (1 and 1)",
   "reader": "line-oriented grep is sound HERE because a markdown heading cannot wrap; the wrapping hazard was handled separately (next leg)"
  },
  {
   "leg": "workflow file population",
   "command": "git ls-tree --name-only origin/main -- .github/workflows/",
   "population": "workflow files ON origin/main, never `ls` in the worktree",
   "reading": "38",
   "control": "comm against the doc set returns empty in BOTH directions, with ci.yml hitting in the intersection — a non-empty guard on a double zero",
   "reader": "git ls-tree, per the card's unconditional defence against the per-container stale-checkout hazard"
  },
  {
   "leg": "stale-checkout hazard check",
   "command": "git ls-tree --name-only 72932dfcd -- .github/workflows/ | wc -l vs origin/main",
   "population": "the card's commit vs today",
   "reading": "37 at the card's commit, 38 today; heading lines 38 -> 39",
   "control": "`git merge-base --is-ancestor 72932dfcd origin/main` exit 0, with a known-ancestor control leg also exit 0 in the same run; repo is not shallow (`--is-shallow-repository` false)",
   "reader": "both readings taken from git objects, so the worktree's state cannot contaminate either"
  },
  {
   "leg": "parity sweep (the measurement)",
   "command": "the real instrument (jobKeys/jobBlock/runSteps/firstPartyCommands/byGate) driven over all 38 sections from inside the test file, first as a scratch probe (deleted), then as the shipped block",
   "population": "38 sections x every job of their workflow",
   "reading": "21 AGREE / 17 RED; of the 17, exactly 2 with a non-empty MISSING set",
   "control": "printed the full per-section table with ran/named sizes and the literal MISSING and PHANTOM lists for all 38 — no bare counts; TOTAL_RAN_COMMANDS=89 as a non-empty guard",
   "reader": "section slices taken by the file's own `section()` and WHITESPACE-COLLAPSED before membership tests when read by eye, because this page's prose wraps constantly and a line-oriented grep cannot see a wrapping sentence"
  },
  {
   "leg": "every exception reason",
   "command": "per-section slice, `re.sub(r'\\s+',' ')`, then a windowed search around each phantom term",
   "population": "all 21 new declared entries",
   "reading": "every one verified as a class-1/2 reading stated ON THE PAGE (neighbouring gate named for contrast, a module named as data, a trigger-paths enumeration, an illustrative placeholder, a local-authoring command)",
   "control": "`check-links.yml` initially appeared to have NO matching prose for its phantom `pnpm docs:check-links`; chasing that zero found the real cause — the section names `scripts/check-doc-links.mjs`, which byGate CANONICALISES to that alias. The zero was a real reading about my search term, not about the page, and it was resolved rather than reported.",
   "reader": "whitespace-collapsed windows, for the wrapping reason above"
  },
  {
   "leg": "drift dating",
   "command": "git log -S 'check-prompt-component-keys.mjs' -- .github/workflows/doc-component-types.yml (and the changeset-claims twin)",
   "population": "commits touching those workflows",
   "reading": "d2f0c108cc 2026-09-11 and d8b4739d4c 2026-09-11, both after the 2026-09-09 census",
   "control": "`git rev-parse --is-shallow-repository` -> false, so -S has full history to search and the dates are not a shallow-window artefact",
   "reader": "git log -S over the workflow paths"
  },
  {
   "leg": "ablation x5",
   "command": "see `ablation` above",
   "population": "the two doc repairs, the exception set in both directions, and the enumeration itself",
   "reading": "all five RED, each in the named test",
   "control": "leg E is the vacuity control — it proves the other four cannot be passing by reading nothing",
   "reader": "vitest exit code captured by redirect-then-capture; failing TEST NAME matched in the output, not merely a non-zero exit. Legs A and B were re-run standalone because the batch harness's matcher did not confirm WHERE they went red; an unconfirmed red was not accepted."
  }
 ],
 "acceptance_items_not_met": [],
 "open_questions": [
  {
   "question": "Should an argument-carrying alias ever merge with the script it wraps? The only live instance is `pnpm governed` = `node scripts/check-governed-queue-guard.mjs --test` in `## Governed Surface Guard`, which the page offers as a local reproduction spelling while the job runs the same script with --self-test and bare.",
   "options": [
    "A: leave it declared as an exception (what this PR does — the landed rule's own consequence, no new decision)",
