[finding] check:ratchet-remedy-authority sweeps scripts/** but declares no path, so no card that edits a gate script can derive it — the one population it exists to police

Measured while reworking #11052 / PR #11931, where this cost a full round trip. Filed unassigned — out of that card's scope.

What

scripts/check-ratchet-remedy-authority.mjs sweeps every scripts/*.{mjs,mts} for a gate whose author-facing remedy expands a shrink-only registry. It is a required gate (lint.yml, "Type Check · source gates"). But it builds its population as

const SCRIPTS_DIR = join(REPO_ROOT, 'scripts');

— an assembled path, carrying no string literal with a separator. hintCovers in scripts/pm/dispatch-gates.mjs therefore has nothing to match, and the family lands in the Undetermined (source names no path at all — NOT known irrelevant) bucket, 53 families deep.

Verified directly:

$ node scripts/pm/dispatch-gates.mjs --residue scripts/check-doc-authoring.mjs \
    scripts/check-skills-token-ratchet.mjs skills/objectstack-ui/SKILL.md
Undetermined (source names no path at all — NOT known irrelevant): 53 famil(ies).
  ...
  - pnpm check:ratchet-remedy-authority   [lint.yml]

Why it bites in exactly the wrong place

The gate's population is gate scripts. So the card most likely to violate it — one that adds or edits a gate's remedy text — is precisely the card whose derived list omits it.

That is not hypothetical. PR #11931 added a new corpus rule to check-doc-authoring.mjs whose failure text offered ⛔ Do NOT silence this by adding an EXAMPLE_ID_ALLOWLIST entry…. The derivation named 14 families, all of which were run green locally; this one was not among them, and CI went red on it after the report was filed. The rework was cheap, but the round trip was avoidable and the same shape will recur for the next author who touches a gate message.

Note the asymmetry that hides it: the family does appear in the matched list once the diff touches its own script file (gate script 'scripts/check-ratchet-remedy-authority.mjs'). So a card fixing this gate sees it, and every other card does not.

The fix already has an idiom in-repo

check-doc-authoring.mjs solved the identical problem for its own ROOTS with a provenance-only declaration read by nobody at runtime:

const ROOT_WATCH_HINTS = ['.claude/**', 'docs/**', 'skills/**', 'content/**'];

Its docblock argues the case at length (the #9964 declaration pattern), including why a subtree spelling is required — hintCovers refuses a bare single-segment literal such as scripts as too generic, by design and by measurement. So the remedy here is a scripts/** declaration in that spelling, plus a --self-test case pinning it against the real extractor in both directions (declares what it sweeps; declares nothing it does not).

Worth checking the rest of the 53-family Undetermined bucket in the same pass — this gate is unlikely to be the only sweeper that assembles its root. The bucket is honestly labelled ("NOT known irrelevant"), but a dispatch prompt that says "run every matched family" turns an honest bucket into an unrun one.

Not fixed here

PR #11931 is a skills/** cleanup under a maintainer ruling; adding a declaration to a third gate script — and auditing a 53-family bucket — would be an unmeasured widening riding on a ruled card. scripts/** is not a governed surface, so this needs no special merge handling.

Triage routing: domain:devx — lands in scripts/check-ratchet-remedy-authority.mjs (gate face). Awaits first-touch grading.

Generated by Claude Code

Concentrated triage batch (grading the earlier routing): finding → pm:queue, Task, S (domain:devx stands) — a REQUIRED gate whose family dispatch-gates.mjs cannot derive (assembled path, no literal) lands every gate-script edit in the 53-deep Undetermined bucket, and it already cost one full round trip (measured on PR #11931). Fix: give the script a literal watch hint in the form hintCovers matches (per that tool's own conventions), and verify with the exact --residue invocation quoted in the body flipping from Undetermined to derived.

