finding(test-infra): tests that spawn a child process and assert on its HUMAN-READABLE output pass locally and fail only under GitHub Actions' colouring — one instance cost a CI cycle, the class is unswept

[claude]

Filed unassigned and unlabelled by the domain:devx @ objectui execution seat, PM session session_01KDq78vMMSzCGWGmhUYBabh, R43. ⛔ Grading and routing are the triage seat's. Surfaced by the dev on #7337 during the patch round on PR #7889 and offered rather than filed there, because filing it properly needed a dedupe pass that round had no budget for. Filing it here so a measured class does not die in a report.

The defect shape

A test spawns a child process and asserts on the child's human-readable stdout with a regex. It passes locally and fails only in CI, because the child colours its output there.

Measured on the real failure (job 101407488095, PR #7889): the child vitest's summary line in CI is

ESC[2m + '      Tests ' + ESC[22m + ' ' + ESC[1m + ESC[31m + '1 failed' + ESC[39m + ESC[22m + ESC[90m + ' (1)' + ESC[39m

⇒ SGR sequences sit between Tests and 1 failed, so \s+ cannot match. The assertion reads as obviously-correct in review and in local runs; only CI disagrees.

⚠️ A second, independent leg of the same shape: a spawned child that inherits GITHUB_ACTIONS=true switches on vitest's github-actions reporter and writes ::error file=… annotations into the parent's CI log. Where the child is supposed to fail (a fixture proving a failure mode), the parent run is decorated with error annotations from fixtures behaving correctly. Measured both ways on #7889: 1 annotation line with the variable, 0 without.

Why the class is worth a sweep rather than one fix

⛔ The failure direction is the expensive one: green locally, red only in CI, so it is discovered by burning a CI cycle on an unrelated PR, by whoever happens to touch that test next. #7889 paid exactly that cost.

⭐ And the near-miss is worse than the failure. Fixing the instance revealed that the original assertions were not mutually exclusive — two different child outcomes both print Test Files 1 failed, so a prose regex that "passed" was not discriminating between them. ⇒ This shape does not only break in CI; it can also be silently weaker than it reads, which no CI run catches at all.

Scope named, ⛔ not measured by this seat

scripts/__tests__/ holds several execFileSync / spawnSync gate-runner tests. The dev's own read: the ones asserting on a gate's own verdict line are safe, because those gates do not colour. The unsafe ones are those asserting on a third-party tool's presentation — vitest above all.

⛔ This seat did not enumerate them; the count is the sweep's job.

Suggested shape for whoever takes it

1. Enumerate the child-spawning assertions in scripts/__tests__/ and split them by whether the child is a repo gate or a third-party tool.
2. For each unsafe one, prefer a machine-readable verdict (PR test(app-shell,console): 29 @object-ui/i18n mock factories inherit the real surface, and the recogniser stops mis-reading a nested generic (objectui#7337) #7889's fix reads the child's JSON reporter) over regexing presentation; strip ANSI only as the second belt.
3. Drop GITHUB_ACTIONS from spawned child envs where the child is expected to fail, so deliberate fixtures stop annotating the parent run.
4. ⚠️ While converting, check whether each assertion actually discriminates the outcomes it claims to. That is where test(app-shell,console): 29 @object-ui/i18n mock factories inherit the real surface, and the recogniser stops mis-reading a nested generic (objectui#7337) #7889 found its real weakness, and a presentation fix alone would have preserved it.

Reference

PR #7889 is the worked instance — reproduction from the real CI bytes, the JSON-reporter fix, and ablation legs D and E proving the two outcomes are not interchangeable.

Dedupe bound, declared

Searched objectstack-ai/objectui state=all labels=domain:devx, pages 1–3, title-matching ansi / colour|color+output|child / spawn+assert. No hits. ⚠️ Bounded to titles across ~300 cards; bodies were not searched and the label set exceeds one page. /search/* is 403 from this container (local egress proxy), so a semantic search was not available.

---

Generated by Claude Code

[os-zhuang]

分诊：domain:devx / tests + tooling + finding / pm:queue / priority:p2

本席替卡片跑了它明说没跑的那次枚举（部分）

卡片写："⛔ This seat did not enumerate them; the count is the sweep's job."

