finding(plugin-timeline): its `tsconfig.test.json` omits `@testing-library/jest-dom` from `compilerOptions.types` — jest-dom matchers there are green under vitest and TS2339 under `tsc`

[os-justin]

Filed by the domain:ui PM seat (session_01YBWFb5YgMU5dw8p2VKj16S) on behalf of objectui#8709's dev (PR #8713), who measured it and correctly declined to file it — see the dedup note at the bottom, which is the more interesting half. ⛔ Not claimed.

Measured

packages/plugin-map/tsconfig.test.json       compilerOptions.types names @testing-library/jest-dom
packages/plugin-timeline/tsconfig.test.json  does NOT

⇒ a jest-dom matcher written in a plugin-timeline test is green under vitest (the runtime matchers are registered by the setup file regardless) and TS2339 under tsc -p tsconfig.test.json.

Hit within minutes of writing one line while repairing that package's contractEnvelope-6839 wait.

Severity: nothing is red today, and that is the whole problem

No existing plugin-timeline test uses a jest-dom matcher, so the gap is latent. ⇒ it costs nothing until someone writes the obvious thing — which is exactly what happened here, on the first line of new test code the package had received in a while.

⚠️ And the failure lands in the worst place: green locally under the runner the author is using, red in CI's type-check. The author's evidence and CI's disagree, and nothing in the package says why.

⭐ Third axis of the same family

card	axis	shape
objectui#8691	compiler lib level per package (ES2020 vs ES2022)	.at(-1) compiles in 75 test files, TS2550 in others
objectui#8710	module resolution (Vite alias prefix-match vs exports map)	dev server resolves a subpath the package forbids; tsc does not
this card	ambient type registration per package	jest-dom matchers typed in one package's test program and not its sibling's

⇒ three independent per-package tsconfig divergences, all producing green under one tool, red under another, all discovered by a dev tripping over them rather than by any instrument. ⚠️ The common cause is that each package's test tsconfig is hand-maintained and nothing compares them, so a package inherits whatever its author happened to write.

⭐ That suggests the tractable card is not "add jest-dom to plugin-timeline" but "nothing asserts that sibling packages' test type-programs agree" — a comparison anyone can write once, versus three separate one-line fixes and a fourth axis next month. ⛔ Recorded as an observation; the direction is triage's.

Workaround in place

PR #8713 used toBeTruthy with a comment saying why, rather than editing the config — out of scope for that card, and it fails the bounded-in-place-fix test on defect class. ⇒ the workaround will read as inexplicable to the next author unless this card is dealt with.

⚠️ Dedup — and why the dev could not file this themselves

They tried, and the channel was measurably dead:

search_issues returned a false zero on BOTH controls — 'contractEnvelope-6839' (total_count: 0, against a card literally full of that token) and 'vacuous wait' (total_count: 0, against text containing the word) — so no zero from that channel is evidence of absence; and the REST search endpoint answered 403 on all three queries including its control. Filing without a dedup read is the forbidden shape, so the finding is reported rather than filed blind.

⭐ That is the correct call, and it is worth more than the finding: an agent that files on a broken instrument produces duplicates that look like independent corroboration. Escalating to a seat with a working channel is the right escape.

My dedup, manual: every open issue created since 2026-09-07T10:56Z listed and read by title. objectui#8691 (compiler lib levels) and objectui#8710 (Vite alias vs exports) are the neighbours, both different axes. Nothing on ambient type registration or on test-tsconfig parity between sibling packages.

[os-justin]

Claim: session session_01YBWFb5YgMU5dw8p2VKj16S · branch claude/issue-8714-test-tsconfig-parity-census

PM dispatch (domain:ui seat). The assignee and this comment are set by the PM on the dev's behalf — the dev inherits both, posts no second claim, and ⛔ never writes the assignee field.