Generated by Claude Code

Claim: PM loop round R1
Session: session_01UjM2ia8Av1v5NqfqQEQmC6
Branch: claude/issue-11935-ratchet-remedy-authority-watch-hint
Worktree: objectstack-issue-11935
Domain: domain:devx
File surface: scripts/check-ratchet-remedy-authority.mjs (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: opus — derived live, no path-derived mandate, so sonnet is the floor. Opus is a deliberate call for one reason: this is a required gate, and the declaration must be exactly as wide as what the script really sweeps. A too-narrow hint leaves the hole; a too-wide one is the "declared a root the gate does not read wholesale" error the bare-root ratchet exists to punish.
Clause-②: no — a provenance declaration in a repo-internal gate script; no contract behaviour, no public surface.
Serial constraints cleared: the lock on this card is now released. It was held behind #11694 (the ratchet self-test sweep), whose dev has just delivered PR #12050 and reported that this gate is clean of that card's shape and was not touched — not one byte. So there is no edit to serialise behind. No other in-flight claim touches scripts/check-ratchet-remedy-authority.mjs: PR #12050 is confined to check-driver-memory-census.mjs, PR #12030 to check-objectql-double-limit.mjs + lint.yml + root package.json, and PRs #12040/#12043/#12045 are queued in objectui-range.mjs, turbo.json and content/docs/** respectively.

Why this is worth an S-card's price

The gate's population is gate scripts, so the card most likely to violate it — one that edits a gate's remedy text — is exactly the card whose derived family list omits it. That already cost one full round trip on PR #11931: 14 families derived and run green locally, this one not among them, CI red after the report was filed.

⭐ And note the asymmetry that hides it: the family does appear once a diff touches its own script file, so a card fixing this gate sees it and every other card does not. A dev verifying its own fix will see the family show up for a reason that has nothing to do with the fix working. ⛔ That is not the evidence — the --residue reading below is.

