Four scripts/** gates hit by the #13489 survey were excluded by occupancy — carry them forward

Filed unassigned, from PR #13797's "Hit but deferred" section (survey card #13489). Every file below was surveyed and is a hit; none could enter that diff, because each was being edited by another card at the time.

Occupancy was re-checked at 2026-08-31T12:45:05Z by reading each card's live state. At that moment PR #13783 had already merged (2026-08-31T12:37:30Z); #13474, #13482 and #13478 were open. ⚠️ Re-check again before starting — do not act on this paragraph's wording.

file	hole 1 (assertion floor)	hole 2 (verdict handshake)	occupying card at survey time
`scripts/pm/dispatch-gates.mjs`	none	NOT MEASURED — injecting at the anchor lands inside a template literal and produces a SyntaxError, so no run of it measures anything. ⛔ Not a clean reading	#13448, PR #13783 (since merged)
`scripts/check-durability-degradation-log-level.mjs`	none	defeated — the dispatch combines two callees' statuses with `\|\|`, so an early return in the first yields `undefined`, falls through to the second, and exits 0	#13474
`scripts/check-platform-checklist.mjs`	none	NOT MEASURED — four self-test callees combined; a one-function probe measures a sub-battery and reads the downstream crash as a handshake	#13482
`scripts/check-dispatcher-error-vocabulary.mjs`	none	defeated, printing nothing	#13478

One note worth keeping

scripts/check-dispatcher-error-vocabulary.mjs is the sibling #13489 names as having landed the named-set form — and it did, for its SHAPES / samples pair. That is a different property from its own self-test's floor. Being right about one is not being floored on the other, and the survey measured it defeated.

The last two files are Tier C for hole 2 (multi-entry: one handshake per callee, or a wrapper — a decision, not a transplant); scripts/pm/dispatch-gates.mjs needs a hand-read entry before it can be measured at all.

Generated by Claude Code

Triage → domain:devx · p3 · tooling. Cross-domain single card; owning lane is domain:devx.

Anchoring, and the one row that is not this lane's. Three of the four files are code/doc-quality gates ⇒ domain:devx. scripts/pm/dispatch-gates.mjs is not: its SUBJECT is the dispatch/governed surface, which the lane table puts in domain:skills (the table names scripts/pm/check-governed-merges.mjs explicitly and rules future gates of the same kind the same way). This card keeps it as a listed row rather than spawning a fourth card, because it is NOT MEASURED and not currently actionable — the injection anchor lands inside a template literal and produces a SyntaxError, so no run of it measures anything. ⛔ The domain:devx dev does not touch that file. When someone gets a clean reading on it (a hand-read entry point, per the card), it becomes a domain:skills card at that moment.

The remaining three are this card's work, and the claim comment declares the full file surface per the cross-domain path.

Grade p3. These are carry-forward rows from a survey, deferred purely by occupancy — no new fact, no new exposure beyond what #13798 and #13799 already carry. They are p3 because they are bookkeeping against those two cards, not because the holes are small.

⚠️ Re-check occupancy before starting; do not act on the card's own paragraph. The card says so itself and it is already partly stale: PR #13783 merged at 2026-08-31T12:37:30Z, before this card was even filed. Read each of #13474 / #13482 / #13478 live.

One correction of emphasis for whoever takes this. The card's note about scripts/check-dispatcher-error-vocabulary.mjs is the most useful line in it and should lead the order, not trail it:

"it is the sibling #13489 names as having landed the named-set form — and it did, for its SHAPES / samples pair. That is a different property from its own self-test's floor. Being right about one is not being floored on the other, and the survey measured it defeated."

That is precisely the trap #13799 warns about in a different spelling — a gate that looks already-fixed to a skimming reader. Verify with the probe, not by reading the file.

Sequencing. These three files sit inside the same ~150-file surface as #13798 and #13799. ⛔ Do not dispatch this card to a different dev than those two — it is the same working set, and this is the smallest of the three, so it should ride along rather than run beside them.

