From 9fcce366eb63c458c739c26433444e7fb1d78594 Mon Sep 17 00:00:00 2001 From: Claude Date: Mon, 2 Nov 2026 23:48:28 +0000 Subject: [PATCH] chore(pm): retire half-state row H34; H2 reports every card it named H34 fired only on a pm-tracked, assigned card with no claim the marker reads, which is exactly the card H2 reports. Delete H34, the near-miss marker and reader only it used, its self-test cases, its band row and its sweep call, and re-point the prose that named it to H2 (assigned) and H1 (unassigned). Co-Authored-By: Claude Claude-Session: https://claude.ai/code/session_01J8VVAmM3BRQQcQSA67VE3L --- scripts/pm/check-half-states.mjs | 414 +++---------------------------- 1 file changed, 33 insertions(+), 381 deletions(-) diff --git a/scripts/pm/check-half-states.mjs b/scripts/pm/check-half-states.mjs index 1f6d84fe3e7..c3c39e8e33b 100644 --- a/scripts/pm/check-half-states.mjs +++ b/scripts/pm/check-half-states.mjs @@ -1197,9 +1197,9 @@ export function h1DispatchedNoAssignee(issue) { * 严格,修法是全舰队向文档拼写收敛,⛔ 不放宽谓词)」. So a dash-written or * fullwidth-written claim is a MALFORMED claim, not an unrecognised dialect, * and the repair direction is the WRITE side. ⛔ Do not widen this class to - * accept another separator — that is the one change the ruling closes. What - * the patrol gained instead is visibility: `h34ClaimShapedNonCanonicalSeparator` - * reports the near miss on its own row without redefining what a claim IS. + * accept another separator — that is the one change the ruling closes. Such a + * card stays visible without redefining what a claim IS: assigned, it is H2's + * row (no claim the marker reads); unassigned, H1's. * * ⚠️ The class was `[::]` until #12090 — U+003A written TWICE, not "ASCII or * fullwidth" as three separate readers (this file's own H33 note, the filing @@ -1207,7 +1207,7 @@ export function h1DispatchedNoAssignee(issue) { * never matched here, measured by codepoint and by a live probe over 172 * dispatched cards (0 fullwidth claims). Collapsed to a single `:` so the code * stops implying an affordance it never honored; behaviour is byte-identical, - * and the fullwidth spelling now surfaces on the H34 row like every other + * and the fullwidth spelling reads as no claim, like every other * non-canonical separator. */ export const CLAIM_COMMENT_MARKER = /^\s*>?\s*Claim(?:ed)?\s*:/mi; @@ -1417,8 +1417,7 @@ export function signatureRefusalText(row, { tool, where = 'the body' } = {}) { // markdown bullet `* Claim: …` would become ` Claim: …` and read as a claim. // This file already rules on that shape from the other side — H20's case // 「a bulleted `Claim:` is still not a claim comment」 pins `- Claim:` as NOT a -// claim, and `CLAIM_NEAR_MISS_MARKER`'s docblock names the bullet hazard by -// name — so undecorating a `*` bullet would accept through decoration exactly +// claim — so undecorating a `*` bullet would accept through decoration exactly // what the `-` bullet is refused for. A bullet is a LIST, not a decoration: // the marker for one is followed by whitespace, the decoration for the other // is not, and that is the whole discriminator. @@ -1432,8 +1431,8 @@ export function signatureRefusalText(row, { tool, where = 'the body' } = {}) { * A markdown LIST ITEM's opening — a bullet or an ordered marker — and the one * prefix `markerMatches` refuses to undecorate through. * - * `[ \t]`, ⛔ never `\s`, on both sides: `\s` matches a NEWLINE, which is the - * defect `CLAIM_NEAR_MISS_MARKER`'s docblock measured the hard way — a hyphen + * `[ \t]`, ⛔ never `\s`, on both sides: `\s` matches a NEWLINE, which is a + * defect this file has measured the hard way — a hyphen * is the commonest line-opening character in this repo's comment bodies, so a * class that can span lines reads an ordinary list as a directive. * @@ -1500,12 +1499,9 @@ export function markerMatches(marker, text) { * word, which is exactly where every measured miss lives. * * ⚠️ Deliberate overlaps, stated rather than hidden: - * · `separator` overlaps H34 on the CLAIM side (`nearMissClaimSeparators`). - * H34 is a ROW with a remedy, gated to cards H2 is already reporting; this - * is a census over every thread the sweep already holds. Two readers of one - * shape answering two different questions, ⛔ not a second vocabulary for - * the claim separator: H34 keeps naming the codepoint and prescribing the - * rewrite. + * · `separator` on the CLAIM side is the shape H2 reports as claimless on an + * assigned card; this is a census over every thread the sweep already + * holds, ⛔ not a second vocabulary for the claim separator. * · `list-item` is the shape `markerMatches` refuses to undecorate through, * named here so that refusal is a reading rather than a silence. * @@ -1776,9 +1772,8 @@ export function releaseAnnouncementHeading(formId, prefix, line) { * * A form