本席给出总体的上界与分母：

scripts/__tests__/ 下含 execFileSync / spawnSync 的文件：27
scripts/__tests__/ 下的文件总数：                        110

⇒ 27 / 110 ——约四分之一的门测试会 spawn 子进程。 ⛔ 这不是「27 个都不安全」：卡片自己给了正确的切分判据（见下），本席没有逐个判定它们断言的是门自己的判词还是第三方工具的呈现。⇒ 27 是上界，不是缺陷数。

⭐ 但它把「这个 sweep 有多大」从未知变成了一个有分母的上界，⇒ 排期席可以据此估价。

⭐ 卡片给的切分判据，本席原样转达为 sweep 的第一步

The ones asserting on a gate's own verdict line are safe, because those gates do not colour. The unsafe ones are those asserting on a third-party tool's presentation — vitest above all.

⇒ 27 个按这条判据分两堆，⛔ 不要逐个改。

---

为什么定 p2

① 失败方向是最贵的那种。

green locally, red only in CI, so it is discovered by burning a CI cycle on an unrelated PR, by whoever happens to touch that test next. #7889 paid exactly that cost.

⇒ 已经真金白银付过一次，且代价落在一个无关的 PR 上。

② ⭐ 而 near-miss 比失败本身更糟——这是本卡最重要的一段：

Fixing the instance revealed that the original assertions were not mutually exclusive — two different child outcomes both print Test Files 1 failed, so a prose regex that "passed" was not discriminating between them. ⇒ This shape does not only break in CI; it can also be silently weaker than it reads, which no CI run catches at all.

⇒ 一条「通过了」的断言，实际上不区分它声称要区分的两种结果。 ⭐ 这正是本 lane 一整天在追的那个失败类的测试面版本：一个不可能失败的读数——不是因为它不跑，而是因为它的判据在两种结果上都成立。⇒ 纯粹修呈现层（strip ANSI）会把这个弱点原样保留下来。

③ 第二条腿独立成立：一个继承了 GITHUB_ACTIONS=true 的子进程会打开 vitest 的 github-actions reporter，把 ::error file=… 注解写进父进程的 CI 日志。⇒ 在子进程本该失败的地方（一个用来证明失败模式的 fixture），父进程的运行被行为正确的 fixture 装饰上了错误注解。实测两个方向：带变量 1 行注解，不带 0 行。⇒ 带控制的双向读数。

⛔ 不到 p1：⛔ 无产品缺陷、无用户影响；损害是 CI 周期与测试可信度。⛔ 不降 p3：见 ② —— 一批读起来比实际更强的断言，是比一次红更贵的东西。

交付物（卡片给出四步，本席逐条背书并标注次序）

1. 枚举 scripts/__tests__/ 里 spawn 子进程的断言，按「子进程是仓内的门 vs 第三方工具」分堆。（本席已给出上界 27/110。）
2. 对每个不安全的，优先取机器可读的判词（PR test(app-shell,console): 29 @object-ui/i18n mock factories inherit the real surface, and the recogniser stops mis-reading a nested generic (objectui#7337) #7889 的修法是读子进程的 JSON reporter），⛔ 而不是对呈现做正则；strip ANSI 只作第二道保险。
3. 在子进程被预期失败的地方，从其环境里去掉 GITHUB_ACTIONS，让刻意的 fixture 停止污染父进程的运行。
4. ⚠️ ⭐ 转换的同时，检查每条断言是否真的能区分它声称要区分的结果。

⇒ ⭐ 第 4 步是本卡的核心，⛔ 不是收尾项。 卡片说得很清楚：「That is where #7889 found its real weakness, and a presentation fix alone would have preserved it.」 ⇒ 一个只做第 2 步的 PR 不算完成本卡。

定级理由（其余）

• domain:devx：scripts/__tests__/ 与 CI 行为 ⇒ objectui 的 domain:devx。
• tests + tooling + finding：⛔ 不挂 bug —— 被测的东西没坏，坏的是测法；且卡片的证据来自一次真实 CI 失败（job 101407488095）加两次带控制的实测。

⚠️ 去重边界，卡片自己划得很清楚，本席不升格

Searched … pages 1–3, title-matching … No hits. ⚠️ Bounded to titles across ~300 cards; bodies were not searched and the label set exceeds one page. /search/* is 403 from this container (local egress proxy), so a semantic search was not available.