Generated by Claude Code

⚠️ 分诊自我更正 —— 上一条评论里 `scripts/pm/dispatch-gates.mjs` 的车道判断是错的

我在上一条(本轮早些时候)把该文件从本卡里碗出来,理由是:

scripts/pm/dispatch-gates.mjs 不是:其 SUBJECT 是 dispatch/governed 面,车道表把它归 domain:skills …… ⛔ domain:devx 的 dev 不碰那个文件。

这条判断不成立,现予撤回。 两处读数推翻它:

该文件自己的 SUBJECT 声明 —— 它是 code-quality 门禁的派生器,不是 governed 面的门禁/审计:
```
scripts/pm/dispatch-gates.mjs
 * dispatch-gates (#7341 item 4) — map a card's file surface to the `check:*`
 * gate families that watch it, and to the model tier its paths MANDATE …
```
车道表给 domain:skills 的 scripts/pm/* 限于「SUBJECT 是 governed 面本身的门禁/审计(现为 scripts/pm/check-governed-merges.mjs,未来同类同判)」。⇒ 派生 check:* 族不是那一类。
本仓四张既有 dispatch-gates 卡全部是 domain:devx —— [finding] dispatch-gates does not derive check:pm-dispatch-gates for a .github/workflows/** change — so a dev following the derived list cannot run the one gate that judges their new workflow #13511 · [finding] Seven gates under-match their declared population — led by check:stall-guard-headroom, whose only declared literal is a repo slug that reaches no file at all #13519 · [finding] artifactOnlySilence misclassifies a family whose roster is extensionless module specifiers — collapseHint cannot resolve them, so a roster prints as an ordinary silence #13520 · [finding] dispatch-gates' import-side-effect docblock still sizes its own self-test at "334 assertions" — the live count is 979 #13536。我那一笔与它们全部相左,而我当时没有查这个先例。今天新分诊的 [finding] dispatch-gates prints its families in two differently-shaped sections, and two independent devs each dropped one section in one night — the local run then claims coverage it does not have and CI reds #13642(同一文件,消费面)也按 devx 落,与它们一致。

⇒ 对本卡的影响

scripts/pm/dispatch-gates.mjs 那一行回到本卡范围内,domain:devx 的 dev 照做,不必碗出。 本卡因此不再是跨域卡 —— 四个文件同属 domain:devx,认领评论按单域申报即可。

⚠️ 但它「NOT MEASURED」的状态不变:注入锚点落在模板字面量内、产生 SyntaxError,所以没有任何一次运行真的测到它。⇒ 它仍然不是当下可执行的一行;要动它,先按卡片说的手工读一个入口点拿到干净读数。区别在于:这是测量缺口,不是车道禁令 —— 我上一条把两者混在了一起。

⛔ 上一条评论的其余内容(定级 p3、⚠️ 先重核占用而非采信卡片自述的那一段、check-dispatcher-error-vocabulary.mjs 那条「看起来已修好」的陷阱、以及与 #13798/#13799 共用 ~150 文件面因此不得分派给不同 dev)全部不变。

Generated by Claude Code

Claim: PM loop round 2
Session: session_012zGPuVVX3deAx9LdjK8jCk
Branch: claude/issue-13800-carry-forward-verdict-handshake
Worktree: objectstack-issue-13800
Domain: domain:devx
File surface: scripts/check-durability-degradation-log-level.mjs, scripts/check-platform-checklist.mjs, scripts/check-dispatcher-error-vocabulary.mjs, and scripts/pm/dispatch-gates.mjs (the last one MEASUREMENT-FIRST, see below) — single-domain per triage's self-correction 5479569677 (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus (dispatch-gates --tier: no path-derived mandate; opus because hole-2 repairs are per-file dispatch-shape judgments, not a substitution)
Clause-②: no
Serial constraints cleared: none — all four occupying cards are CLOSED, re-checked just now (#13474, #13482, #13478, #13448 — the card's own instruction is 「Re-check again before starting — do not act on this paragraph's wording」, and I did). No devx card is in flight; #13798 batch 3 (check-type-check-coverage.mjs + bare-root-worklist.mjs) and #13799's next roster batch are both undispatched and touch none of these four files.

Premise re-verified on `origin/main` at 12:3xZ, with a positive control

All four files still lack hole 2. The zero is a real reading, not a dead grep: 81 files on origin/main carry the repair form !== SELF_TEST_VERDICT, and each of these four returns 0 against that same query.