whose remainder is empty is skipped: 「Release:」 alone carries no * record, and reporting it would fill the clause with headings. That is the - * conservative half, `looksLikeClaimContent`'s judgement at H34 applied one - * notch looser — this is a census, and a threshold as strict as H34's would - * recreate the silence it exists to end. + * conservative half — this is a census, and a stricter content threshold + * would recreate the silence it exists to end. * * @param {{ id?: number, body?: string }[]} commentRows — REST rows, NOT bodies: * the row names the COMMENT, and an id is what a seat opens. @@ -8155,9 +8150,8 @@ export function isTriageRulingComment(body) { * `CLAIM_COMMENT_MARKER` wants the colon directly after the word, and here the * word is followed by ` pointer`, so the marker has never matched it. Under the * same 2026-08-11 ruling that governs the dash spellings it is a MALFORMED - * claim — and unlike those it is invisible to H34 as well, which reports only a - * punctuation separator (that gap was measured and deliberately left open; see - * H34's header). The paragraph's conclusion is unaffected: it rests on the + * claim, read as no claim at all like the dash spellings. The paragraph's + * conclusion is unaffected: it rests on the * first specimen, which is real, and on the live count above. * * ## ⚠️ The inherited blind spot, and what it is NOT (#12090) @@ -8183,9 +8177,9 @@ export function isTriageRulingComment(body) { * takes on every unrecognised spelling (H17's extractor, H20's branch shape), * never a fabricated row. * - * Where those claims DO surface is H34, which reports the near miss as its own - * observation and points the seat at the canonical spelling — visibility - * without redefining the vocabulary. + * Where those claims DO surface is H2 (an assigned card with no claim the + * marker reads) or H1 (an unassigned one) — visibility without redefining the + * vocabulary. * * @param {{ body?: string, created_at?: string }[]} commentRows * @returns {{ createdAt: string|null } | null} @@ -8257,257 +8251,6 @@ export function h33ClaimPredatesRuling(issue, commentRows) { ); } -// --------------------------------------------------------------------------- -// H34 — a claim-shaped comment with a NON-CANONICAL SEPARATOR (#12090). -// -// The card that filed this measured 24 of 54 open `pm:dispatched` cards -// carrying 「Claim — , session …」 with an EM DASH, invisible to -// `CLAIM_COMMENT_MARKER`, and proposed widening the class. That proposal is -// CLOSED, and this row is what was taken instead. -// -// ## Why the predicate is not widened — the ruling, verbatim -// -// `.claude/skills/pm-dispatch/SKILL.md` (step 4 of the claim protocol) records -// 「首行以字面 `Claim:` 开头是机器判据(维护者 2026-08-11 裁定;巡查谓词只认这 -// 一个拼写且保持严格,修法是全舰队向文档拼写收敛,⛔ 不放宽谓词)」. Under that -// ruling a dash-written claim is a MALFORMED claim, H2's row on it is a CORRECT -// report, and the repair is the write side converging on the documented -// spelling. Widening the reader would make the enforced protocol drift away -// from the written one permanently — the declared-≠-enforced shape this repo -// treats as a defect everywhere else — and would do it by editing a maintainer -// ruling's own text, which is not a patrol script's call. -// -// ## …and why the gap is nevertheless real, which is what this row closes -// -// The cost the card found is not only H2's loudness. `governingClaim` gates on -// the same marker before reading `Branch:`, so a card claimed with a dash buys -// no ref probe and can never produce an H20 「no remote ref」 or H27 「dead -// claim」 row: it sits OUTSIDE dispatch-liveness entirely, silently, for as long -// as it is in flight. H33 inherits the same blindness. So the malformed claim -// costs one noisy row and three quiet ones, and the quiet ones are the -// expensive direction. -// -// This row makes the malformed spelling VISIBLE without redefining what a claim -// is. `Claim:` stays the single machine criterion — H2/H20/H27/H33 are -// untouched, byte for byte — and the near miss is reported as its own, -// separately-named observation whose remedy points at the canonical spelling. -// The file therefore never disagrees with itself about its own vocabulary, -// which is the objection that kept the gap open. -// -// ## Measured, fresh, at dispatch time (2026-08-25 ~15:00Z) -// -// Over the LIVE open `pm:dispatched` population (45 cards): 40 colon-form, -// 1 dash-only, 0 hyphen, 0 fullwidth, 4 with no claim in any spelling. The -// filing card's 24-of-54 has already decayed to 1-of-45 in a day — the -// maintainer's prescribed convergence is working, which is precisely why the -// widening was not worth a ruling. Over the 128 `pm:dispatched` cards closed -// since 2026-08-24 the shape is recurring rather than historical: 13 dash-only -// claims, clustered in identifiable seat batches. So this row's expected steady -// state is quiet — H23's register, not H30's — and its value is that a seat -// writing the wrong spelling learns about it while the card is still in flight, -// which is how a fleet converges. -// -// ## Deliberately NOT the whole near-miss space -// -// Two suppressions, both under-reporting on purpose (H17's and H20's standing -// direction — a fabricated row sends a reader to check something that was never -// there): -// -// • A thread that ALSO carries a canonical claim is silent here. That card is -// machine-visible, so the row would have no remedy to offer. The residual -// it accepts: a RE-claim written with a dash over an older colon claim -// leaves `governingClaim` reading the stale one. Narrower than the shape -// this row is for, and naming it would cost a row on every card that ever -// wrote a dash claim once. -// • A claim-shaped line whose remainder carries none of the protocol's own -// content (a session reference, the word "seat", or a protocol-shaped -// branch anywhere in the comment) is silent. 「Claim - see above」 in prose -// is not evidence that a claim was attempted. -// -// ## The WORD position — measured 2026-08-27, and NOT widened -// -// A filed finding asked whether the separator class should cover a WORD where -// the punctuation goes, on the strength of 「Claim pointer: folded into the -// 11678 family dispatch」 — a string `latestClaimComment`'s header cited as a -// claim the marker reads, which it never was. Measured before answering, -// because the header above says this row's strictness was paid for once and -// a widening is exactly what it was paid to prevent. -// -// The census: 161 pm-tracked cards — 13 open `pm:dispatched`, 76 open -// `pm:queue`, 72 `pm:dispatched` closed since 2026-08-26 — and all 343 comments -// on them, classified by what follows a line-opening claim word. 19 canonical -// `Claim:`; 2 with a declared punctuation separator (both EM DASH, both on -// threads with no canonical claim — this row's live population, and it fires); -// and ZERO with a distinct word in the separator position. The 「Claim pointer:」 -// specimen occurs nowhere on the board: its only live instance is the finding -// card quoting this file quoting it. -// -// So the widening buys nothing measurable, and it would spend the conservative -// half's whole margin to buy it. `Claim` followed by a word is ordinary English -// — 「Claiming this card」, 「Claim comments are …」 — where `Claim` followed by -// an em dash is not; `looksLikeClaimContent` would carry that load for a -// population of zero. RECORDED, NOT WIDENED. -// -// The same census found three OTHER shapes invisible to both markers, recorded -// so a later reader can reopen the question on numbers instead of re-measuring -// — and each is already harmless for a DIFFERENT reason, which is the actual -// finding: -// -// • 2 「Claim (dev): …」 openings. Both threads ALSO carry a canonical claim, -// so the first suppression above already covers them: they are a SECOND -// claim comment on a machine-visible card, never a substitute for one. -// • 2 inflected openings (「Claiming this card. session …」) and 1 decorated -// with backticks (「`Claim:` devx@objectstack seat …」). All three sit on -// UNASSIGNED cards, which this row and H2 both decline to judge — the -// residual already declared above. -// -// Three different shapes, three different reasons, none of them this one. Each -// is its own question with its own noise floor if it ever acquires a population. -// --------------------------------------------------------------------------- - -/** - * The separators a claim-shaped line is measured to carry INSTEAD of the - * canonical colon, each with the name the row prints. Codepoints are spelled - * out rather than pasted: these characters are visually near-identical to each - * other and to the colon, and a row that says 「EM DASH (U+2014)」 tells a seat - * what to search for in a way a rendered glyph cannot. - * - * U+FF1A is here, and it is the reason it can be: the claim marker's class - * looked like it accepted it for the whole of this file's life and never did - * (see `CLAIM_COMMENT_MARKER`). In a protocol whose own SKILL.md is written in - * Chinese, 「Claim：」 is a plausible spelling; it was invisible to every reader - * and is now merely non-canonical, which is a state a seat can act on. - */ -export const NON_CANONICAL_CLAIM_SEPARATORS = [ - ['—', 'EM DASH (U+2014)'], - ['–', 'EN DASH (U+2013)'], - ['-', 'HYPHEN-MINUS (U+002D)'], - ['：', 'FULLWIDTH COLON (U+FF1A)'], -]; - -/** - * A claim-shaped opening whose separator is not the canonical colon. - * - * Mirrors `CLAIM_COMMENT_MARKER`'s strictness deliberately — same optional - * blockquote, same must-BEGIN-with-the-word anchor, same