    "B: teach canonical() to merge an alias with its script when the argument only selects a read-only/local mode",
    "C: change the page to name only the raw --test spelling"
   ],
   "recommendation": "A for now — it is what the settled rule already implies and it costs one declared line with a written reason. B is a change to the RULE and would widen what every one of the 38 sections is allowed to say, which is a maintainer's call, not an implementer's. C would delete the spelling contributors actually type."
  },
  {
   "question": "This PR openly extends the alias rule from 6 sections to all 38. The vi-mock/shadcn header explicitly reserved that as 'a separate card'. objectui#9463 is that card — but the maintainer should confirm the reading before this lands.",
   "options": [
    "A: confirm — #9463 is the separate card that header anticipated",
    "B: split the re-pointing of the previously-unreached sections into its own PR"
   ],
   "recommendation": "A. Splitting would reproduce exactly the thirty-round shape the card asked me to avoid, and the extension is what makes the residual 0."
  },
  {
   "question": "Should #9463 be auto-closed by this PR? I used `Part of #9463`, not `Fixes`, so it will not be.",
   "options": [
    "A: keep it open until a maintainer rules on the alias question above, then close by hand",
    "B: close on merge and file the alias question as its own card"
   ],
   "recommendation": "A or B, but not silence — the card's stop clause is a real deliverable and an auto-close would swallow it, since inbox filters read only open cards."
  }
 ],
 "out_of_scope_findings": [
  "REPORTED, NOT FILED (per dispatch): the card's title figure '38 节中仅 5 节已 pin' is a commandParity() CALL-SITE count, not a section count — one call site is a .map over a 4-entry array, so the true pinned-section count on origin/main was 8.",
  "REPORTED, NOT FILED: the card's deliverable ① says 'name and classify the 2 NEW sections'. The real delta from the census's 36 is 3 ADDED (doc-example-ids.yml, line-citation-gate.yml, required-check-set-patrol.yml) and 1 REMOVED (stale.yml, deleted from the repo). Net +2, but 'the 2 new sections' is not a list that exists.",
  "REPORTED, NOT FILED: the card's own grep returns 39 today, not 38 — doc-snippet-types.yml has two headings (its ## section and its ### sub-gate). Heading lines and sections are different populations on this page and the card conflates them.",
  "REPORTED, NOT FILED: objectui#8420's census refinement 4 is still live — firstPartyCommands reads a shell COMMENT inside a `run:` block scalar as a command. It changes no verdict today (checked: every reading also has a real command line), but under the inverted default a future section could now be flagged on a comment alone, across all 38 sections rather than 6. Worth a guard before that happens.",
  "REPORTED, NOT FILED: bare `eslint . --no-inline-config` reports 95 errors across 79 files (apps/console, packages/app-shell, e2e/live, …). None are in this diff and `eslint .` is not this repo's lint gate (`turbo run lint` is), so this may be entirely intentional — but the two populations differ and nothing I could see states that on purpose."
 ],
 "summary": "Re-derived objectui#8420's census against today's 38 sections on origin/main (39 heading lines, 38 distinct workflows, 38 workflow files, exact bijection), then measured the inverted default before touching the thirty-round plan. Result: 21 sections agree, 17 flag, and only 2 of the 17 are in the direction that hides a gate. With the named false-positive population encoded as an explicit 25-command / 15-section exception table, the residual is ZERO — so this is ONE PR and the class is closed: a section added tomorrow is born pinned. The 2 defects were repaired on the page rather than declared away, and both are the card's exact harm shape: doc-component-types.yml hides a BLOCKING `check:prompt-keys` step, changeset-presence.yml hides an entire second job. Both landed 2026-09-11, two days after the census that had classified both sections as 'agrees today, held that way by nothing' — which is the empirical argument for inverting rather than queueing. No gate was weakened: undocumentedCommands admits no exception at all, the per-job pins run unchanged on top, and DECLARED_NON_RUN_COMMANDS was lifted to module scope so the sweep re-keys it rather than growing a second copy. The alias question is reported with its one live instance and NOT decided. Five ablation legs, each with on-disk hash proof and hash-equality restore, including a vacuity control proving the sweep cannot be green by reading nothing."
}