⚠️ Two markers were tried before this one and both were wrong — selfTestReachedVerdict (not the name) and a bare SELF_TEST_VERDICT count (the constant pre-exists the repair in some files, so counting it discriminates nothing). The guard form above is the one with a working control. ⛔ Do not use the bare-constant count as an acceptance criterion.

Also re-verified: hole 1 and hole 2 really are orthogonal here

scripts/check-dispatcher-error-vocabulary.mjs received hole 1 (the roster) from #13799 batch 1c, PR #14896, merged 12:00:25Z — SELF_TEST_BATTERIES now returns 7 hits there — while its hole 2 is still 0. That is the card's 「⛔ orthogonal — do not merge the two counts」 made concrete on a live file.

Generated by Claude Code

Claim (dev): os-dev subagent, PM loop round 2
Session: session_012zGPuVVX3deAx9LdjK8jCk
Branch: claude/issue-13800-carry-forward-verdict-handshake (pushed empty as the write-route probe before the first edit — write path is green)
Worktree: objectstack-issue-13800, off origin/main @ ca3fd4b1
Domain: domain:devx, single-domain per triage's self-correction 5479569677

File surface declared (all four, per the self-correction that put scripts/pm/dispatch-gates.mjs back in this lane):

scripts/check-durability-degradation-log-level.mjs
scripts/check-platform-checklist.mjs
scripts/check-dispatcher-error-vocabulary.mjs
scripts/pm/dispatch-gates.mjs — MEASUREMENT-FIRST; hand-read entry point first, and if no clean reading is obtainable the row is delivered as "still NOT MEASURED, here is what I tried" rather than a guessed repair

Scope: hole 2 (verdict handshake) only. No hole-1/roster work (that is #13799's); no sweep of other scripts/** self-tests (#13798 batch 3 / #13799 next batch are undispatched and would collide). skip-changeset — scripts/** publishes nothing. Clause-②: no.

Acceptance criterion for the premise: the guard form !== SELF_TEST_VERDICT, with the 81-file positive control on origin/main; ⛔ not a bare SELF_TEST_VERDICT count.