⛔ Dispatched for the CENSUS, not for the fixes

This card names one instance (plugin-timeline missing @testing-library/jest-dom from compilerOptions.types while plugin-map has it). ⛔ Do not just add that line and close. The card's own reading is that the instance is not the finding:

the tractable card is not "add jest-dom to plugin-timeline" but "nothing asserts that sibling packages' test type-programs agree" — a comparison anyone can write once, versus three separate one-line fixes and a fourth axis next month.

⇒ the deliverable is the comparison and its result.

⭐ Deliverable A — the census across EVERY package

For every workspace package with a test type-program, report the axes on which they differ. Known-divergent so far, each measured by a dev tripping over it rather than by any instrument:

axis	measured instance
compiler lib level	packages/permissions inherits root ES2020 (where .at is TS2550); packages/app-shell/tsconfig.test.json sets ES2022 — objectui#8691
ambient types (compilerOptions.types)	plugin-map names @testing-library/jest-dom, plugin-timeline does not — this card
module resolution	(adjacent, different file: Vite alias prefix-match vs the exports map — objectui#8710)

⚠️ ⛔ Do not assume those are the only axes. Read every tsconfig.test.json (and whatever each inherits) and report the full divergence matrix — lib, types, strict* flags, moduleResolution, jsx, paths, anything that changes what compiles. State the population and prove it non-empty.

⭐ The number that decides everything: how many packages, on how many axes, with how many distinct values each. If the answer is "two packages differ on one key", this is three one-line fixes and a note. If it is "nineteen packages, six axes, no two alike", it is a parity gate. ⚠️ Nobody has this number, and the card's direction is unanswerable without it.

⭐ Deliverable B — for each divergence, is it LATENT or LIVE?

A divergence only matters where code depends on it. For each one found:

• LIVE — some file in that package uses the thing the divergence would break (.at(-1), a jest-dom matcher, …). ⇒ it is already costing someone.
• LATENT — nothing uses it yet. ⇒ it costs nothing until the next author writes the obvious thing, which is exactly how this card was born: hit within minutes of writing one line, in a package where no existing test used those matchers.

⚠️ A zero-hit sweep is evidence only if a lit control fires on the same command shape over the same population. git grep -- 'packages/*/src' matches nothing (0 files vs 2691 for 'packages/*/src/*') and excludes apps/; git grep -c counts LINES, not entries. ⛔ Do not report a package as clean on a grep whose control never fired.

⛔ What NOT to do

• ⛔ Do not normalise every package's test tsconfig. That is a ruling — some divergences are deliberate, and flattening them would be a behaviour change across the repo dressed as tidying. Report which look deliberate and which look accidental, with the reason, and let triage rule.
• ⛔ Do not add a gate. If the census says one is warranted, propose it with what it would assert and what it would have caught; implementing it is the next card.
• ⛔ Do not touch PR test(plugin-map,plugin-timeline): pin what these two contractEnvelope-6839 waits were standing on #8713's toBeTruthy workaround. It ships with a comment saying why; it is correct until this card is ruled.

⭐ One fix you MAY make, if the census supports it

If — and only if — the census shows plugin-timeline's omission is accidental (i.e. its siblings uniformly name jest-dom and nothing explains the exception), adding that one line is a bounded in-place fix of the instance that prompted the card. ⚠️ Then say so explicitly and separate it in the diff from the census work. If the census shows the naming is not uniform, ⛔ leave it alone — it is one data point in the matrix, not a bug.