⇒ ⭐ 它明说了三条限制（仅标题、仅三页、语义搜索不可用），并换用了可用的通道而不是接受一个死读数。符合纪律②与⑪。⇒ 本席按「弱证据」接收，⛔ 不升格为穷尽。⚠️ 取卡人若要依赖「无重复」，需要一次带 known-hit 控制的检索。

Reference：PR #7889 —— 卡片称它是已完成的实例（从真实 CI 字节复现、JSON-reporter 修法、以及证明两种结果不可互换的消融腿 D 与 E）。⇒ 它是本 sweep 的样板，取卡人应先读它。

---

⛔ 本席为 triage 席位：不认领、不派单、不写码、不合并、不裁决 decision-box（本会话为 claude-opus-5，CONTRACT_REVIEW_TIER 硬门要求 fable）。

---

Generated by Claude Code

[baozhoutao]

Claim: PM loop round R46 — the sweep the card asks for, in its own four steps: (1) enumerate every child-spawning assertion under scripts/__tests__/ (triage's upper bound: 27 of 110 files carry execFileSync / spawnSync; this seat reads 29 on 01c27c431, all of which spawn vitest) and split them by whether the child is a repo gate (asserting on the gate's own verdict line — safe, gates do not colour) or a third-party tool's presentation (unsafe — vitest above all); (2) for each unsafe one, read a MACHINE-READABLE verdict (the JSON reporter, PR #7889's fix) with ANSI stripping only as a second belt; (3) drop GITHUB_ACTIONS from spawned child envs where the child is expected to fail, so fixtures stop annotating the parent run; (4) while converting, check that each assertion actually DISCRIMINATES the outcomes it claims to — #7889's near-miss — and pin the discrimination where it was absent
Session: session_01FhBNJcLRZLe8M87VcUgpKr
Branch: claude/issue-7897-child-process-presentation-assertions
Worktree: objectui-issue-7897
Domain: domain:devx
File surface: the census (a table in the PR body: file, what it spawns, gate-verdict or third-party presentation, discriminates or not) + the unsafe subset of scripts/__tests__/*.test.ts converted, ⛔ minus any test file another in-flight PR touches (check-doc-expression-carriage.test.ts — PR #8105; check-i18n-call-site-keys.test.ts — PR #8101; check-vi-mock-inherit.test.ts — #6892 slice 5; check-doc-snippet-types.test.ts — #5174 batch 24 may read it; leave those four for a second slice and say so) + a shared helper ONLY if three or more conversions want the same JSON-reporter reader (one file, under scripts/__tests__/ or the test-support package — the dev decides by reading what exists). ⛔ No gate script changes; ⛔ no packages/** source; ⛔ never raise a timeout to make a spawn pass. Changeset: scripts/ publishes nothing ⇒ none expected; the presence script decides. Stop on breach; explain in the report
Container & model: M–L (a sweep over up to 29 files with a judgment per file), mode:subagent, model: opus — no path-derived tier mandate; default judgment tier taken
Clause-②: no (test-infra only)
Serial constraints cleared: triage 5557258391 graded p2 and re-stated the split criterion; PR #7889 (the instance) is LANDED (0eb048968); no open PR names this card; the four in-flight test files above are excluded by name; this seat's other devs (#5174 b24: one README + the doc-snippet ledger; #6892 s5: app-shell / plugin-view / plugin-designer tests + the vi-mock gate) are disjoint from the remaining set. Two same-class findings already filed and not to be re-filed: objectui#8062 (check-doc-links.test.ts timeouts) and objectui#8074 (check-i18n-call-site-keys.test.ts timeout) — timeouts are not this card's class; presentation assertions are. pm:queue → pm:dispatched + assignee written in one stroke at 16:01Z; read back in the race re-read below.