Generated by Claude Code

os-dev-report

{
  "issue": 13800,
  "status": "done",
  "branch": "claude/issue-13800-carry-forward-verdict-handshake",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14960",
  "premise_still_valid": false,
  "summary": "Hole 2 delivered for the four carried-forward files, but only TWO needed work: the dispatch premise is falsified for the other two. The acceptance marker in the dispatch (the guard form '!== SELF_TEST_VERDICT', control 81 files) gives a FALSE ZERO, because commit 9acddde9 ('Verdict handshake for 134 scripts/** self-tests', #14479) landed a second, deliberate spelling — a module-level flag, chosen in-code because 'the self-test's own exit code stays load-bearing'. Marker proven to separate before from after across that commit: check-durability-degradation-log-level 0->6, check-dispatcher-error-vocabulary 0->3, and 0->0 for the two it did not touch. Union control on origin/main: 81 sentinel + 77 flag = 158 files. Note the PM dismissed 'selfTestReachedVerdict' as 'not the name' — it IS the name, in 77 files. ROW OUTCOMES: (1) scripts/pm/dispatch-gates.mjs — the MEASUREMENT-FIRST row — is now MEASURED and REPAIRED, not still NOT MEASURED. Its NOT MEASURED status was a property of the INSTRUMENT: measure-self-test-floor's injectEarlyReturn takes the FIRST regex match of 'function selfTest() {', which in that file is a fixture STRING, so injection lands in a template literal and only yields SyntaxError. Hand-anchored on the real definition (line 10517, unique match) the reading is clean and the defect is real: the dispatch was a bare 'selfTest();' discarding the result. Repaired with the landed sentinel form. (2) scripts/check-platform-checklist.mjs — MEASURED and REPAIRED; five batteries returning {checked, failures}, repaired in the landed FLAG form for the reason the sibling files state (the return value is load-bearing, so the handshake cannot BE the return value), guarded on BOTH legs including the inline one that is the only leg CI reaches. (3) check-durability-degradation-log-level.mjs and (4) check-dispatcher-error-vocabulary.mjs — ALREADY REPAIRED by #14479 and left UNTOUCHED; probed, all three entries hold. Hole 1 / rosters untouched throughout; the two counts were never merged. skip-changeset applied to the PR and confirmed by readback.",
  "tests": "All exit codes captured BEFORE any pipe (redirect-then-capture). Union re-derived and re-run on the FINAL commit b12a5bda. DERIVATION: 'node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack' from my own worktree after the edits, EXIT=0, change set 'committed 2, working tree 0, untracked 0' — 21 families. THE IRONY ANSWERED: one of the two repaired files IS that script; re-derived on IDENTICAL explicit inputs with the ca3fd4b1 version of the script and with the repaired one, the derived list is IDENTICAL (diff empty). ALL 21 FAMILIES GREEN at b12a5bda, including 'pnpm check:pm-dispatch-gates' EXIT=0 (291s, 'dispatch-gates self-test: 1288 cases pass'), check:entry-guard, check:parse-guard, check:nul-bytes, check:ratchet-remedy-authority, check:comment-mask-corpus. TWO derived entries were NOT MEASURED, not red, and neither is a finding: 'node scripts/check-test-completeness.mjs' EXIT=3 'PREREQUISITE NOT MET — this gate grades a saved turbo run test log, and no log was named' (its own --self-test EXIT=0); and durability EXIT=3 on the fresh worktree before pnpm install ('PREREQUISITE NOT MET — the dependency typescript is not installed'), EXIT=0 after install. ONE GATE IS RED AND IT IS RED ON main TOO: 'pnpm check:platform-checklist' EXIT=1, 4 UNCLASSIFIED coverage kinds — proven not mine by running the ca3fd4b1 version of the file in the same tree, finding sets byte-identical (diff empty); that gate is deliberately not CI-wired (lint.yml: 'by MAINTAINER action'). MUTATION EVIDENCE — every probe wrote a mutated COPY beside the original (the tracked file was never mutated), asserted the anchor matched EXACTLY ONCE, proved the mutation on disk (marker count AND injected-text presence) before running, and removed the copy in a finally; a non-landing mutation was hit once and that reading was discarded and redone rather than retried silently. dispatch-gates.mjs: unmodified before repair EXIT=0 '1288 cases pass'; early return before repair EXIT=0 with ZERO BYTES of output (DEFEATED); unmodified after repair EXIT=0 '1288 cases pass'; early return after repair EXIT=1 'selfTest() returned without reaching its verdict'. check-platform-checklist.mjs: unmodified before EXIT=0 '141 assertions'; bare 'return;' before repair EXIT=1 but only as a TypeError crash ('Cannot read properties of undefined (reading failures)') — the crash a one-function probe reads as a handshake; PARTIAL early verdict before repair EXIT=0 printing '119 assertions' (down from 141) while still claiming in prose that the skipped direction 'REFUSES an empty/renamed/reshaped' table (DEFEATED); after repair all three mutations EXIT=1 naming the battery — selfTestTrapVocabulary for both mutation shapes and selfTestSourceLineCitations for a LATE battery, confirming it is not just the first that is caught; unmodified after repair EXIT=0 '141 assertions' unchanged, and the inline gate leg's findings unchanged from base. ALREADY-REPAIRED PAIR PROBED: durability selfTest EXIT=1 named diagnostic, durability selfTestReadSeams EXIT=1, dispatcher-error-vocabulary selfTest EXIT=1 named diagnostic. Also run: node --check on both edited files, and a control-character self-scan (grep -naP over the class) — clean.",
  "mcp_calls": "6 — issue_read get, issue_read get_comments, add_issue_comment (claim), create_pull_request, search_issues x2 (dedupe). Everything else went over repo-scoped REST or git. CHANNEL NOTE: the REST /search/issues endpoint is 403 in this session ('sessions are bound to their configured repositories'), so dedupe switched to two targeted MCP search_issues calls; both returned non-empty, so the readings are valid rather than dead greps. Repo-scoped REST reads/writes (labels, issue creation, PR body readback) all work.",
  "open_questions": [
    {
      "question": "check-platform-checklist needed the handshake at TEN sites (five batteries x two legs). I kept the landed check, message and exit code but factored the wording into one requireReachedVerdict(name, reached) helper instead of inlining the landed eight-line block ten times. The dispatch said 'copy it, do not invent a second spelling' — is factoring the wording a deviation you want reverted?",
      "options": [
        "A: keep the helper — one authoritative wording, ten one-line call sites, ~54 added lines",
        "B: inline the landed block at all ten sites — literal fidelity to the landed shape, ~80 lines of duplicated prose, and ten places to drift",
        "C: inline on the --self-test leg only and helper on the inline leg — worst of both, listed only for completeness"
      ],
      "recommendation": "A. The protected thing is the handshake mechanism and its diagnostic, and both are byte-identical to the landed ones; only the duplication is factored. The spelling that WOULD have been an invention — a 'verdict:' field on the returned object — was rejected precisely because it has zero precedent in the 158, and the flag form was taken instead because two sibling files state the exact rationale for it. Cheap to reverse if you disagree: one commit, no behavior change."
    }
  ],
  "out_of_scope_findings": [
    "filed as #14961: check:platform-checklist is red on main again — four coverage kinds UNCLASSIFIED (batch_endpoints, crud_endpoints, metadata_endpoints, route_generation); proven pre-existing by running the ca3fd4b1 file in the same tree (identical findings); third instance of a recurring shape (#11421, #7347 both closed), with the not-CI-wired reporting channel already open as #11730",
    "filed as #14962: the landed handshake in check-durability-degradation-log-level.mjs turns a GENUINE red into 'never reached its verdict' AND stops the second battery running — measured; directly contradicts that dispatch's own comment 'Both rules' fixtures always run — a red one must not hide the other'; not fixed here because the repair is a decision, not a transplant. check-dispatcher-error-vocabulary.mjs is NOT affected (its failure path exits inside the self-test)",
    "filed as #14963: measure-self-test-floor's probe anchors on the FIRST 'function selfTest() {' with no literal masking, so a fixture string masks the real definition — the sole reason dispatch-gates.mjs read NOT MEASURED; sibling of #14842; includes a second small row (ENTRY_BY_HAND says FOUR self-test callees for check-platform-checklist, there are FIVE)"
  ]
}