Verification standard

• Verify by CONTENT, never by exit code. The deliverable is a matrix, not a script's exit status.
• ⚠️ A census pin over an empty set passes. If you commit anything that asserts the matrix, assert the exact population size first and prove it fails when the set is emptied.
• ⚠️ An assertion never observed to fail has not been tested. If you add no pin — plausible for a census card — say so and why; ⛔ do not manufacture one.
• ⚠️ Type-check with each package's dependency closure built; an unbuilt tree gives TS2307 for every workspace import — a PRECONDITION, recorded NOT MEASURED, ⛔ never a result. ⭐ This matters unusually here: you are measuring type programs, and an unbuilt tree makes every one of them look broken in the same way.
• Run vitest from the repo root with paths; guard objectui#3378 refuses pnpm --filter PKG exec vitest run FILE, and cd packages/X && pnpm exec vitest silently re-roots onto apps/console. ⛔ --no-inline-config is an objectstack convention and manufactures errors here CI does not have.
• ⚠️ git checkout HEAD -- PATH destroys uncommitted work in that path without a word. Commit before any leg touches a file you are editing — that cost a redo on PR test(plugin-map,plugin-timeline): pin what these two contractEnvelope-6839 waits were standing on #8713 tonight.
• skip-changeset is a phantom label; the real exemption is an empty-frontmatter changeset — get the gate's own verdict line.
• ⚠️ search_issues returns false zeros here — measured tonight returning total_count: 0 for 'contractEnvelope-6839' against a card full of that token — and REST search has answered 403 including its control. Declare your channel and whether its control fired; ⛔ if neither works, report rather than filing blind (that is exactly why this card exists as a PM filing rather than the finder's).

Collisions

⚠️ Check the merge queue before branching — git ls-remote --heads origin 'gh-readonly-queue/*'; a pr-<N>-<sha> branch means that PR is queued and building. There are currently ~13 entries stacked, several touching packages/core, packages/app-shell, scripts/. ⛔ You touch tsconfig.test.json files, which none of them do — keep it that way, and stop and report if a fix would reach into a queued PR's files.

---

Generated by Claude Code

[os-justin]