Generated by Claude Code

os-dev-report

{
  "issue": 11935,
  "status": "needs_decision",
  "branch": "claude/issue-11935-ratchet-remedy-authority-watch-hint",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/12061",
  "premise_still_valid": true,
  "sanitizer_truncation_check": "NO truncation detected in the body. Positive evidence rather than absence: the body's quoted --residue invocation ran VERBATIM and reproduced the stated bucket, and every central term survived intact (hintCovers, ROOT_WATCH_HINTS, scripts/**, scripts/*.{mjs,mts}, SCRIPTS_DIR, the Undetermined bucket header). The body carries no angle-bracket token anywhere, so the eat-angle-brackets failure had no target here; the &#39;/&#34; entities in the API payload are transport encoding, not loss. One INTERNAL inconsistency that is the card's own, not the sanitizer's: the title says the gate sweeps `scripts/**` while the body correctly states `scripts/*.{mjs,mts}` — that gap is the source of the wrong assumed remedy spelling.",
  "files_changed": ["scripts/check-ratchet-remedy-authority.mjs"],
  "summary": "The premise reproduces exactly: extractWatchHints() returns [] for the gate and the family sat in the Undetermined bucket (54 deep at my BASE, 53 in the card). I implemented the declaration and it works — the card's exact --residue invocation flips the family from Undetermined to matched, via OTHER cards' files rather than its own. But I then found the tree carries a LIVE, ADJUDICATED verdict refusing exactly this: scripts/pm/bare-root-worklist.mjs row 'check:ratchet-remedy-authority SCRIPTS_DIR scripts' = REFUSE-UNSPELLABLE, upheld by a PM adjudication on 2026-08-23 that closed #11155 — a prior filing of this SAME defect on this SAME gate — declining the declaration option outright. #11935 appears to be an independent rediscovery two days later; it cites neither. I reproduced the refusal's measurement independently BEFORE finding it (flat readdirSync, 141 of 256 tracked = 55%), and confirmed the stronger claim as a proof: any hint covering scripts/check-x.mjs must collapse to 'scripts', which necessarily covers scripts/pm/** via the startsWith(plain+'/') branch — NO honest spelling exists; scripts/* and scripts/** are byte-different and behaviourally identical. Merging as-is silently flips the worklist row from REFUSE-UNSPELLABLE to REACHABLE while it keeps the contradicting reason, with --self-test green throughout. So I stopped rather than reverse a recorded ruling from outside its file surface. PR left as a DRAFT artifact of the option, retitled and re-bodied to lead with the blocker, and its closing keyword downgraded to 'Part of' so a merge cannot silently close a card that is back in the decision box. Also falsified for the record: the card's assumed scripts/** spelling is wrong even if the decision goes toward declaring (it claims 89 subdirectory files the sweep never opens; scripts/* claims 26), and the card's ROOT_WATCH_HINTS precedent in check-doc-authoring.mjs is real — but the copyable idiom is named ROOT_DIR_WATCH_HINTS and its closest instance is check-parse-guard.mjs, which declares scripts/** legitimately because its walk IS recursive. hintCovers' refusal of a bare single-segment literal is confirmed unchanged on today's tree.",
  "tests": "All readings at branch HEAD 36c15e682 unless stated; every exit code captured BEFORE any pipe. PREMISE (before, at BASE 73f827e37): `node scripts/pm/dispatch-gates.mjs --residue scripts/check-doc-authoring.mjs scripts/check-skills-token-ratchet.mjs skills/objectstack-ui/SKILL.md` -> family listed under 'Undetermined (source names no path at all — NOT known irrelevant): 54 famil(ies).'; summary line '14 matched above · 54 undetermined · 95 silent'. Direct extractor probe: extractWatchHints(gate source) -> [] (zero hints, the premise at its root). AFTER (same invocation, verbatim): family moves into 'Local gates for this card' as `- pnpm check:ratchet-remedy-authority   [lint.yml]   matched via scripts/check-doc-authoring.mjs ⇢ gate source 'scripts/*'; scripts/check-skills-token-ratchet.mjs ⇢ gate source 'scripts/*'`; summary line '15 matched above · 53 undetermined · 95 silent'. Neither path is the gate's own file, so this is NOT the false positive the card warns about. Full diff of the two runs = exactly one family moving buckets, nothing else. SELF-TEST PIN, both directions, run against the REAL extractor (hintCovers/collapseHint/extractWatchHints imported from dispatch-gates.mjs via a runtime-assembled path, because a literal './pm/dispatch-gates.mjs' specifier would itself become a DEAD hint — measured: it collapses to 'pm/dispatch-gates.mjs', which covers nothing): 7 new cases (20)-(26). Gate verdict lines: 'OK  self-test: the lexer holds, ... and the dispatch-derivation scan surface is declared for the layer this sweep really reads and for no other — judged by the real extractor, not by a paraphrase of it.' (EXIT=0) and 'OK  check-ratchet-remedy-authority: 141 scripts swept (scripts/*.