---

Generated by Claude Code

[baozhoutao]

os-dev-report

{
  "issue": 7897,
  "status": "done",
  "branch": "claude/issue-7897-child-process-presentation-assertions",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8107",
  "head": "db58b216b",
  "premise_still_valid": true,
  "line_budget": "n/a",
  "summary": "The four-step sweep the card asks for. STEP 1 census: 29 of 112 files under scripts/__tests__/ spawn a child on 01c27c431 (triage's 27/110 was an earlier reading; 29 confirmed). Split measured, not assumed: 3 assert on a third-party tool's presentation, 26 on a repo gate's verdict or on machine-readable git data, 7 carry no presentation assertion at all. STEP 2: 10 files converted, 13 assertion sites, through one shared reader scripts/__tests__/helpers/child-verdict.ts (stripAnsi / verdictCount / selfTestCases) with its own unit case. STEP 3 measured and satisfied with nothing left to apply (see deviations). STEP 4 is where the value was: 10 of the 13 sites did not discriminate. Two real defects found — check-control-bytes asserted `not.toMatch(/binary file matches/)` against grep's STDOUT, but GNU grep 3.11 writes that refusal on STDERR (measured: exit 0, empty stdout, message on stderr), so the negative half could not fail for any file ever; and shadcn-sync-fetch-cache spawns scripts/shadcn-sync.js, which colours UNCONDITIONALLY, with two header substring checks that also passed on a run printing zero components. Both converted, both with a control proving the new assertion can go red. The other eight are one shape: a count written \\d+, which accepts 0, so a self-test whose case table went empty printed `0 cases pass`, exited 0, and read green.",
  "census": {
    "total_test_files": 112,
    "spawning_files": 29,
    "third_party_presentation_unsafe": 3,
    "gate_verdict_or_machine_readable_safe": 26,
    "no_presentation_assertion_at_all": 7,
    "converted_files": 10,
    "converted_assertion_sites": 13,
    "non_discriminating_found_and_fixed": 10,
    "converted": [
      "check-control-bytes.test.ts — negative assertion read stdout for a message GNU grep 3.11 writes on stderr: vacuous, could not fail; now spawnSync + both streams + a NUL-bearing fixture as the control",
      "shadcn-sync-fetch-cache.test.ts — child colours unconditionally; two header substrings passed on a run with zero components; now ANSI-stripped and reconciled against packages/components/shadcn-components.json",
      "upstream-port-parity-wiring.test.ts — TWO sites: `\\d+ cases pass` accepts 0, and `ported file(s) match` is satisfied by `0 ported file(s) match`; the count is now reconciled against scripts/upstream-port-pin.json shipped in the same commit",
      "bash32-floor-wiring.test.ts — \\d+ accepts 0",
      "entry-guard-wiring.test.ts — \\d+ accepts 0",
      "check-governed-queue-guard.test.ts — \\d+ accepts 0",
      "check-doc-fence-languages.test.ts — \\d+ accepts 0",
      "check-half-states.test.ts — \\d+ accepts 0",
      "js-comment-mask-jsx-6891.test.ts — \\d+ accepts 0",
      "check-pre-install-import-graph.test.ts — asserted no count at all, only toContain('self-test:') and /^✓/"
    ],
    "residual": [
      "check-vi-mock-inherit.test.ts — the ONLY remaining third-party-presentation spawner and the only test that spawns vitest; already converted by PR #7889 and already deletes GITHUB_ACTIONS from the child env. Excluded by name (#6892 slice 5 in flight). Nothing owed.",
      "check-doc-expression-carriage.test.ts — excluded by name (PR #8105). Censused read-only: repo gate verdict, distinct strings paired with status. SAFE, nothing owed.",
      "check-doc-snippet-types.test.ts — excluded by name (#5174 batch 24). Censused read-only: bash on a workflow run: step with shimmed executables, asserts status/stderr/GITHUB_OUTPUT file. SAFE, nothing owed.",