absence of `g` — so - * that the two markers partition claim-shaped openings rather than overlapping. - * The prose control the strictness was paid for (#7488) is still excluded here - * for the same reason it is there: 「the next seat should claim: only after the - * ruling lands」 does not BEGIN with the word. - * - * ⚠️ `[ \t]*`, NOT `\s*`, on both sides of the word. `\s` matches a NEWLINE, so - * `Claim\n- something` would put a markdown BULLET's hyphen in the separator - * position and read an ordinary list as a malformed claim — and a hyphen is the - * commonest line-opening character in this repo's comment bodies, where the - * colon marker never had to care (a line starting with `:` is not a thing - * anyone writes). - * - * Stated precisely, because it is easy to over-claim: `nearMissClaimSeparators` - * already runs this regex per SPLIT LINE, so the reader below is safe either - * way and the character class is not what saves it. What the class protects is - * this constant AS AN EXPORT — it is `m`-flagged and reusable, and a future - * caller doing `CLAIM_NEAR_MISS_MARKER.test(body)` over a whole comment gets - * the right answer only because of it. Both properties are pinned separately in - * the self-test, at the regex and at the reader, so neither can go green on the - * other's behalf. - * - * Capture groups: 1 = the separator, 2 = the rest of the line. - */ -export const CLAIM_NEAR_MISS_MARKER = new RegExp( - `^[ \\t]*>?[ \\t]*Claim(?:ed)?[ \\t]*([${NON_CANONICAL_CLAIM_SEPARATORS.map(([ch]) => `\\u${ch.codePointAt(0).toString(16).padStart(4, '0')}`).join('')}])[ \\t]*(.*)$`, - 'mi' -); - -/** - * Does this line's remainder carry the claim protocol's own content? - * - * The conservative half of the detection. The protocol's claim comment is - * required to name a session ID and a branch, and the measured dash claims do: - * 「Claim — skills seat `session_01RM…`. Folded dispatch …」 and 「Claim — - * domain:cli lane execution seat, session 019siH5jDmk5hrayvfyojUqR, round - * R36」. Requiring one of those tokens is what separates an attempted claim - * from a line that merely opens with the word. - * - * A protocol-shaped branch is accepted from ANYWHERE in the comment rather than - * from the opening line, because the template puts `Branch:` on its own line — - * the same reason `claimedBranches` reads a directive line rather than the - * first one. - */ -export function looksLikeClaimContent(remainder, body) { - if (/\bsessions?\b|\bseat\b/iu.test(String(remainder ?? ''))) return true; - return claimedBranches(body).length > 0; -} - -/** - * Which non-canonical separators this comment body opens a claim-shaped line - * with — de-duplicated, in the order declared, empty when none does. - * - * @param {string} body - * @returns {string[]} the printable separator NAMES - */ -export function nearMissClaimSeparators(body) { - const text = String(body ?? ''); - const found = []; - for (const line of text.split('\n')) { - const hit = CLAIM_NEAR_MISS_MARKER.exec(line); - if (!hit) continue; - if (!looksLikeClaimContent(hit[2], text)) continue; - const name = NON_CANONICAL_CLAIM_SEPARATORS.find(([ch]) => ch === hit[1])?.[1]; - if (name && !found.includes(name)) found.push(name); - } - return found; -} - -/** - * H34 — null when clean, else the finding sentence. - * - * Gated on the SAME population H2 judges (pm-tracked, assigned) and read off - * the SAME comment bodies that branch already fetched, so the row costs no - * request at all. It fires only where H2 fires — no canonical claim on the - * thread — which makes every H34 row a companion that EXPLAINS its card's H2 - * row rather than a second accusation about the same fact. - * - * The residual that gate accepts, stated rather than dropped: an UNASSIGNED - * `pm:dispatched` card buys no comment fetch here and so is never judged. That - * card is H1's finding in its own right, and its dispatch is already the thing - * being questioned. - * - * @param {object} issue — an OPEN issue. - * @param {(string|null|undefined)[]} commentBodies - */ -export function h34ClaimShapedNonCanonicalSeparator(issue, commentBodies) { - const labels = labelNames(issue ?? {}); - const pmTracked = labels.some((l) => l === 'pm:queue' || l === 'pm:dispatched'); - if (!pmTracked || (issue?.assignees ?? []).length === 0) return null; - if (!Array.isArray(commentBodies)) return null; - // A card with a readable claim is machine-visible; the row has no remedy for it. - if (commentBodies.some((b) => markerMatches(CLAIM_COMMENT_MARKER, String(b ?? '')))) return null; - - const separators = []; - let lines = 0; - for (const body of commentBodies) { - const names = nearMissClaimSeparators(body); - if (names.length === 0) continue; - lines += 1; - for (const name of names) if (!separators.includes(name)) separators.push(name); - } - if (separators.length === 0) return null; - - return ( - `a claim-shaped comment whose separator is NOT the canonical colon — ${lines} comment(s), ` + - `${separators.join(' + ')} where the protocol writes \`Claim:\` — and NO comment on this thread ` + - 'matches the marker. So this card reads as CLAIMLESS to H2 (its row on this card is correct, not a ' + - 'false positive) and sits outside dispatch-liveness entirely: `governingClaim` gates on the same ' + - 'marker before reading `Branch:`, so H20 can never report a missing remote ref for it and H27 can ' + - 'never report a dead claim — the two rows that exist to catch an abandoned dispatch. Remedy, and it ' + - 'is the WRITE side: the claiming seat re-posts (or edits) the claim so its FIRST line begins with ' + - 'the literal `Claim:`, per SKILL.md step 4 — 「首行以字面 `Claim:` 开头是机器判据(维护者 ' + - '2026-08-11 裁定;巡查谓词只认这一个拼写且保持严格,修法是全舰队向文档拼写收敛,⛔ 不放宽谓词)」. ' + - '⛔ The patrol predicate is NOT widened to accept the separator this card used; that is the one ' + - 'repair the ruling closes. Report-only: ⛔ never a label written from this script.' - ); -} - // --------------------------------------------------------------------------- // H36 — two open PRs holding the same changed file, one side already ACCEPTED // or armed (#12286). @@ -10339,8 +10082,8 @@ export function h43GovernedReviewRequestGap(pr, governed, approvers, reviewed = // above — so an author test cannot separate a seat's reading from the // maintainer's prose. It is not merely unavailable, it is ill-posed. // -// ⇒ So this row recognises the ARTEFACT, exactly as `CLAIM_COMMENT_MARKER`, H33, -// H34 and H37 already do, and filters by nothing else. That is also what the +// ⇒ So this row recognises the ARTEFACT, exactly as `CLAIM_COMMENT_MARKER`, H33 +// and H37 already do, and filters by nothing else. That is also what the // rule itself says: it is scoped to 「认领、派发令、复核、轮次报告、座位贴段落」 // — five artefacts — and says nothing whatever about who typed them. // @@ -11967,8 +11710,8 @@ export function h49PartialLandingUnreleased(issue, mergedPrs, openPrs, commentRo // the claim, and exactly the population whose thread the // dispatch-liveness loop already buys for H20/H27/H33. // OUT a closed card; a card with no `Claim:` this file's marker reads -// (H2's row where the card is assigned, H34's where the separator -// is the reason); a claim whose stamp does not parse — it cannot be +// (H2's row where the card is assigned, H1's where it is +// not); a claim whose stamp does not parse — it cannot be // placed against the cutoff, so the row DECLINES rather than // accuses; a predecessor whose id does not read — the "actual" half // of the sentence would be invented. @@ -15797,7 +15540,7 @@ export function h72OffTableMailboxLines(reading, nowMs = Date.now()) { // login (H43's header), and the seat-post sticker login does not match comment // authorship. ⚠️ The direction is what makes the two rows agree rather than // collide: H44 refused an author test for FINDING seat artefacts and that -// refusal stands — this row finds them exactly the structural way H44, H33, H34 +// refusal stands — this row finds them exactly the structural way H44, H33 // and H37 do, and since the re-keying it does not read the author as the defect // at all. // @@ -19574,7 +19317,6 @@ export const HALF_STATE_FAMILY_BAND = Object.freeze({ H24: 'state', H29: 'state', H30: 'state', - H34: 'state', H37: 'state', H44: 'state', H45: 'state', @@ -19592,10 +19334,8 @@ export const HALF_STATE_FAMILY_BAND = Object.freeze({ H49: 'state', H50: 'state', - // H60 is a `state` for H34's reason, one field over. H34 reads a claim whose - // SEPARATOR the marker does not accept; this one reads a claim whose BRANCH - // line the directive reader does not accept. Both are a live card that - // contradicts itself — claimed to a human, unclaimed (or claimed elsewhere) + // H60 is a `state`: it reads a claim whose BRANCH line the directive reader + // does not accept. That is a live card that contradicts itself — claimed to a human, unclaimed (or claimed elsewhere) // to the machine — and both are repaired on the board by the claiming seat // re-writing one comment. ⛔ Not `stall`: this row does not claim the card is // stopped, and it fires on a card whose dev may be working perfectly well; @@ -23429,13 +23169,6 @@ async function sweepInto(findings, seen, seenPrs, seenMerged, seenUnscoped, seen if (h2AssigneeNoClaimComment(issue, comments)) { findings.push([issue, 'H2', 'assignee set but no claim comment on the thread']); } - // H34 (#12090) — the same bodies, asked WHY H2 is firing. Free by - // construction: it reads the page already in hand, on exactly H2's - // population, and fires only on cards H2 is already reporting. Judged - // here rather than in the dispatched loop below so a `pm:queue` claim - // written with a dash is covered too, and so the two rows land together. - const nearMiss = h34ClaimShapedNonCanonicalSeparator(issue, comments); - if (nearMiss) findings.push([issue, 'H34', nearMiss]); } // H17 — the trigger-file index. Gathering only: the card's own body plus @@ -25339,8 +25072,8 @@ async function selfTest() { const h32row = (...args) => String(h32SeatIdleOverQueue(...args) ?? ''); const h38row = (...args) => String(h38SeatPostStale(...args) ?? ''); const h33row = (...args) => String(h33ClaimPredatesRuling(...args) ?? ''); - // H34's wrapper (the pattern this generalizes) stays beside its own block, - // as do H8's `halvesRow` and H27's `dead27Row` — all three wrap a helper + // H8's `halvesRow` and H27's `dead27Row` stay beside their own blocks — + // both wrap a helper // that is itself declared locally, next to the fixtures it closes over. // H29/H30 use a different reader, `says()`, declared at those blocks: // it returns a DESCRIBING string (`NO MESSAGE (null)`) rather than `''`, so @@ -27864,7 +27597,7 @@ async function selfTest() { // The governing claim — the MOST RECENT one, H19's double-check ① reasoning. const gov = (rows) => governingClaim(rows); // ⚠️ `?? []` rather than a bare `.branches` on the three-valued return, for - // the reason H34's row wrapper carries: a case built to go NULL under a + // this reason: a case built to go NULL under a // mutation must report a NAMED failing case, not throw and abort the suite // before the remaining cases run — measured while ablating the #16170 anchor, // where the reverse run died on a TypeError naming neither row nor mutation. @@ -28282,7 +28015,7 @@ async function selfTest() { t('⛔ H60: …and the claim marker still matches it, which is what makes this a STATE and not an absence', CLAIM_COMMENT_MARKER.test(INLINE_16322_NEW.body), true); t('⛔ H60: #16170\'s bulleted directive still parses, so its card stays clean', h60ClaimBranchUnparsed(dispatchedCard(), gov60([claimRow60(1, '2026-09-01T00:00:00Z', 'Claim: PM loop\n- Branch: `claude/issue-15511-zh-gap-helptext`')])), null); t('⛔ H60: …and the blockquoted template spelling too', h60ClaimBranchUnparsed(dispatchedCard(), gov60([claimRow60(1, '2026-09-01T00:00:00Z', 'Claim: PM loop\n> Branch: `claude/issue-6752-x`')])), null); - t('H60 band: the row is registered as a `state`, beside H34\'s claim-shape row', familyBand('H60'), familyBand('H34')); + t('H60 band: the row is registered as a `state`', familyBand('H60'), 'state'); // -- H62 (#17009): a decision card whose FACE carries no four-facet block --- // @@ -31564,92 +31297,11 @@ async function selfTest() { // H2/H20/H27 use — a file that disagrees with itself about what a claim IS // would be the worse defect. The consequence is UNDER-reporting, this file's // standing direction for an unrecognised spelling (H17/H20), never a - // fabricated row. Where the malformed claim becomes VISIBLE is H34 (#12090), - // whose cases sit below. + // fabricated row. Where the malformed claim becomes VISIBLE is H2's row on + // an assigned card (and H1's on an unassigned one). t('H33 claim: an EM-DASH claim is invisible to the shared marker (malformed, by ruling)', latestClaimComment([row33(CLAIM_AT, 'Claim — skills seat `session_x`.')]), null); t('H33: …so a card claimed that way under-reports rather than fabricating', h33ClaimPredatesRuling(dispatched33(), [row33(CLAIM_AT, 'Claim — skills seat `session_x`.'), rulingRow33]), null); - // -- H34 — a claim-shaped comment with a non-canonical separator (#12090) --- - // The acceptance pair for this row is the FOUR-WAY split below: the marker's - // verdict and H34's verdict are asserted on the same specimen every time, so - // no case can go green by quietly redefining what a claim is. - const tracked34 = (labels = ['pm:dispatched']) => issue(labels, ['os-zhuang']); - // ⚠️ `?? ''` rather than a bare `.includes` on the predicate's return. A row - // that goes NULL under a mutation must report a NAMED failing case, not throw - // and abort the suite before the remaining 200 cases run — measured while - // reverse-verifying this row: widening the marker turned a red suite into a - // TypeError whose message named neither the row nor the mutation. - const h34row = (...args) => String(h34ClaimShapedNonCanonicalSeparator(...args) ?? ''); - const DASH_CLAIM = 'Claim — skills seat `session_01RM`. Folded dispatch, patrol-predicates pack.'; - const COLON_CLAIM = 'Claim: skills seat `session_01RM`.