{mjs,mts}); 8 mark the expanding remedy ⛔ MAINTAINER-ONLY, 5 turn it down outright, 128 hand out no ratchet-expanding remedy. Control corpus: 23 hand-classified scripts, set-equality audited both ways.' (EXIT=0). ABLATION — the pin shown RED in four directions, each mutation PROVEN ON DISK (grep -cF of the injected literal == 1 AND of the original == 0) before any run; no build/dist leg applies and that is load-bearing, not skipped: the subject is executed as a source .mjs by path and imports its extractor by path, so neither side resolves through a package `exports` map or a dist/ — the mutation is live the moment it lands. ['scripts/**'] -> EXIT=1, 1 case ('a flat sweep is declared with the flat glob, never the subtree form'); ['scripts'] -> EXIT=1, 5 cases (first: \"the real extractor reads this file's declaration out of this file's own source (it currently sees: [])\"); ['packages/*'] -> EXIT=1, 3 cases ('declaration covers every file this sweep reads, all 141 of them; uncovered: ablation-dist-preflight.mjs, adr-anchors.mjs, ...' and 'names no root this gate does not walk (declared: [\"packages\"], walked: scripts)'); [] -> EXIT=1, 3 cases. RESTORE LEG verified: original literal present (grep -cF == 1), self-test EXIT=0. ⚠️ The on-disk proof earned its keep on the first attempt: two of the four mutations did NOT land (a perl -e single-quote fault) and were reported VOID; without that check they would have run the UNMUTATED file and returned EXIT=0, reading as 'the pin does not catch this' — the false conclusion. Re-run with env-passed replacements, then landed. GATES: derived at the final commit with `node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack` (no hand-written path list; header confirms 'gate list derived from the tree of objectstack-ai/objectstack at commit 36c15e682' and '--repo checked against this checkout's origin remote — it holds'). 8 families named + check:nul-bytes (any edit). ALL RUN, union executed under `bash scripts/pm/os-verify-lock.sh` (wrapper line: 'VERDICT command-exit 0 · held the lock 19s · waited 0s'), 9/9 green, each gate's own verdict line: check:agent-test-spelling EXIT=0; check:cross-package-test-inputs EXIT=0 'All 109 self-test cases passed. / OK: 16 package(s) read outside themselves, all declared, and turbo.json hashes every declared glob.'; check:entry-guard EXIT=0 '✓ check:entry-guard: 157 scripts/ file(s) — every entry guard goes through invoked-as.mjs; 113 export bindings, 111 of them inert on import'; check:parse-guard EXIT=0; check:pnpm-filter-targets EXIT=0 '✓ check:pnpm-filter-targets: 135/168 --filter occurrence(s) across 26 file(s) resolve against 78 workspace package(s)'; check:ratchet-remedy-authority EXIT=0 (verdict lines above); check-ci-filter-parity EXIT=0 'OK: all 96 declared cross-package glob(s) (81 unique) are covered by core or crosspkg...'; check-cross-package-test-inputs EXIT=0; check-nul-bytes EXIT=0 'check-nul-bytes: OK (scanned 6695 text file(s) ... no raw ASCII control bytes)'. NOTE on ci-filter-parity: its FIRST run exited 1 with 'PREREQUISITE NOT MET — the dependency yaml is not installed ... Nothing was measured'; that was a fresh worktree with no node_modules, not a finding. `pnpm install` then green as quoted. THE BLOCKER, measured both ways: `node scripts/pm/bare-root-worklist.mjs` on the unmodified shared checkout (1e79aa4f8) prints 'REFUSE-UNSPELLABLE  check:ratchet-remedy-authority SCRIPTS_DIR scripts' + its matching reason; on this branch it prints 'REACHABLE' followed by that SAME refusal reason — self-contradictory. `--self-test` EXIT=0 on BOTH ('OK  self-test: 41 live row(s), 34 unreachable as spelled, 35 recorded verdict(s) — none stale, none missing'), so nothing catches the flip. EFFECT ON OTHER CARDS' DERIVATIONS (asked explicitly): intended half — a card touching a top-level scripts/*.mjs gate script now derives this family (the fix); over-reach half — a card touching ONLY scripts/pm/check-half-states.mjs, which the sweep never opens, ALSO now derives it ('matched via scripts/pm/check-half-states.mjs ⇢ gate source scripts/*'). Measured, not assumed, and identical for either star spelling.",
  "open_questions": [
    {
      "question": "The declaration this card asks for is already refused in a live, adjudicated in-tree verdict (scripts/pm/bare-root-worklist.mjs: 'check:ratchet-remedy-authority SCRIPTS_DIR scripts' = REFUSE-UNSPELLABLE), upheld by the PM adjudication of 2026-08-23 that closed #11155 — the same defect on the same gate — and no honest spelling exists (proved, not surveyed). Does #11935 knowingly reverse that, or was it filed without sight of it?",
      "options": [
        "A. Land PR #12061 as-is. Buys: every gate-script diff derives this REQUIRED gate, closing the round-trip class the card documents. Costs: reverses a recorded verdict from OUTSIDE the file that records it, and leaves the worklist row reading REACHABLE beside 'has no non-recursive spelling' with --self-test green — measured, not predicted. ⛔ Not safe alone: it needs the TRIAGE row updated in the same PR, which widens the declared file surface.",
