fix(pm): the claim reader reads one declared act — Release: through the shared reading, the prose channel retired, cross-author live claims named (row C9)

[os-tesla]

Fixes #18862
Fixes #18829
Part of #18773 — the READER half only. The governed-text half of ruling A (AGENTS.md's release clause, SKILL.md's yielding line at its line ceiling, the core-rules.md twin) remains open on that card: Tier H, held by the seat, no governed file is touched here.

Clause-②: no

What this PR does

One skills round, one PR on scripts/pm/check-clause2-carriers.mjs, as the three director records of the maintainer's 「同意」 order (batches #154 / #156). The claim arbiter used to infer three ownership acts the protocol never declared — a supersession between seats, a prose retraction, a stripped prefix — and this PR makes it read the declared act only.

1. Cross-author live claims with no Release: between — row C9 ([finding] the #18828 triage's p1 escalation condition is MET — 12 open cards across both boards carry LIVE Claim: comments from two or more DIFFERENT authors with no retraction between them, and every reader prints the newer one green as a supersession; a cross-session ownership transfer has no reader #18862, ruling b, 5725370464). claimHandovers reads the live, attributable claims (membership decided by claimRetractions, exactly as C8 and governance read it) and names every point where the author changes. A hand-over whose taking claim is dated strictly after CROSS_AUTHOR_CLAIM_ROW_EFFECTIVE_AT = '2026-09-19T03:45Z' (the frozen UTC minute the row was authored — it names the rule's landing window; the seat reconciles pairs between it and the merge by hand) is a C9 row, exit 4, with the ruling's remedy sentence: the holder posts Release:; the taker posts nothing until then; a taker that yields posts its own Release: with 去向 「让先到者」. A hand-over dated at or before the instant is a C9-BEFORE-EFFECTIVE note and a claim.handover record field — listed, informational, never the exit. C8 is unchanged in behaviour; its direction-(b) pin (silent on cross-author) is replaced by the pin that C9 is the reader that speaks — never both for two authors with one claim each.
2. One undecoration definition ([finding] check-clause2-carriers keeps two undecoration paths that disagree on 5 of 12 spellings — the retraction line strips a leading sigil / list marker / heading / underscore and the shared claim reading refuses them — both directions carry a live specimen, so converging is a protocol decision, not a cleanup #18829, ruling A, 5725677876). The retraction channel reads Release: through the sibling's markerMatches, never through a stripper of its own; the second stripper is deleted from the tree. The sigil-led [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 shape is a named near-miss row, leading-sigil, in OWNERSHIP_MARKER_NEAR_MISS_FORMS (scripts/pm/check-half-states.mjs, one roster member and its BATTERY68 pins — no predicate, no H-row, no other export, per the claim amendment 5738916776). The twelve spellings the card compared now read the same on every side of the reader (5 disagreements → 0, pinned on the pool and on the retraction index); - Claim: / * Claim: stay not a claim (H20).
3. The prose retraction channel retires ([finding] the protocol names Release: as the act that takes a card out of a seat's hands, but a claim RETRACTED before work starts is written as prose — the arbiter now reads that prose (PR #18770), and whether the protocol should REQUIRE the Release: spelling for a retraction is a governed-text question nobody has ruled #18773, ruling A, 5725677697 — reader half). The verb roster, the comment-id scan and the act-verb reading PR fix(pm): a retracted claim leaves the governing-claim pool, so a withdrawn claim never governs #18770 taught are deleted, not disabled. A retraction is read only as a Release: line by the claim's own author, later than the claim; it needs no comment id (it names its target by authorship) — that requirement did not exist before and is not added. The [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 specimen is replayed both ways: the prose line no longer retracts (the withdrawn claim stands again and governs — the cost ruling A priced and accepted, ⛔ C), the same line in the declared act takes it out, and the thread is listed by C9's note rather than read silently.

Measured, not assumed

• A live consequence of item 2, replayed from the REST rows (read 2026-09-19T03:30Z through the proxy): four of the twelve pairs carried the earlier holder's Release: all along, bolded or backticked — finding(spec): zod z.record() SILENTLY DROPS a __proto__ key from its parse OUTPUT while reporting success — ObjectSchema accepts the document and hands back a different one #17852 (os-warren, 5700605769), objectui#7804 (os-justin, 5707789209 and 5710740327), objectui#7848 (claude[bot], 5617804323), objectui#7924 (os-warren, 5613413938) — and the raw RELEASE_COMMENT_MARKER.test the retraction channel used could not see any of them. With the shared reading, objectui#7848 and feat(driver-memory)!: refuse to boot into a multi-tenant deployment (#6915) #7924 clear by this change alone (one author left, nothing named, no seat posts anything); on finding(spec): zod z.record() SILENTLY DROPS a __proto__ key from its parse OUTPUT while reporting success — ObjectSchema accepts the document and hands back a different one #17852 the pair named today is os-litant → os-elon-musk (2026-09-18T22:04Z), not the os-warren / os-litant pair the sweep counted; on objectui#7804 what stands is os-tesla ×4 (C8) and os-sam.
• The ten remaining measured pairs are all listed today (none judged — every taking claim predates the instant), and every one re-dated sixty days forward is judged: the row is reachable on the real shapes, not only on synthetic ones. objectui#9370's 「⛔ NOT a re-claim」 line is named anyway — the reader reads order, not intent.
• [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 (the [finding] the governing-claim selector keeps a RETRACTED claim in the pool — a withdrawn claim outranks the live one, names a branch that does not exist, and the real claimant reads as SUPERSEDED #18719 specimen) becomes a listed cross-author pair once the prose channel retires (os-litant → os-bill, 2026-09-17), cleared by os-bill's own Release: — added to the seat's reconciliation list below.
• Offline acceptance runs (--pair-json, head 4cb58bc): a synthetic pair with the taker dated after the instant → exit 4, ✗ C9 — with the remedy sentence; the real spec: the evaluated-slot rule of #15430 reaches only the flow-node ledger — every other ExpressionInputSchema slot an engine evaluates (formula expression, validation / hook / sharing condition, visibleWhen…) still accepts an ast-only or blank-source envelope #15811 rows → exit 0, ℹ️ C9-BEFORE-EFFECTIVE — and claim.handover: … LISTED; two claims by one author → exit 4, ✗ C8 — only, claim.handover: none.