\nBranch: `claude/issue-12090-x`'; - const PROSE_CONTROL = 'the next seat should claim: only after the ruling lands'; - const FULLWIDTH_CLAIM = 'Claim：skills seat `session_01RM`, round R36.'; - - // 1. dash claim -> the near-miss row fires AND the marker stays false. - t('H34: an EM-DASH claim fires the near-miss row', typeof h34ClaimShapedNonCanonicalSeparator(tracked34(), [DASH_CLAIM]), 'string'); - t('H34: …and CLAIM_COMMENT_MARKER is still false on it (⛔ 不放宽谓词)', CLAIM_COMMENT_MARKER.test(DASH_CLAIM), false); - t('H34: …and the row NAMES the separator by codepoint', h34row(tracked34(), [DASH_CLAIM]).includes('EM DASH (U+2014)'), true); - t('H34: …and points the remedy at the canonical spelling', h34row(tracked34(), [DASH_CLAIM]).includes('begins with the literal `Claim:`'), true); - t('H34: …and says the predicate is NOT widened', h34row(tracked34(), [DASH_CLAIM]).includes('NOT widened'), true); - // …and H2 still reports the card, correctly, as claimless. - t('H34: H2 still fires on the same card, and that row is CORRECT', h2AssigneeNoClaimComment(tracked34(), [DASH_CLAIM]), true); - - // 2. colon claim -> the marker is true and the near-miss row is silent. - t('H34: a COLON claim leaves the near-miss row silent', h34ClaimShapedNonCanonicalSeparator(tracked34(), [COLON_CLAIM]), null); - t('H34: …because the marker reads it', CLAIM_COMMENT_MARKER.test(COLON_CLAIM), true); - t('H34: …and H2 is clean', h2AssigneeNoClaimComment(tracked34(), [COLON_CLAIM]), false); - - // 3. the #7488 prose control -> NEITHER reader sees a claim. - t('H34: the prose control fires neither reader (H34)', h34ClaimShapedNonCanonicalSeparator(tracked34(), [PROSE_CONTROL]), null); - t('H34: …nor the marker', CLAIM_COMMENT_MARKER.test(PROSE_CONTROL), false); - - // 4. fullwidth colon -> a near miss, which is what the collapsed `[::]` class - // makes honest: U+FF1A never matched the marker and now says so out loud. - t('H34: a FULLWIDTH-COLON claim fires the near-miss row', typeof h34ClaimShapedNonCanonicalSeparator(tracked34(), [FULLWIDTH_CLAIM]), 'string'); - t('H34: …named as such', h34row(tracked34(), [FULLWIDTH_CLAIM]).includes('FULLWIDTH COLON (U+FF1A)'), true); - t('H34: …and the marker has never matched it (behaviour unchanged by #12090)', CLAIM_COMMENT_MARKER.test(FULLWIDTH_CLAIM), false); - - // The remaining separators, each pinned by name. - t('H34: an EN DASH is a near miss', nearMissClaimSeparators('Claim – skills seat, session 019x').join(','), 'EN DASH (U+2013)'); - t('H34: a HYPHEN-MINUS is a near miss', nearMissClaimSeparators('Claim - skills seat, session 019x').join(','), 'HYPHEN-MINUS (U+002D)'); - t('H34: `Claimed —` is one too', nearMissClaimSeparators('Claimed — by the skills seat, session 019x').join(','), 'EM DASH (U+2014)'); - t('H34: a BLOCKQUOTED near miss reads (the template is a blockquote)', nearMissClaimSeparators('> Claim — skills seat, session 019x').join(','), 'EM DASH (U+2014)'); - - // ⚠️ The `[ \t]*` decision, pinned TWICE and deliberately: once on the - // exported regex (where the character class is what answers) and once on the - // reader (where the per-line split answers). A single case would let the - // regex be relaxed to `\s*` with the suite still green — measured: the - // reader-level case alone survives that mutation untouched. - t('H34: the exported marker does not span lines (the class, not the split)', CLAIM_NEAR_MISS_MARKER.test('Claim\n- skills seat, session 019x'), false); - t('H34: a markdown bullet on the NEXT line is not a separator', nearMissClaimSeparators('Claim\n- skills seat, session 019x').length, 0); - t('H34: …the same shape with a colon claim is unaffected', CLAIM_COMMENT_MARKER.test('Claim\n- skills seat'), false); - - // The conservative content half: a claim-shaped opening with none of the - // protocol's own content is NOT evidence that a claim was attempted. - t('H34: a contentless claim-shaped line is silent', nearMissClaimSeparators('Claim - see above').length, 0); - t('H34: …but a `Branch:` line elsewhere in the comment is content', nearMissClaimSeparators('Claim - see above\nBranch: `claude/issue-12090-x`').join(','), 'HYPHEN-MINUS (U+002D)'); - - // Suppression: a thread that ALSO carries a readable claim is machine-visible, - // so the row has no remedy to offer and stays quiet. - t('H34: a thread carrying BOTH spellings is silent', h34ClaimShapedNonCanonicalSeparator(tracked34(), [DASH_CLAIM, COLON_CLAIM]), null); - // Gates, mirroring H2's exactly — same population, same declines. - t('H34: `pm:queue` is in scope too', typeof h34ClaimShapedNonCanonicalSeparator(tracked34(['pm:queue']), [DASH_CLAIM]), 'string'); - t('H34: an untracked card is out of scope', h34ClaimShapedNonCanonicalSeparator(issue(['domain:skills'], ['os-zhuang']), [DASH_CLAIM]), null); - t('H34: an UNASSIGNED card is out of scope (H1 owns that card)', h34ClaimShapedNonCanonicalSeparator(issue(['pm:dispatched']), [DASH_CLAIM]), null); - t('H34: an unconsulted thread declines', h34ClaimShapedNonCanonicalSeparator(tracked34(), undefined), null); - t('H34: an unreadable thread declines', h34ClaimShapedNonCanonicalSeparator(tracked34(), null), null); - t('H34: an empty thread is clean', h34ClaimShapedNonCanonicalSeparator(tracked34(), []), null); - t('H34: a null body among the comments does not throw', h34ClaimShapedNonCanonicalSeparator(tracked34(), [null, DASH_CLAIM]) !