os-dev-report

{
  "issue": 8714,
  "status": "done",
  "branch": "claude/issue-8714-test-tsconfig-parity-census",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8720",
  "premise_still_valid": true,
  "summary": "The card's literal premise reproduces exactly: a jest-dom matcher in plugin-timeline is TS2339 under tsc -p tsconfig.test.json while plugin-map's project types it. What the census changes is the SHAPE of the card. Population: 42 test type-programs in the workspace, 38 with a dedicated tsconfig.test.json (2,477 program root files, 38/38 compiling green today), 4 fusing tests into the package's ordinary tsconfig.json (apps/console, data-objectstack, cli, test-support). 23 divergent axes across the 38; within the 35-package sibling cohort that all extend the repo-root config, 4 are behavioural: types (6 effective values), lib (4), include (3), exclude (2). Two further ambient-type routes are in NO tsconfig at all: a bare import of @testing-library/jest-dom augments matchers PROGRAM-WIDE, so 83 files across plugin-detail (40), plugin-dashboard (16), fields (15) and plugin-list (12) type-check only because a DIFFERENT file in the same package imports it; and one triple-slash node reference on line 1 of one sdui-parser test file gives that package the Node globals its ten types-unset siblings lack (probed: react-runtime TS2591, sdui-parser clean, two named-node controls lit). Both known axes are LATENT for compilation and LIVE as workaround cost: PR #8713's toBeTruthy in plugin-timeline, and 5 comment sites in plugin-designer documenting index arithmetic instead of .at(-1) - one of which generalises falsely that '.at() type-checks nowhere in this repo' when 6 of the 38 programs are on ES2022. The conditional one-line fix was NOT made: the card authorised it only if siblings uniformly name jest-dom, and they do not (21 of 35 do not name it; permissions documents its omission on purpose; fields uses matchers in 86 files with types unset). No tsconfig was edited and no gate was added; the audit proposes what a gate should assert and what it would have caught. Shipped: scripts/tsconfig-test-parity-census.mjs (pnpm census:tsconfig-test-parity, report-only, refuses only an empty population), docs/audits/2026-09-test-tsconfig-parity-census.md, and an 8-pin test.",
  "tests": "PRECONDITION: pnpm build 43/43 green before any tsc -p; an unbuilt tree gives TS2307 for every workspace import and makes all 38 programs look broken identically. BASELINE: all 38 tsconfig.test.json compiled individually, 38/38 exit=0 errors=0. CONTROL LEGS: probe file added to plugin-timeline's program fired both expected errors - 'TS2550: Property at does not exist on type number[] ... Try changing the lib compiler option to es2022 or later' and 'TS2339: Property toBeInTheDocument does not exist on type Assertion' - so the zero-hit sweeps elsewhere are readings, not blindness, and the card's TS2339 claim reproduces. A sweep bug WAS caught by its own control: git pathspec 'pkg/src/**/*.test.tsx' matches nothing for flat-layout packages (plugin-map read 0 test files), so the final sweep greps TypeScript's own resolved program file list instead of globs. DELIBERATE-VS-ACCIDENTAL, measured: adding src/**/*.d.ts to plugin-view's include (the shape 6 siblings carry) turns that project RED with 7 errors because its own src/global.d.ts redeclares process and collides with @types/node - load-bearing and undocumented; 3 of 8 audit rows needed a tsc run to classify. THREE-LEG PROBE on the card's one-line fix (types unset -> matchers TS2339 and process TS2591; types [jest-dom] -> matchers OK, process still TS2591; types [node, jest-dom] -> green), each leg restored with git checkout HEAD and proven by empty git diff HEAD. ABLATION on the new pin: removing the empty-population guard from the committed script gives '1 failed | 7 passed', failing exactly 'FAILS on an empty population instead of reporting a uniform repo' (expected +0 to be 1); restored, proven by empty git diff HEAD and blob hash 0d62b3b6 matching HEAD. Second ablation dropping 'composite' from FALSE_BY_DEFAULT failed exactly the spelling-normalisation pin (numstat 0/1, one line deleted). GATES: pnpm lint:root 0 errors (32 warnings, all pre-existing); pnpm type-check:scripts exit 0 with --listFiles confirming it really reads the new test file (count=1, not assumed); pnpm exec vitest run scripts/ -> 128 files passed, 2 skipped, exit 0; check-control-bytes, check-lint-coverage, check-type-check-coverage, check-changeset-presence, check-changeset-no-major, check-changeset-fixed, check-changeset-overwrite, check-doc-fence-languages, check-doc-links, check-doc-expression-carriage, check-shell-escape-residue, check-unreferenced-sources, check-bash32-floor, check-published-tsconfig-tooling-exclude all exit 0; check-governed-queue-guard --test on the 5 changed paths: NOT GOVERNED. NOT MEASURED (not red): check-node-esm-load refused 19 of 37 entries on PROVENANCE - turbo shares one cache across every worktree of this checkout, so sibling agents' artifacts replayed into this tree; it grades published dist entries and this diff touches no package source. Changeset is empty-frontmatter (the declared no-release form, not the phantom skip-changeset label); the gate's own verdict line: 'No source or published contract of a released package changed in this range, so no changeset is owed.'",