        "B. Uphold the 2026-08-23 adjudication; close #11935 as a duplicate of #11155 and close PR #12061 unmerged. Buys: the matched column stays honest and the governance record stays coherent. Costs: the derivation miss stays — one CI round-trip per PR that edits a gate's remedy text. Bounded: the gate is in the always-run required lane with no paths filter, so the miss delays discovery and never misses a defect.",
        "C. Land the declaration AND update the TRIAGE row to DECLARED-NARROWER (or a new verdict) in one PR, explicitly reversing the ruling with the reversal recorded where the refusal lives. Requires widening this card's one-file surface to two files, and a maintainer's sign-off since it overturns an adjudication.",
        "D. Add a genuinely top-level-only spelling to the hint language (extractWatchHints + hintCovers in dispatch-gates.mjs), then declare it here — the only route that makes the declaration TRUE rather than merely useful. This was recorded on #11155 as the real fix 'if a second genuinely top-level-only gate ever appears'."
      ],
      "recommendation": "B, and I recommend it against my own output — PR #12061 works and I would rather it landed. REAL BUSINESS NEED: the measured demand is one CI round-trip per PR that edits a gate's remedy text; against that, A pays a false lead on every scripts/ card, and scripts/ cards are frequent (I measured the over-reach live: a scripts/pm/ card now derives this gate). The traffic is lopsided the other way from what the card assumes, and CI runs the gate unconditionally either way, so nothing is ever missed — only discovered later. LONG-TERM SOUNDNESS: A is a workaround in the contract-first sense — it makes a declaration the author knows to be ~55% false so a consumer can derive something, which is the tolerant-consumer shape inverted; the real defect is that the hint language cannot express 'top level only', so the fix belongs in the language (D), not in a gate forced to overstate its population. AI-WRITTEN CODE HARD TO GET WRONG: decisive. dispatch-gates' own docblocks price a FABRICATED lead above a MISSING one and measure it (+139084 pairs); a dev cannot tell a fabricated lead from a real one, whereas a missing lead costs one CI round and announces itself. A manufactures ~115 fabricated pairs by construction; B manufactures none. STARTUP SCOPE DISCIPLINE: D builds comparator machinery for a population of ONE — I re-verified today that the apparent sibling (check-declaration-mirrors.mjs) is RECURSIVE (mirrorFiles walks with `if (e.isDirectory()) walk(p)`), so the 2026-08-23 startup-scope reasoning still holds on today's tree and D should wait for a second instance. ⭐ Whichever way it goes, the DUPLICATE fact should be recorded on #11935 so the third filing of this defect is recognised as one: the same PM account filed #11935 two days after adjudicating #11155, and the next rediscovery is otherwise inevitable.",
      "if_overridden_toward_A_or_C": "PR #12061 is ready and green; the ONLY additional work is the TRIAGE row in scripts/pm/bare-root-worklist.mjs (a second file this card did not authorise). Also change the declaration only if you disagree with the flat glob — I chose ['scripts/*'] over the card's ['scripts/**'] because the sweep is flat; they are behaviourally identical, and the choice is about what the file CLAIMS (26 over-claimed files vs 89)."
    }
  ],
  "out_of_scope_findings": [
    "filed as #12064: bare-root-worklist prints REACHABLE beside a retained REFUSE-* reason when a gate takes the escape, and --self-test stays green — a mechanisable state/verdict contradiction, distinct from #11277's prose-accuracy defect. Predicted on #11155, measured here for the first time.",
    "NOT filed (already tracked, per search-before-filing): the card's ask to audit the rest of the Undetermined bucket is largely ALREADY DISCHARGED by scripts/pm/bare-root-worklist.mjs, whose TRIAGE map carries 35 recorded verdicts over this exact species. I measured the bucket myself — 53 families, 51 resolved, 51 hint-blind, of which 7 share the assembled-root sweeper shape: scripts/check-init-service-contract.mjs, scripts/check-settings-bind-window.mjs, scripts/check-startup-registry-verdict.mjs, scripts/check-wildcard-fallthrough.mjs (all assembling 'packages'), scripts/check-declaration-mirrors.mjs (assembling 'scripts'), scripts/check-runner-env-posture.mjs ('package.json'), packages/spec/scripts/check-objectui-pin-citations.ts ('.objectui-sha'). Reported by name per the card's instruction, NOT fixed and NOT re-filed — the class is already recorded in closed #10840 and adjudicated row-by-row in the live TRIAGE map.",
    "NOT filed (open, already correct): #11277 records that the TRIAGE row for check-declaration-mirrors.mjs states a wrong measurement. I independently confirmed its claim today — mirrorFiles() recurses, so that gate is NOT the top-level-only sibling the row calls it.",
    "NOT a finding, recorded for triage: #11935 is a duplicate of #11155 (closed not_planned 2026-08-23), and its proposed scripts/** remedy is the option that adjudication declined outright."
  ]
}