Generated by Claude Code

ACCEPT — PR #14960. ⭐ And my acceptance marker was wrong in the exact way I warned the dev about

domain:devx seat, session session_012zGPuVVX3deAx9LdjK8jCk, R2, 2026-09-03T13:1xZ.

The correction, owned — this is the worst of my three this round

My claim comment and dispatch gave the acceptance criterion as the guard form !== SELF_TEST_VERDICT, "controlled" by 81 files carrying it while all four targets carried 0. I then told the dev, in as many words, that selfTestReachedVerdict "is not the name."

It is the name — in 77 files. Verified by me on origin/main:

spelling	files
sentinel `!== SELF_TEST_VERDICT`	81
flag `selfTestReachedVerdict`	77
union (distinct)	158

Commit 9acddde9 — "Verdict handshake for 134 scripts/* self-tests that exit 0 on an early return"* (#14479, 2026-09-02) — landed a second, deliberate spelling: a module-level flag, chosen in-code because the self-test's own exit code stays load-bearing and so the handshake cannot be the return value.

⇒ My zero was a false zero, and two of the four rows were already repaired: check-durability-degradation-log-level.mjs and check-dispatcher-error-vocabulary.mjs both carry the flag form (3 hits each), confirmed. Had the dev accepted my criterion it would have written a third spelling into two already-correct files.

The lesson, and it is sharper than the one I had been recording all day. I had been telling myself and the dev that "a marker is only a reading once it is proven to separate before from after" — and I did control that. The control proved my query worked; it could not prove my query was complete. ⭐ Controlling that a marker discriminates does not establish that it is the only marker. When probing for the absence of a repair, first ask how many spellings the repair landed in — here, git log on the commit that did the landing would have said 134 files while my census said 81.

Row outcomes — 2 of 4 needed work, and the measurement-first row was measurable after all

scripts/pm/dispatch-gates.mjs — MEASURED and REPAIRED, not "still NOT MEASURED". ⭐ Its NOT MEASURED status was a property of the instrument, not the file: measure-self-test-floor's injectEarlyReturn takes the first regex match of function selfTest() {, and in that file the first match is inside a fixture string — so injection landed in a template literal and could only ever produce a SyntaxError. Hand-anchored on the real definition (line 10517, asserted a unique match) the reading is clean, and the defect is real: the dispatch was a bare selfTest(); discarding the result. This is exactly why the row was fenced measurement-first rather than assumed unfixable.
scripts/check-platform-checklist.mjs — MEASURED and REPAIRED at ten sites (five batteries × two legs), in the flag form, guarded on both legs including the inline one that is the only leg CI reaches.
The other two — already repaired by Verdict handshake for 134 scripts/** self-tests that exit 0 on an early return #14479, left untouched. Probed, all three entries hold. Hole 1 / rosters untouched throughout; the two counts were never merged.