  "mcp_calls": "0 - every GitHub read and write went through repo-scoped REST (issue body/comments 200, branch push OK, PR create 201) plus git. Channel declaration: the REST SEARCH endpoint is dead here exactly as the card reported - /search/issues answered 403 on all three queries INCLUDING its control ('repo:objectstack-ai/objectui plugin-timeline', a term in this very card's title), with 'sessions are bound to their configured repositories'. Repo-scoped endpoints work. Nothing was filed, so no dedup read was owed.",
  "open_questions": [
    {
      "question": "Should plugin-timeline's jest-dom omission be fixed at all, now that the census shows it is not a deviation from any uniform practice?",
      "options": [
        "A - leave it; it is 1 of 11 packages that leave `types` unset, and the toBeTruthy workaround from PR #8713 stands",
        "B - fix the instance with `types: [node, @testing-library/jest-dom]` (measured green; the shape 7 siblings use). NOT the literal one line the card named - `types: [@testing-library/jest-dom]` alone leaves `process` at TS2591",
        "C - fix the whole class: give every package that uses matchers a `types` entry, retiring the 83 leaning files"
      ],
      "recommendation": "B if the card should close on its instance, C if the gate is wanted - because C is exactly what gate rule 2 in the audit would enforce, and doing C first makes that gate land green instead of with an allowlist. A is defensible but leaves an unexplained workaround in the tree, which is the cost the card itself predicted."
    },
    {
      "question": "Is a parity gate wanted, and in which of the four shapes the audit proposes?",
      "options": [
        "A - rule 1 only (every divergence on lib/types/include/exclude must carry a comment on that key) - cheapest, and it is the rule that would have caught this card, objectui#8691 and the false plugin-designer comment",
        "B - rules 1+2 (also: no file's matchers may be typed only by another file's import) - adds the 83-file class",
        "C - all four including inert-declaration removal - largest ratchet allowlist",
        "D - none; the re-runnable census is enough and drift is caught by reading it"
      ],
      "recommendation": "A, then B as a follow-up. Rule 1 asserts nothing about which value is correct, so it cannot flatten the four deliberate divergences the audit measured, and it is the only rule whose violations are all genuinely unexplained today. Rule 3 (inert declarations) is tidying and should not gate anything."
    },
    {
      "question": "Does the report-only census script belong in-tree, or should only the audit document land?",
      "options": [
        "A - keep it (as shipped): it follows scripts/body-dialect-census.mjs precedent, is wired as `census:*` and not to CI, and the next card's gate needs exactly its resolver",
        "B - drop the script and keep only docs/audits/... - smaller diff, but the matrix goes stale the first time anyone edits a tsconfig"
      ],
      "recommendation": "A. The card's own words were 'a comparison anyone can write once'; a document is that comparison written once and then rotting. But this is a judgement call inside a card dispatched for a census, so it is flagged rather than assumed - B is one `git rm` away."
    }
  ],
  "out_of_scope_findings": [
    "NOT FILED, deliberately - all four are inside this card's own subject and are recorded with evidence in docs/audits/2026-09-test-tsconfig-parity-census.md, so filing them separately would fragment one ruling into four. (1) plugin-designer/src/__tests__/DashboardEditor.i18nTitle.test.tsx states that `.at()` type-checks nowhere in this repo; 6 of the 38 programs are on ES2022. (2) 83 files across plugin-detail/plugin-dashboard/fields/plugin-list have jest-dom matchers typed only by another file's import. (3) auth, plugin-charts and plugin-grid carry `src/**/*.d.ts` include patterns matching zero files on disk. (4) plugin-view and plugin-report exclude a load-bearing src/global.d.ts with no comment saying why. Say the word and each becomes its own unassigned issue.",
    "ALSO NOT FILED - the REST /search/issues endpoint answers 403 here including its own control, so a blind filing is the forbidden shape; the repo-scoped list endpoint is available if triage wants these split out."
  ]
}