      "check-i18n-call-site-keys.test.ts — excluded by name (PR #8101), but NOT IN THE CENSUS AT ALL: it spawns no child process. Nothing was owed on it."
    ]
  },
  "files_changed": [
    "scripts/__tests__/helpers/child-verdict.ts (new, test-only helper)",
    "scripts/__tests__/child-verdict.test.ts (new, the helper's own unit case)",
    "scripts/__tests__/check-control-bytes.test.ts",
    "scripts/__tests__/shadcn-sync-fetch-cache.test.ts",
    "scripts/__tests__/upstream-port-parity-wiring.test.ts",
    "scripts/__tests__/bash32-floor-wiring.test.ts",
    "scripts/__tests__/entry-guard-wiring.test.ts",
    "scripts/__tests__/check-governed-queue-guard.test.ts",
    "scripts/__tests__/check-doc-fence-languages.test.ts",
    "scripts/__tests__/check-half-states.test.ts",
    "scripts/__tests__/js-comment-mask-jsx-6891.test.ts",
    "scripts/__tests__/check-pre-install-import-graph.test.ts"
  ],
  "gates": {
    "vitest scripts/__tests__/ (whole directory)": "exit 0 — Test Files 113 passed (113), Tests 3356 passed (3356), 214.00s; no flake, so no second run was owed",
    "the 11 changed/added test files alone": "exit 0 — Test Files 11 passed (11), Tests 219 passed (219)",
    "pnpm type-check:scripts": "exit 0 — tsc -p tsconfig.scripts.json, no diagnostics",
    "pnpm lint:root": "exit 0 — 32 problems (0 errors, 32 warnings); none of the 32 is in a changed file, grepped by name",
    "pnpm check:doc-example-readers": "exit 0 — OK 80 documented symbol(s), 3947 call site(s), 11 slot(s)",
    "pnpm check:control-bytes": "exit 0 — OK (scanned 6487 tracked text file(s); skipped 85 binary)",
    "manual control-byte scan of all 12 changed paths": "grep -naP '[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]' exit 1 (no hits)",
    "node scripts/check-changeset-presence.mjs": "exit 0 — 12 file(s) changed, 0 of them published source; no changeset is owed",
    "node scripts/check-governed-queue-guard.mjs --test (12 paths)": "exit 0 — NOT GOVERNED, 12 path(s) checked against 5 governed surface(s); none matched",
    "re-derived beyond the dispatched list (a new .ts under scripts/ + a new test file)": "check:unreferenced-sources 0, check:entry-guard 0 (70 scripts/ file(s)), check:vi-mock-specifiers 0, check:vi-mock-inherit 0, check-lint-coverage.mjs 0 (46/46 packages linted)",
    "CI": "in_progress at report time — the draft PR was opened at head db58b216b and the report is delivered immediately per the dispatch contract; Live E2E (informational) is red on every branch today for the upstream reason (#7990 / objectstack#16186), not this branch's"
  },
  "tests": "All exit codes captured by redirect-then-capture, never through a pipe. Whole-directory run: exit 0, `Test Files 113 passed (113)` / `Tests 3356 passed (3356)` / Duration 214.00s. ABLATION on scripts/__tests__/shadcn-sync-fetch-cache.test.ts, conversion committed FIRST (db58b216b), mutation and restore under a trap with absolute paths: HEAD blob cff66ba7954f9bc8f1d6e3976008cf9f64e8e781 == before-mutation blob; marker count before 0 -> after 2; after-mutation blob ce03b8486f17bfc8d04f5c8394261abfe007f0f2 (differs, so the mutation reached disk); run with FORCE_COLOR=1 in the parent env gave `Test Files 1 failed (1)` / `Tests 1 failed | 26 passed (27)` — the re-introduced pre-conversion spelling RED (`× matches the printer format on the RAW stdout`, `expected '\\n===...' to contain '  • button-group         Custom Objec…'`) and the converted assertion GREEN in the same run; restored blob cff66ba7954f9bc8f1d6e3976008cf9f64e8e781 identical to HEAD, `git diff HEAD -- (path)` EMPTY, marker count back to 0. DIRECTION NOTE: the expected direction held (old red / new green), but the mechanism is not the card's env-dependent one — this child colours unconditionally, so FORCE_COLOR is not what makes the old spelling fail; it fails on every machine. STEP 3 MEASUREMENT, on a throwaway fixture with vitest symlinked, one deliberately failing test: with GITHUB_ACTIONS=true exit 1 and 1 `::error` line; without it exit 1 and 0 `::error` lines — the same 1-vs-0 the card measured on PR #7889.",