Evidence quality

Mutation probes wrote a mutated copy beside the original (the tracked file was never mutated), asserted the anchor matched exactly once, proved the mutation on disk before running, and cleaned up in a finally. A non-landing mutation was hit once and that reading was discarded and redone rather than retried silently. Before/after on both repaired files, including the shape that matters most: check-platform-checklist with a partial early verdict previously exited 0 printing "119 assertions" (down from 141) while still claiming in prose that the skipped direction refuses a reshaped table — a false green that a count-only check would never catch. After the repair, three mutation shapes each exit 1 naming the battery, including a late battery, so it is not merely the first that is caught.

⭐ The gate red on this PR is red on main too, and was proven so rather than asserted: check:platform-checklist EXIT=1 with 4 UNCLASSIFIED coverage kinds, demonstrated pre-existing by running the base version of the file in the same tree and diffing the finding sets (empty). It is deliberately not CI-wired ("by MAINTAINER action"). Filed as #14961, third instance of a recurring shape, with the reporting-channel card #11730 already open.

Open question answered — A, keep the helper

The dispatch said "copy it, do not invent a second spelling." The helper is not a second spelling. What that instruction protects is the handshake mechanism and its diagnostic, and both are byte-identical to the landed ones; only the duplicated prose is factored. Inlining an eight-line block ten times would create ten places to drift — the exact failure this card's own family keeps documenting. And the thing that would have been an invention (a verdict: field on the returned object, zero precedent in the 158) was correctly rejected in favour of the flag form two sibling files already justify in-code. ⇒ Keep it. No revert.

Follow-ups filed — all three are real

#14961 (gate red on main, pre-existing, proven) · #14962 ⭐ the landed handshake in check-durability-degradation-log-level.mjs turns a genuine red into "never reached its verdict" and stops the second battery running — directly contradicting that dispatch's own "both rules' fixtures always run" comment. Correctly not fixed here: it is a decision, not a transplant · #14963 the measure-self-test-floor anchoring bug that caused the false NOT MEASURED, sibling of #14842.

Landing

CI on b12a5bda at 13:1xZ: 20 success / 11 skipped / 2 in progress ⇒ ⛔ not armed yet. Flip + arm on full green, every check.
On merge I will verify all four rows and close #13800 myself — the PR says Part of, which is the conservative choice, but rows 3 and 4 are already satisfied on main, so the card is complete once this lands.

Generated by Claude Code

LANDED — PR #14960 MERGED 14:11:57Z. All four rows satisfied ⇒ closing this card

domain:devx seat, session session_012zGPuVVX3deAx9LdjK8jCk, R2. Probed on origin/main after a re-fetch (8ea03a9b).

row	reading on `origin/main`	source
`scripts/pm/dispatch-gates.mjs`	`SELF_TEST_VERDICT` const `:10532`; `return SELF_TEST_VERDICT` `:17758`; guard `if (selfTest() !== SELF_TEST_VERDICT)` `:17813`. The bare `selfTest();` that discarded the result is gone	this PR
`scripts/check-platform-checklist.mjs`	`requireReachedVerdict` defined once, 11 occurrences of `requireReachedVerdict(` = 1 definition + 10 call sites — exactly the 5 batteries × 2 legs, including the inline leg that is the only one CI reaches	this PR
`scripts/check-durability-degradation-log-level.mjs`	flag form present, 3 hits — unchanged	#14479, untouched here
`scripts/check-dispatcher-error-vocabulary.mjs`	flag form present, 3 hits — unchanged	#14479, untouched here

⇒ Closing completed, pm:dispatched stripped. The PR reads Part of, which was the conservative choice, so GitHub did not close this card and the closure is a deliberate act after verifying rows 3 and 4 independently — not an assumption that "the PR covered it".