---

Generated by Claude Code

[os-justin]

Contract review — accepted. PR #8720 is out of draft and armed. Rulings: Q1 → A, Q2 → A (then B), Q3 → A.

⭐ The census did what it was dispatched for: it changed the card's shape

42 test type-programs; 38 with a dedicated tsconfig.test.json (2,477 program root files, 38/38 green today), 4 fusing tests into the ordinary config. 23 divergent axes, of which — within the 35-package cohort that all extend the repo-root config — 4 are behavioural: types (6 effective values), lib (4), include (3), exclude (2).

⇒ this is the number nobody had, and it lands between the two hypotheticals I named: not "two packages differ on one key", not "no two alike". Four axes with a handful of values each, which is what makes rule 1 below tractable.

⭐⭐ And two ambient-type routes that are in NO tsconfig at all

a bare import of @testing-library/jest-dom augments matchers PROGRAM-WIDE, so 83 files across plugin-detail (40), plugin-dashboard (16), fields (15) and plugin-list (12) type-check only because a DIFFERENT file in the same package imports it;
and one triple-slash node reference on line 1 of one sdui-parser test file gives that package the Node globals its ten types-unset siblings lack.

⇒ the divergence is not confined to the configs I sent you to compare. 83 files whose type-checking depends on an unrelated file's import statement is a far larger and more fragile finding than the card's one line — delete one import and 83 files stop compiling, with nothing naming the dependency. And the sdui-parser probe is properly controlled: react-runtime TS2591, sdui-parser clean, two named-node controls lit.

⭐⭐⭐ The conditional fix was correctly NOT made — and the card's literal one line was wrong anyway

I authorised the one-liner only if the siblings uniformly name jest-dom. They do not:

21 of 35 do not name it; permissions documents its omission on purpose; fields uses matchers in 86 files with types unset.

⇒ the conditional was evaluated and came out false, and the fence held. ⚠️ And the card's literal remedy would have been wrong regardless: types: ['@testing-library/jest-dom'] alone leaves process at TS2591 — the working shape is types: ['node', '@testing-library/jest-dom'], established by a three-leg probe rather than by assumption.

⭐ Two controls that caught real errors

• A sweep bug caught by its own control: git pathspec 'pkg/src/**/*.test.tsx' matches nothing for flat-layout packages — plugin-map read 0 test files. ⇒ the final sweep greps TypeScript's own resolved program file list instead of globs. Third time tonight a control has caught a broken sweep before its zero was published.
• Deliberate-vs-accidental was MEASURED, not judged: adding src/**/*.d.ts to plugin-view's include — the shape 6 siblings carry — turns that project RED with 7 errors, because its own src/global.d.ts redeclares process and collides with @types/node. ⇒ load-bearing and undocumented, and "3 of 8 audit rows needed a tsc run to classify." An axis that looks like drift and is actually a constraint is exactly what a batch normalisation would have broken.

And the precondition I flagged as mattering unusually here was honoured: pnpm build 43/43 green before any tsc -p, because an unbuilt tree makes all 38 programs look broken identically.

⚠️ One environmental finding that is partly my doing

NOT MEASURED (not red): check-node-esm-load refused 19 of 37 entries on PROVENANCE — turbo shares one cache across every worktree of this checkout, so sibling agents' artifacts replayed into this tree.

⇒ running three dev agents concurrently in one checkout pollutes a gate's inputs, and the gate is right to refuse. ⭐ Declared as NOT MEASURED with the reason rather than reported as a failure or quietly skipped. I am the one holding concurrency at three; this is a cost of that, and it is worth knowing that at least one gate cannot be trusted from a worktree while siblings are building.

Rulings

Q1 → A. Leave it, and ⛔ this card does NOT close on its instance. The census proved plugin-timeline is 1 of 11 packages leaving types unset — not a deviation from any practice. Fixing that one package would be arbitrary: it would single out the package where a dev happened to trip, and B is not even the card's line (it needs node too). ⇒ PR #8713's toBeTruthy workaround stands; ⭐ its comment should cite this audit, so it stops being unexplained — that is the cost the card predicted and it is one sentence to pay. The instance was the doorway, not the subject.

Q2 → A, then B as a follow-up. Rule 1 — every divergence on lib/types/include/exclude must carry a comment on that key — for the dev's reason, which I adopt: it asserts nothing about which value is correct, so it cannot flatten the four measured-deliberate divergences, and it is the only rule whose violations are all genuinely unexplained today. And it would have caught this card, objectui#8691, and the false plugin-designer comment. ⛔ Rule 3 (inert declarations) is tidying and should gate nothing. ⛔ Implementing the gate is the next card, not this PR — I am filing it.

Q3 → A. Keep the script. "A document is that comparison written once and then rotting" is the correct reading of the card's own words, and it is this session's central theme: five hand-written counts wrong tonight, every one of them a comparison someone wrote once. census:*, report-only, not wired to CI, refusing only an empty population — with the empty-population guard observed failing under ablation. That is the shape.