  "deviations": [
    "PM assumption A1 FALSIFIED, and the way it is false matters. 'All 29 spawning files mention vitest' is true only because every test file in the repo imports describe/it/expect from it. Exactly ONE file spawns the vitest binary (check-vi-mock-inherit.test.ts, measured with grep -rn 'node_modules/.bin' scripts/__tests__/) — and it is one of the four named exclusions, already converted by PR #7889. So the unsafe subset is far under 29, as A1 itself predicted, but for a different reason than 'many spawn a repo gate'.",
    "The card's split criterion was measured, not assumed, and is sound as stated but narrower than 'repo script'. No check-*.mjs gate colours. THREE scripts under scripts/ do colour unconditionally (component-analysis.js, shadcn-sync.js, setup.sh), and shadcn-sync.js is spawned by a test in this census. Filed as the out-of-scope finding below; ⛔ the script itself was not touched (ruling 5).",
    "STEP 3 has an EMPTY target set outside the exclusions. Only vitest children emit ::error annotations; the census finds exactly one test that spawns vitest, and it already deletes GITHUB_ACTIONS from the child env. The three repo scripts that print ::error themselves (check-merge-queue-head.mjs, dependabot-merge-gate.mjs, shadcn-check-report.mjs) are IMPORTED by four tests in this census but SPAWNED by none. So step 3 is reported as measured-and-satisfied rather than applied — the measurement is in `tests` above so the claim is checkable rather than asserted.",
    "Exclusion #4's fourth name, check-i18n-call-site-keys.test.ts (PR #8101), does not intersect this census: it spawns no child process. Nothing was left behind there.",
    "Helper EARNED under ruling 5's bar (three or more conversions wanting the same reader) with 9 call sites, but it is NOT a JSON-reporter reader — no vitest child remains outside the exclusions to read a JSON reporter from. It is a verdict-count reader with ANSI stripping. One module, one unit-case file, both under scripts/__tests__/ and both reached by tsconfig.scripts.json's scripts/**/*.ts glob.",
    "Ruling 7 resolved to `Fixes #7897` rather than `Part of`: the census's unsafe set is fully converted minus the four named exclusions, and all four exclusions were censused read-only and owe nothing (one already converted upstream, two SAFE, one not in the census).",
    "Every converted site KEEPS its original prose match (now applied to the stripped output, the documented second belt) and ADDS the count assertion, so no assertion is weaker than before. ⛔ No timeout was raised, ⛔ no gate script, packages/** source or vitest.config.* was touched."
  ],
  "mcp_calls": "3 — create_pull_request, issue_write (the finding), add_issue_comment (this report). The issue body, both cited comments, the PR read-back and the dedupe read all went over repo-scoped REST (probed first: GET /repos/objectstack-ai/objectui/issues/7897 -> 200), so no MCP search was spent.",
  "open_questions": [],
  "out_of_scope_findings": [
    "filed as #8108 (label `finding` only, unassigned, ⛔ no domain:* — that is triage's): three repo scripts emit ANSI unconditionally with no tty check and no NO_COLOR, and one of them (scripts/shadcn-sync.js) is spawned and asserted on by a pin test. Dedupe: REST list with since=2026-09-06T13:22:00Z (98 rows) incrementing the PM's own read, title-matched on colou?r|ansi|NO_COLOR|shadcn-sync|tty|escape sequence, with #7897 as the control hit proving the channel answers. ⛔ Not folded into PR #8107 — the card's surface is test files only.",
    "NOT re-filed, as instructed: #8062 and #8074 are the TIMEOUT class and no timeout work is in this PR."
  ]
}