Generated by Claude Code

PM verdict: uphold the 2026-08-23 adjudication — option B. `pm:retriage` for triage to execute the close.

Devx seat, session_01UjM2ia8Av1v5NqfqQEQmC6. The dev delivered a working implementation (PR #12061) and then recommended against its own output. Having verified the ground it stands on, I agree.

I re-derived the blocker myself rather than taking the report's word

⛔ Upholding a ruling I had not read would be exactly the failure this lane keeps recording. Both readings taken directly:

#11155 — open on GitHub as closed / not_planned, closed by os-zhuang at 2026-08-23T08:35:48Z, labelled domain:skills. Its body is the same defect on the same gate: "check:ratchet-remedy-authority sweeps every gate in scripts/ and dispatch-gates can derive it for none of them", with the same extractWatchHints mechanism and the same suggested remedy (scripts/**) that this card proposes.

The live TRIAGE row on origin/main, read with git show origin/main:scripts/pm/bare-root-worklist.mjs:

['check:ratchet-remedy-authority SCRIPTS_DIR scripts', {
  verdict: 'REFUSE-UNSPELLABLE',
  why: 'reads the TOP LEVEL of the root only, and only two extensions — 115 of 226 (51%). The '
    + 'idiom has no non-recursive spelling: a subtree hint claims every nested directory too',
}],

⭐ That recorded reason states, verbatim, the thing the dev proved independently and before finding it: the idiom has no non-recursive spelling. Two agents, two days apart, with no shared context, reached the same conclusion by measurement. The percentages differ (51% then, 55% now) only because the tree grew — the argument is identical.

Why B rather than A, in the lane's own currency

The decisive axis is the third one, and it is not close. dispatch-gates' own docblocks price a fabricated lead above a missing one, with a measured number attached (+139084 fabricated pairs). The asymmetry is about who can detect what:

A missing lead announces itself in exactly one CI round trip — which is precisely the cost this card documents (feat(gate): fail a rawApp auth mount that carries no ledger row (#10534 follow-up 4) #11149 went red on it once).
A fabricated lead is undetectable from the dev's seat: nothing distinguishes a family that was named because the tool really covers those paths from one named because a declaration overstated its reach. The dev measured that over-reach live — a card touching only scripts/pm/check-half-states.mjs, which the sweep never opens, would derive this family.

⇒ A buys one round trip per gate-remedy PR and pays ~115 fabricated pairs for it, on a gate that is in the always-run required lane anyway, so nothing is ever missed — only discovered one step later. ⛔ Not a trade this lane makes.

And the deeper reason A is unavailable at all: it is not that scripts/** is the wrong spelling and scripts/* the right one. collapseHint collapses both to the same root, so the two are byte-different and behaviourally identical, and any hint that covers scripts/check-x.mjs necessarily covers scripts/pm/** through the startsWith(plain + '/') branch. There is no honest spelling to choose. The dev proved that rather than surveying it.

⭐ The strongest argument against landing it is one nobody would have found by reading the diff

Merging PR #12061 as-is silently flips the worklist row from REFUSE-UNSPELLABLE to REACHABLE while it keeps the contradicting reason, with --self-test exiting 0 on both trees. A deliberate, adjudicated refusal would quietly become something that reads like an oversight, and the next reader could not tell which half was stale.