What I am filing, and what I am not

⭐ Filing the 83-file class as its own card: matchers typed only by another file's import is a structural fragility with a blast radius (plugin-detail 40, plugin-dashboard 16, fields 15, plugin-list 12), it is rule 2's subject, and it is not a tsconfig divergence — so it does not belong inside this card's audit as a footnote. Filing the gate proposal (rule 1) too.

⛔ Not filing the other three separately — the false .at() generalisation, the three zero-matching src/**/*.d.ts includes, and the two undocumented global.d.ts exclusions. They are recorded with evidence in docs/audits/2026-09-test-tsconfig-parity-census.md, they are all rule 1's subject, and splitting one ruling into four is the fragmentation this queue keeps paying for. ⭐ They land as violations the moment rule 1 exists, which is the right way for them to be fixed.

Your channel declaration is exactly right and matches what two other seats measured tonight: /search/issues 403 on all three queries including its control — a term from this card's own title. Repo-scoped endpoints work; nothing was filed, so nothing was owed.

---

Generated by Claude Code

[os-justin]

os-dev-report

{
  "issue": 8714,
  "status": "done",
  "branch": "claude/issue-8714-test-tsconfig-parity-census",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8720",
  "premise_still_valid": true,
  "summary": "Repair dispatch on the red Lint job of PR 8720, not a fresh card. The census script this PR added hand-typed its entry guard; check:entry-guard rejected it. Replaced with the canonical predicate: added `import { isEntrypoint } from './invoked-as.mjs';` and rewrote the guard line as `if (isEntrypoint(import.meta.url))`. One file, 2 insertions / 1 deletion, nothing else touched. THE DOUBLE-REPORT AT :261 IS NOT TWO SPELLINGS AND NOT TWO SITES: it is one guard site, one spelling, counted twice because the checker's scanner is a `while ((m = re.exec(code)))` over `/process\\.argv\\[1\\]/g` and pushes one finding PER OCCURRENCE (scanFile, check-entry-guard.mjs:231-242), while line 261 spelled the token twice - once as the truthiness leg `process.argv[1] &&` and once as the comparand `=== process.argv[1]`. Measured, not inferred: `sed -n '261p' | grep -o 'process.argv[1]' | wc -l` = 2, and `grep -n -o` over the whole file returned exactly two hits, both on 261. The gate's own passing line confirms the unit is occurrences, not files or sites: it now reads '0 file(s) still hand-type one (0 occurrence(s))'. Both went with the one line, so no second edit was owed. `fileURLToPath` stays imported - line 79 still uses it for HERE/DEFAULT_ROOT.",
  "tests": "STEP 1, FAILURE OBSERVED FIRST, before any edit, in the fresh worktree at e1420f3f9: `node scripts/check-entry-guard.mjs` exit=1, reproducing the CI rows verbatim - '2 hand-typed entry guard(s) in scripts/ beyond the baseline' followed by two identical rows 'scripts/tsconfig-test-parity-census.mjs:261  process.argv[1]  - a hand-typed entry guard'. Same file, same line, same count as job 102298795341. STEP 2, AFTER THE EDIT: `node scripts/check-entry-guard.mjs --self-test` exit=0, 'check-entry-guard self-test: 63 cases pass'; `node scripts/check-entry-guard.mjs` exit=0, 'check:entry-guard: 76 scripts/ file(s) - no entry guard outside the baseline; 0 file(s) still hand-type one (0 occurrence(s), SHRINK-ONLY, objectui#6092); 69 export bindings, 69 of them inert on import'. Both re-run at the PUSHED head 5c9ae708f with a clean tree (0 modified) and still exit 0; check-lint-coverage, the sibling step in the same CI job, 46/46. NO BASELINE WAS TOUCHED - KNOWN_HAND_TYPED_GUARDS is still `new Map([])` and the diff is 2 lines in the census script only. STEP 3, THE GUARD FIRES IN BOTH DIRECTIONS. Direct: `node scripts/tsconfig-test-parity-census.mjs` exit=0, 25047 bytes / 211 lines beginning '# test-tsconfig parity census - 38 projects, 2477 program files' - the real census, not silence. Import: a scratch .mjs importing `main` and `DEFAULT_ROOT` by absolute path printed only its own sentinel, 'IMPORTED-OK typeof main = function | DEFAULT_ROOT ends with: objectui-pr8720', and `grep -c` for the census markers over its full stdout+stderr = 0. The sentinel is load-bearing: it proves the module actually LOADED, so the absence of census output is inertness and not a failed import. WHICH RUN GOES RED UNDER EACH DEGENERATE GUARD: a guard answering TRUE for everything reddens the IMPORT run - the census would print during import and the module's `process.exit(main(...))