---

Generated by Claude Code

[os-try-charles]

ACCEPT — PR objectui#9467, reviewed against GitHub and against the tree, ⛔ not against the report

domain:devx execution seat, session session_013VGeMu3p6qEFWR6K6GGLaW, round R58, 2026-09-14T06:52Z. ⚠️ Reviewing seat ≠ dispatching seat: this card was claimed and dispatched by session_01FhBNJcLRZLe8M87VcUgpKr, which handed the seat over at 06:16Z. I am the recorded reviewer and inherited the claim; the assignee and Claim: at 5659574246 stand unchanged.

PR head 3a3873755a, base e3cb47624 — equal to origin/main's tip at review time, so every reading below is a reading on this PR's actual base. Draft, 2 files, +411/−38. CI: 33 checks, all settled — 30 success / 3 skipped / 0 red. mergeable_state: clean.

Path face — taken with get_files, ⛔ not from the PR body

content/docs/guide/ci-cd-pipeline.md · scripts/__tests__/ci-cd-pipeline-doc.test.ts. Neither is governed surface (⛔ no docs/adr/**, .claude/**, skills/**, AGENTS.md, CLAUDE.md) ⇒ the ordinary landing path, ⛔ not the four-piece governed route. ⭐ Taken by reading the file list rather than trusting Governed Surface Queue Guard's green, because this lane's own standing fact is that the guard's pull_request leg exits 0 by design — its green is not evidence a PR is ungoverned.

Clause-②: no. Path limb: the diff touches no packages/spec/src/**, no error-code ledger, no *.zod.ts. Declaration limb: ⚠️ the inherited claim comment carries no Clause-② line at all — recorded as a carrier gap in the claim I inherited, ⛔ not papered over. Substantively the answer is no and the enqueue gate does not fire: nothing here widens an accept set. undocumentedCommands — the direction that hides a gate — admits no exception, verified in the diff.

What I re-measured myself — and one reading I made STRONGER than the dev's

⛔ The report was not taken on trust. Three load-bearing claims, each re-derived on origin/main with a firing control:

claim	my reading	control
doc-component-types.yml runs check-prompt-component-keys.mjs in its own step	✅ :114 run: node scripts/check-prompt-component-keys.mjs	nonsense spelling 0, real spelling 1
…and that step is blocking	✅ continue-on-error occurrences in the whole file: 0	run: occurrences 2 ⇒ the probe read the file; the zero is a reading
changeset-presence.yml grew a second job	✅ :96 changeset-presence and :142 changeset-claims → :168 run: node scripts/check-changeset-claims.mjs	both job keys named, ⛔ not counted
bijection at 38	✅ 39 heading lines / 38 workflow files	matches the report exactly

⭐ The dating argument — the one the whole "invert rather than queue" case rests on — I re-derived by a different method, because the dev's method is not available to me. Their evidence was git log -S with --is-shallow-repository reading false. ⚠️ In this container it reads true, and -S over a shallow history can only ever produce a false negative. So rather than repeat an instrument I cannot trust here, I read the tree itself at the census era:

shallow horizon reaches 2026-09-02  ⇒ the 2026-09-09 census IS inside my window
at aeaa0f64c (2026-09-10, i.e. AFTER the census — the conservative direction):
  check-prompt-component-keys  ->  0     CONTROL check-doc-component-types  ->  4
  changeset-claims (job)       ->  0     CONTROL changeset-presence (job)   ->  6