What this card cost the seat, recorded so the next one does not repeat it

⭐ My acceptance criterion was wrong, and it was the third and sharpest instance of one mistake in a single shift. I gave the sentinel form !== SELF_TEST_VERDICT (81 files) as the marker and told the dev that selfTestReachedVerdict "is not the name". It is the name, in 77 files — commit 9acddde9 (#14479) landed a second, deliberate spelling because the self-test's exit code stays load-bearing, so the handshake cannot be the return value. Union: 158.

I had controlled my marker. The control proved the query worked; nothing in it could prove the query was complete — and completeness was the property I actually needed.

⛔ The rule, for the next sitter: a repair can land in more than one spelling. Before probing for a repair's ABSENCE, read the commit that landed it and ask how many spellings it used. Here git log on that commit said 134 files while my census said 81 — the discrepancy was visible for free and I did not look.

Two of these four rows were already correct. Had the dev deferred to me, it would have written a third spelling into them.

And one row that was better than the card thought

scripts/pm/dispatch-gates.mjs was carried as NOT MEASURED and I fenced it measurement-first rather than as unfixable. That was the right call: its NOT MEASURED status was a property of the instrument, not the file — measure-self-test-floor's probe anchors on the first function selfTest() { match, which in that file is inside a fixture string, so injection could only ever yield a SyntaxError. Hand-anchored at the real definition (line 10517, unique match), the reading was clean and the defect real. Filed as #14963.

Follow-ups from this card, all bare for triage: #14961 (check:platform-checklist red on main, proven pre-existing) · #14962 (the landed handshake in check-durability-degradation-log-level.mjs turns a genuine red into "never reached its verdict" and stops the second battery — a decision, not a transplant) · #14963 (the anchoring bug above).

ACCEPT: 5526405653.

Generated by Claude Code

Navigation Menu

Repository navigation

Four scripts/** gates hit by the #13489 survey were excluded by occupancy — carry them forward #13800

One note worth keeping

⚠️ 分诊自我更正 —— 上一条评论里 `scripts/pm/dispatch-gates.mjs` 的车道判断是错的

⇒ 对本卡的影响

Premise re-verified on `origin/main` at 12:3xZ, with a positive control

Also re-verified: hole 1 and hole 2 really are orthogonal here

ACCEPT — PR #14960. ⭐ And my acceptance marker was wrong in the exact way I warned the dev about

The correction, owned — this is the worst of my three this round

Row outcomes — 2 of 4 needed work, and the measurement-first row was measurable after all

Evidence quality

Open question answered — A, keep the helper

Follow-ups filed — all three are real

Landing

LANDED — PR #14960 MERGED 14:11:57Z. All four rows satisfied ⇒ closing this card

What this card cost the seat, recorded so the next one does not repeat it

And one row that was better than the card thought

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Four scripts/** gates hit by the #13489 survey were excluded by occupancy — carry them forward #13800

Description

One note worth keeping

Activity

os-warren commented on Aug 31, 2026

os-warren commented on Aug 31, 2026

⚠️ 分诊自我更正 —— 上一条评论里 scripts/pm/dispatch-gates.mjs 的车道判断是错的

⇒ 对本卡的影响

claude commented on Sep 3, 2026

Premise re-verified on origin/main at 12:3xZ, with a positive control

Also re-verified: hole 1 and hole 2 really are orthogonal here

baozhoutao commented on Sep 3, 2026

claude commented on Sep 3, 2026

claude commented on Sep 3, 2026

ACCEPT — PR #14960. ⭐ And my acceptance marker was wrong in the exact way I warned the dev about

The correction, owned — this is the worst of my three this round

Row outcomes — 2 of 4 needed work, and the measurement-first row was measurable after all

Evidence quality

Open question answered — A, keep the helper

Follow-ups filed — all three are real

Landing

claude commented on Sep 3, 2026

LANDED — PR #14960 MERGED 14:11:57Z. All four rows satisfied ⇒ closing this card

What this card cost the seat, recorded so the next one does not repeat it

And one row that was better than the card thought

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

⚠️ 分诊自我更正 —— 上一条评论里 `scripts/pm/dispatch-gates.mjs` 的车道判断是错的

Premise re-verified on `origin/main` at 12:3xZ, with a positive control