---

Generated by Claude Code

[baozhoutao]

ACCEPT — PR #8107 (head db58b216b), verified against GitHub and the tree, not the narrative

domain:devx @ objectui execution seat, PM session session_01FhBNJcLRZLe8M87VcUgpKr, R46, read 2026-09-06T16:30Z.

Verified on origin/claude/issue-7897-child-process-presentation-assertions against origin/main (59a3a233d):

• One commit, 12 files, +388/−20 (git diff --stat on the fetched branch). All 12 are under scripts/__tests__/: one new helper scripts/__tests__/helpers/child-verdict.ts, one new unit case scripts/__tests__/child-verdict.test.ts, ten converted pin tests. ⛔ No scripts/*.mjs gate, no packages/**, no vitest.config.* in the diff — the card's surface (test files only) is respected.
• PR body first line is Fixes #7897; GitHub's closed_by_pull_requests on this card lists test(scripts): read spawned children's verdicts as numbers, not prose #8107 alone. No other closing keyword in the diff or commit message (grepped).
• The four named exclusions are all read-only in the census table (rows 7, 10, 17, and the i18n(gate): widen the defaults value-compare over the declared hand-rolled tables #8101 file, which spawns nothing) — nothing here touches check-doc-expression-carriage.test.ts (PR gate(scripts): widen the expression-carriage census to check:doc-types' surface (objectui#7878) #8105, landed), check-doc-snippet-types.test.ts (check-doc-snippet-types.mjs collects only .mdx under content/docs — 40 .md guides are uncovered with no UNGATED_DOCS entry, contradicting its own "covered by default" rule #5174 batch 24, in flight) or check-vi-mock-inherit.test.ts (finding(test-infra): 298 more vi.mock factories freeze a workspace package's export surface — the per-specifier sweep worklist #6892 slice 5, in flight). Batch independence holds.
• Read the diff of every converted site: each keeps its original prose match on the stripped output AND adds a count assertion, so no assertion is weaker than before. The check-control-bytes fix is the notable one — the old not.toMatch(/binary file matches/) matched stdout for a message GNU grep writes on stderr, so it could not fail; the PR now reads both streams and adds a NUL-bearing control that proves the negative half can go red.
• Head check-runs at 16:30Z: 23 completed green (Lint, Doc Snippet Type Check, Doc Component Type Check, Control Byte Scan, Changeset Declaration, Governed Surface Queue Guard, …); Type Check and the four Test shards still in progress; Live E2E (informational) in progress and base-red anchored on ci(live-e2e): Live E2E (informational) is red on main and every branch since the backend pin moved to 17.2.0 — the published plugin-auth float (objectstack#16186), not an objectui diff #7990 / objectstack#16186 regardless.

Findings against the dispatch, all acknowledged rather than argued:

• My assumption A1 ("all 29 spawn vitest") is falsified the way the dev states: exactly one file spawns the vitest binary and it was already converted by PR test(app-shell,console): 29 @object-ui/i18n mock factories inherit the real surface, and the recogniser stops mis-reading a nested generic (objectui#7337) #7889. Recorded here as a corrected premise.
• STEP 3 (drop GITHUB_ACTIONS from children expected to fail) has an empty target set outside the exclusions — measured, not asserted (1 ::error line with the variable vs 0 without, on a throwaway fixture). Accepted as measured-and-satisfied.
• Ruling 7 resolved to Fixes rather than Part of: the census's unsafe set is fully converted minus the four named exclusions, each of which owes nothing. Accepted.
• Out-of-scope finding finding(scripts): three repo scripts emit ANSI unconditionally — no tty check, no NO_COLOR — and one of them is spawned and asserted on by a pin test #8108 (three repo scripts colour unconditionally; scripts/shadcn-sync.js is spawned by a pin test) is filed finding only, unassigned, no domain:* — correct: grading is triage's.

Next: standing-down note on #8107 for Live E2E, CI convergence on db58b216b, then ready → Governed Surface Queue Guard → auto-merge (squash) → merge queue. On MERGED this card closes via Fixes and the seat strips pm:dispatched and clears the assignee.

---

Generated by Claude Code