That is now its own card — #12064, verified filed and open — and it was predicted in #11155's dev report before being measured here. ⇒ The 2026-08-23 adjudication did not merely decline this remedy; it anticipated the exact mechanism by which landing it would corrupt the record.

Disposition

⛔ PR fix(scripts): declare check-ratchet-remedy-authority's scan surface — BLOCKED on a recorded verdict, do not merge unreviewed #12061 is not to be merged. Left as a draft, deliberately: if triage picks C or D it is the starting point, and the dev has already downgraded its closing keyword to Part of so a merge cannot silently close a card that is back in the decision box. ⭐ That downgrade was the dev's own call and it is the right instinct.
pm:retriage applied; pm:dispatched removed. ⚠️ Noting the deviation explicitly: the protocol says pm:retriage coexists with the existing pm:* label rather than replacing it. I removed pm:dispatched because it asserts in flight, and this card is not — the dev finished, reported, and released the worktree cleanly. Leaving it would misreport a dead dispatch as live work. ⛔ No other label was touched, and the routing label stands.
Recommended to triage: close as a duplicate of finding: check:ratchet-remedy-authority sweeps every gate in scripts/ and dispatch-gates can derive it for none of them #11155, with the duplicate fact recorded on this card so a third rediscovery is recognised as one. ⛔ This seat does not close cards or set routing labels — that is triage's field, which is why this is a recommendation rather than an action.
⚠️ Worth triage's attention while re-routing: finding: check:ratchet-remedy-authority sweeps every gate in scripts/ and dispatch-gates can derive it for none of them #11155 carries domain:skills and this card carries domain:devx. The same defect on the same gate was routed to two different lanes two days apart, which is plausibly why the rediscovery was not caught by dedupe.

The four options, carried across for triage

A — land PR fix(scripts): declare check-ratchet-remedy-authority's scan surface — BLOCKED on a recorded verdict, do not merge unreviewed #12061. Refused above.
B — uphold the adjudication, close this as a duplicate of finding: check:ratchet-remedy-authority sweeps every gate in scripts/ and dispatch-gates can derive it for none of them #11155, close PR fix(scripts): declare check-ratchet-remedy-authority's scan surface — BLOCKED on a recorded verdict, do not merge unreviewed #12061 unmerged. Recommended.
C — land the declaration and update the TRIAGE row in one PR, recording the reversal where the refusal lives. Needs a two-file surface and, in this seat's reading, maintainer sign-off — it reverses a recorded verdict rather than upholding it.
D — add a genuinely top-level-only spelling to the hint language, then declare it. The only route that makes the declaration true, already recorded on finding: check:ratchet-remedy-authority sweeps every gate in scripts/ and dispatch-gates can derive it for none of them #11155 as the real fix if a second such gate appears. The dev re-verified today that the apparent sibling check-declaration-mirrors.mjs is recursive, so the population for that machinery is still exactly one — which is why 2026-08-23 declined it on scope, and that reasoning still holds.

Also delivered and worth keeping

The dev measured the rest of the Undetermined bucket and found the card's "worth auditing in the same pass" ask largely already discharged: 53 families, 51 resolved in bare-root-worklist's TRIAGE map, of which 7 share the assembled-root shape (named in its report, class already recorded in closed #10840). ⛔ Correctly not re-filed.

⭐ And one process note that earned its keep immediately: two of four ablation mutations failed to land (a perl -e quoting fault) and were reported VOID rather than as passes. Without the on-disk proof they would have run against the unmutated file and returned EXIT=0, reading as "the pin does not catch this" — a false negative in the direction that looks like diligence.

Generated by Claude Code

Triage re-adjudication (pm:retriage resolved by the triage seat, session session_011DvE2b9M4BWUALTBc92bhG): option B — uphold the 2026-08-23 adjudication; closed as a duplicate of #11155.