⇒ Both defects genuinely did not exist when the census ran. The census did not miss them. ⭐ This is the argument for inverting the default, and it is now verified on the tree with both controls firing, ⛔ not inherited from a -S run on a differently-shaped clone. ⚠️ Recorded because this lane's errata is explicit that a zero from a command not proven able to return non-zero is not a reading — both zeros here carry a non-zero control on the same file in the same command.

The two caveats the outgoing brief fenced this PR behind — both answered

1. The alias question is reported, ⛔ not decided — verified in the diff, not in the prose. SWEEP_DECLARED_NON_RUN_COMMANDS carries governed-surface-guard.yml: pnpm governed as a declared exception whose reason ends verbatim: "⛔ Whether such an alias should ever merge is a decision about the RULE — reported, not taken here." The settled rule is applied unchanged (a pure node scripts/<file> wrapper merges; an argument-carrying alias selects a mode and does not). ⇒ Landing this does not silently rule the alias question — declaring is the landed rule's own consequence, and merging would have been the new decision. The brief's fence is satisfied by the code, not by a promise.

2. The 6→38 widening is in scope, and openly flagged. The vi-mock/shadcn header reserved re-pointing the other sections as "a separate card". objectui#9463 is that card — its own title is the remaining pin appetite across the current denominator. ⇒ ⛔ not a silent side effect. It is declared in the PR body and in the test's own docblock.

The instrument cannot go green by reading nothing

Five ablation legs, each with on-disk hash proof and hash-equality restore. Leg E is the one that matters: the heading regex is mutated to .yZZ so the sweep reads zero sections → 3 failed, including both parser floors. ⇒ the other four legs cannot be passing vacuously. Two legs (A, B) were re-run standalone because the batch harness did not confirm where they went red — ⭐ an unconfirmed red was not accepted, which is this lane's own discipline applied by the dev to itself.

gates_weakened: NONE. DECLARED_NON_RUN_COMMANDS moves to module scope and the sweep re-keys it rather than growing a second copy — with a floor asserting the re-key actually arrives, so an empty re-key fails loudly instead of silently reporting four real phantoms as undeclared.

Deviations from the report — recorded, none blocking

• ⚠️ The dev's open_questions are not blockers: (1) is the alias question, correctly left declared; (2) and (3) are scope confirmations this review answers above. ⛔ None meets the escalation bar, so ⛔ none is being put to the maintainer as a decision card by this ACCEPT.
• ⭐ The card's own title figure ("仅 5 节已 pin") is a call-site count, not a section count — the true figure was 8. The dev corrected the card rather than matching it. ⛔ Not a defect; recorded because the card's remaining prose still carries the wrong denominator.
• The five out_of_scope_findings are REPORTED, NOT FILED per the dispatch. One of them — firstPartyCommands reading a shell comment inside a run: block scalar as a command — is now class-(a)-shaped under the inverted default, because it can flag a section on a comment alone across all 38 sections rather than 6. ⇒ this seat files it rather than letting it die in a PR body; see the follow-up below.

⇒ ACCEPT. Next: flip ready, re-wait for the post-flip Governed Surface Queue Guard re-run (a pre-flip green is stale in this repo), arm auto-merge, confirm entry from the added_to_merge_queue timeline event — ⛔ never from enable_pr_auto_merge's echo — and a landing probe written before the merge and re-run on the actual <merge>^.

⚠️ Part of #9463, ⛔ not Fixes — deliberate and correct. On merge this card does not auto-close: it returns to pm:queue with a Release: line naming what landed and what remains, so the alias stop-clause survives the merge instead of being swallowed by it.

---

Generated by Claude Code