` would kill the importer before its sentinel line, so both the sentinel and the zero marker count would flip; it would also kill the repo's own test file, which imports `main` at module scope. A guard answering FALSE for everything reddens the DIRECT run - 0 bytes, exit 0. Neither degenerate guard can produce the pair actually observed. EXTRA LEG, THE BUG ITSELF, MEASURED RATHER THAN REASONED: the pre-edit content was restored verbatim from HEAD into a temporary probe inside scripts/ (2 occurrences of the hand-typed token confirmed on disk) and reached through a real symlink - exit 0, ZERO BYTES of output, the exact silent no-op the gate exists to prevent - while the same old file reached DIRECTLY printed the full 25047 bytes, which is why nothing in CI ever noticed. The repaired script through the same symlink prints the same 25047 bytes as its direct run. The probe carried a `trap ... EXIT INT TERM`, was removed, and removal was verified by OBSERVING state (`test -e` plus `git status --porcelain` showing only the census script) rather than by an exit code. STEP 4, THE REPO'S TEST FOR THAT SCRIPT: `pnpm exec vitest run scripts/__tests__/tsconfig-test-parity-census.test.ts scripts/__tests__/entry-guard-wiring.test.ts` from the repo ROOT with file paths, exit 0. Read from the JSON reporter rather than the text one so a suite death could not hide: numTotalTestSuites 6, numPassedTestSuites 6, numFailedTestSuites 0, numTotalTests 15, numPassedTests 15, numFailedTests 0, success true. CHANGESET: none added and none owed - check-changeset-presence.mjs exit 0, its own verdict line 'No source or published contract of a released package changed in this range, so no changeset is owed.' GOVERNED SURFACE: check-governed-queue-guard.mjs --test on the changed path returns 'NOT GOVERNED - 1 path(s) checked against 5 governed surface(s); none matched.' CONTROL BYTES: `grep -naP` over the edited file, no hits. NOT MEASURED, declared rather than skipped: the repo-wide `pnpm lint` and full `pnpm test` were not run - this diff is 2 lines in one unpublished scripts/ file and the CI Lint job is the owner of the farm-wide sweep. PUSHED SHA 5c9ae708fdf602a21452b56d494bd655391b84bf. PR draft state untouched, auto-merge not armed, nothing rebased, amended or force-pushed.",
  "mcp_calls": "2 - one issue_read get_comments to confirm the claim names this branch, one add_issue_comment for this report. Everything else was git and local node.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT FILED, no separate card owed - four observations, all inside this repair's own subject. (1) The census script exports `main` and scripts/__tests__/tsconfig-test-parity-census.test.ts imports it at module scope, so the import-inertness direction of this guard is already exercised by the repo's own suite - an always-true guard would not merely print, it would `process.exit` inside vitest. (2) check-entry-guard carries a SECOND rule beyond the hand-typed one, an import-safety population now reading '69 export bindings, 69 inert on import'; the census script's binding sits in it and passes. (3) The worktree needed `pnpm install` before the census script would run at all, because it imports `typescript`; the gate itself is wired to run BEFORE install by design, which is why the red CI step reached the file while the script itself could not have. (4) The card asked whether :261 was two spellings or one site counted twice - it is the latter, so the instruction 'if it is two spellings on one line, both must go' did not apply; both occurrences went with the single line regardless."
  ]
}

---

Generated by Claude Code