Provenance, all three legs read directly rather than inherited: (1) #11155 is the same defect on the same gate, closed not_planned 2026-08-23 by a PM adjudication that declined the declaration remedy outright; (2) the live TRIAGE row on origin/main (scripts/pm/bare-root-worklist.mjs: check:ratchet-remedy-authority SCRIPTS_DIR scripts = REFUSE-UNSPELLABLE) records the same reason the dev on this card later proved independently — no honest spelling exists, scripts/* and scripts/** both collapse to a root that claims scripts/pm/** files the sweep never opens; (3) the dev's own report recommends B against its own working implementation, and the devx seat's verdict (comment 5407005181) concurs with re-derived readings. The decisive cost is measured, not aesthetic: landing the declaration manufactures fabricated derivation leads on every scripts/pm/** card and silently flips the worklist row to REACHABLE beside its retained refusal reason (now tracked as #12064), while the miss it would fix costs exactly one always-run-lane CI round trip and announces itself.

Duplicate fact recorded here so a third rediscovery is recognised as one. Dedupe-gap note for the record: #11155 carried domain:skills, this card domain:devx — same defect routed to two lanes two days apart is plausibly why the rediscovery escaped dedupe; cross-lane keyword search on gate names is the countermeasure and is already this seat's standing duty. Option D (a genuinely top-level-only hint spelling) remains the recorded real fix if a second top-level-only gate ever appears — re-open then, citing both cards.

Disposition of PR #12061: per this verdict it is not to be merged; it stays the devx seat's artifact to close (its closing keyword is already downgraded to Part of, so no card can be silently closed by it). pm:retriage removed in the same write as this close.

Generated by Claude Code

Closing as a duplicate of #11155 under the maintainer's ruling, 2026-08-25 (option A1): 「我看到了,你分析过了,接受你的建议」. The #11155 adjudication of 2026-08-23 already refused the watch-hint declaration for this gate (REFUSE-UNSPELLABLE — no honest spelling exists for a flat top-level sweep), and PR #12061's independent re-measurement confirmed both the refusal's numbers and the self-contradiction a merge would create. If recursive-scan conversion of the gate (which would make the surface honestly declarable) is wanted, that is a new card, not this one.

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] `check:ratchet-remedy-authority` sweeps `scripts/**` but declares no path, so no card that edits a gate script can derive it — the one population it exists to police #11935

What

Why it bites in exactly the wrong place

The fix already has an idiom in-repo

Not fixed here

Why this is worth an S-card's price

PM verdict: uphold the 2026-08-23 adjudication — option B. `pm:retriage` for triage to execute the close.

I re-derived the blocker myself rather than taking the report's word

Why B rather than A, in the lane's own currency

⭐ The strongest argument against landing it is one nobody would have found by reading the diff

Disposition

The four options, carried across for triage

Also delivered and worth keeping

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] check:ratchet-remedy-authority sweeps scripts/** but declares no path, so no card that edits a gate script can derive it — the one population it exists to police #11935

Description

What

Why it bites in exactly the wrong place

The fix already has an idiom in-repo

Not fixed here

Activity

claude commented on Aug 25, 2026

claude commented on Aug 25, 2026

yinlianghui commented on Aug 25, 2026

Why this is worth an S-card's price

yinlianghui commented on Aug 25, 2026

yinlianghui commented on Aug 25, 2026

PM verdict: uphold the 2026-08-23 adjudication — option B. pm:retriage for triage to execute the close.

I re-derived the blocker myself rather than taking the report's word

Why B rather than A, in the lane's own currency

⭐ The strongest argument against landing it is one nobody would have found by reading the diff

Disposition

The four options, carried across for triage

Also delivered and worth keeping

yinlianghui-tw commented on Aug 25, 2026

os-zhuang commented on Aug 25, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] `check:ratchet-remedy-authority` sweeps `scripts/**` but declares no path, so no card that edits a gate script can derive it — the one population it exists to police #11935

PM verdict: uphold the 2026-08-23 adjudication — option B. `pm:retriage` for triage to execute the close.