== null, true); - // Two malformed comments, two separators, counted and de-duplicated. - const twoNearMiss = h34row(tracked34(), [DASH_CLAIM, 'Claim – other seat, session 019y']); - t('H34: multiple malformed comments are counted', twoNearMiss.includes('2 comment(s)'), true); - t('H34: …and both separators named', twoNearMiss.includes('EM DASH (U+2014) + EN DASH (U+2013)'), true); - t('H34: a separator is named once, not per comment', h34row(tracked34(), [DASH_CLAIM, DASH_CLAIM]).includes('EM DASH (U+2014) + EM DASH'), false); - // No `g` flag on the shared marker — the state bug the colon marker's header - // names, asserted here because this regex is exported and reused per line. - t('H34: the near-miss marker carries no `g` flag', CLAIM_NEAR_MISS_MARKER.global, false); - t('H34: …so repeated reads of one line agree', nearMissClaimSeparators(DASH_CLAIM).join() === nearMissClaimSeparators(DASH_CLAIM).join(), true); - // -- H36 — cross-lane same-file holds (report-only patrol input, #12286) --- const h36pr = (number, opts = {}) => ({ number, @@ -33883,7 +33535,7 @@ Doubles as the fire's **write self-check** (step 0). \`201\` is not the reading. t('H50 order: the predecessor is the previous ROW, whatever its kind', h50([GRADING50, cm50(77, 'os-dev-report\n{}', '2026-09-07T11:30:00Z'), cm50(2, claim50('77'), T50_AFTER)]), null); t('H50 order: the documented blockquote claim spelling reads the same', h50([TRIAGE50, GRADING50, cm50(4, `> ${claim50('5569655227').split('\n').join('\n> ')}`, T50_AFTER)]), null); t('H50 order: a thread with no claim at all is silent (H2\'s row where assigned)', h50([TRIAGE50, GRADING50]), null); - t('H50 order: a dash-written claim is invisible here, as it is to H2 (H34 reports it)', h50([GRADING50, cm50(2, 'Claim — PM loop round 1\nThread-read: 1', T50_AFTER)]), null); + t('H50 order: a dash-written claim is invisible here, as it is to H2 (H2 reports the card claimless)', h50([GRADING50, cm50(2, 'Claim — PM loop round 1\nThread-read: 1', T50_AFTER)]), null); // Population — `pm:dispatched`, open. t('H50 population: an open `pm:dispatched` card is in', h50SpeaksAbout(DISPATCHED50), true); @@ -36187,7 +35839,7 @@ Doubles as the fire's **write self-check** (step 0). \`201\` is not the reading. b(BATTERY68, 'control: …and the blockquoted prose control too', markerMatches(CLAIM_COMMENT_MARKER, '> the next seat should claim: only after the ruling lands'), false); b(BATTERY68, 'control: ⛔ `Released:` is still a MALFORMED release, ⛔ not a dialect', markerMatches(RELEASE_COMMENT_MARKER, 'Released: session `session_x`'), false); b(BATTERY68, 'control: ⛔ the FULLWIDTH colon still does not read — the 2026-08-11 ruling is untouched', markerMatches(RELEASE_COMMENT_MARKER, 'Release：session `session_x`'), false); - b(BATTERY68, 'control: ⛔ nor a dash-written claim, which is H34\'s row and ⛔ not a widening', markerMatches(CLAIM_COMMENT_MARKER, 'Claim — skills seat, session 019x'), false); + b(BATTERY68, 'control: ⛔ nor a dash-written claim, which is H2\'s claimless card and ⛔ not a widening', markerMatches(CLAIM_COMMENT_MARKER, 'Claim — skills seat, session 019x'), false); // THE LIST-ITEM GUARD — `undecorateProseLine` strips EVERY `*`, so a `*` // bullet would become a directive. H20 pins `- Claim:` as NOT a claim; a `*` @@ -36253,7 +35905,7 @@ Doubles as the fire's **write self-check** (step 0). \`201\` is not the reading. b(BATTERY68, '#18831 ordering: the two additions are LAST on the roster, in that order', OWNERSHIP_MARKER_NEAR_MISS_FORMS.slice(-2).map((f) => f.id).join(','), 'heading-bare,bare-word'); b(BATTERY68, '#18831 ordering: a heading WITH the canonical colon is still `heading`, ⛔ not `heading-bare`', miss68('## Claim: seat, session `session_x`').form, 'heading'); b(BATTERY68, '#18831 ordering: …an inflected word with a colon is still `inflected-word`', miss68('Claiming: seat, session `session_x`').form, 'inflected-word'); - b(BATTERY68, '#18831 ordering: …and a dash-written claim is still `separator`, which is H34\'s row and ⛔ not this addition\'s', miss68('Claim — skills seat, session 019x').form, 'separator'); + b(BATTERY68, '#18831 ordering: …and a dash-written claim is still `separator`, which is H2\'s claimless card and ⛔ not this addition\'s', miss68('Claim — skills seat, session 019x').form, 'separator'); b(BATTERY68, '#18831 ordering: …a `-` bulleted claim with no colon is named by NEITHER addition — a list is not a heading and not a line start', ownershipMarkerNearMisses(rows68('- Claiming the producer half, seat `domain:spec`')).length, 0); // THE `leading-sigil` FORM (#18829 A) — the shape the sibling's retired