Verification

• node scripts/pm/check-clause2-carriers.mjs --self-test: 978 → 1046 cases pass, exit 0 (head 4cb58bc). Battery roster 33 → 34 (#18862: cross-author LIVE claims…, floor 52); #18719 battery rewritten around the one channel (floor 36 → 46 by count); #18764 two-paths pins replaced by the twelve-spelling replay; #18828 direction (b) replaced.
• node scripts/pm/check-half-states.mjs --self-test: 5043 → 5063 cases pass, exit 0; H2/H47/H66 decorated ownership marker 193/184 → 213/184; roster pin …separator,leading-sigil,heading-bare,bare-word.
• Reverse verification, both files, through scripts/ablation-replace.mjs (mutation proven on disk, restored to the HEAD blob, git diff HEAD empty after each): judged: dated === 'after', → judged: false, in clause2 reds 21 of 1046 cases (every judged-direction pin); id: 'leading-sigil', → id: 'leading-sigil-ablated', in the sibling reds 6 of 5063 (the exact-roster pin and the form's own pins).
• Absence, tree-wide at 4cb58bc: git grep -n undecorateRetractionLine → 0 (control undecorateProseLine → 20); git grep -n RETRACTION_PROSE_ANCHORS → 0 (control RELEASE_COMMENT_MARKER → 64). No control bytes in either file (grep -naP over the C0/DEL range → 0).
• Lint, narrowed and measured (head 4cb58bc): npx eslint --no-inline-config --format json over the two changed files → exit 0, 2 files, 0 errors, 0 warnings. Population: eslint.config.mjs' **/*.{ts,tsx,mts,cts,js,jsx,mjs,cjs} object minus NEVER_LINTED, so both .mjs files are in it; the count is the JSON output's length (2). Invariance: the config states it 「never enables type-aware linting (no parserOptions.project, no typed @typescript-eslint rules) for ANY file」, so this diff moves no untouched file's verdict — the narrowing excludes nothing.
• Derived gate families (node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands from the merge-base: 42 commands — eight more than the dispatch's list, pulled in by the check-half-states.mjs touch — plus check:pm-governed-merges, named by the dispatch): the sweep was in progress when this PR was opened; every exit code, with the --ran reconciliation, is in the os-dev-report comment on [finding] the #18828 triage's p1 escalation condition is MET — 12 open cards across both boards carry LIVE Claim: comments from two or more DIFFERENT authors with no retraction between them, and every reader prints the newer one green as a supersession; a cross-session ownership transfer has no reader #18862.

Acceptance notes

• For the seat's reconciliation sweep (ruling item 2, the seat's act): objectui#7848 and feat(driver-memory)!: refuse to boot into a multi-tenant deployment (#6915) #7924 need no Release: — they clear on landing; finding(spec): zod z.record() SILENTLY DROPS a __proto__ key from its parse OUTPUT while reporting success — ObjectSchema accepts the document and hands back a different one #17852's live pair is now os-litant / os-elon-musk; objectui#7804 is os-tesla / os-sam beside os-tesla's C8; [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 (closed) joins the informational population as os-litant / os-bill.
• Deviation from the dispatch's mechanism assumption, reported rather than forced: the dispatch expected the [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 prose line itself (「🚨 **撤回上一条认领…」) to be reported by the near-miss row. It cannot be without reading the verb 撤回 — the channel ruling A on [finding] the protocol names Release: as the act that takes a card out of a seat's hands, but a claim RETRACTED before work starts is written as prose — the arbiter now reads that prose (PR #18770), and whether the protocol should REQUIRE the Release: spelling for a retraction is a governed-text question nobody has ruled #18773 retires (⛔ B). So the named row is the sigil-led spelling of the declared act (the same line with Release: in place of the verb, sigil and bold intact — pinned as leading-sigil in both files), the prose specimen is pinned as named by no form with the reason, and its thread is loud through C9's note instead. The ruling's intent — the shape is named, not stripped; the thread is not silent — holds on both counts.
• The live --pair N path: node's built-in fetch does not read HTTPS_PROXY; the script already prints the --use-env-proxy hint. Noted, not filed — not a defect in this reader. 承接者:无.
• --pair-json documents that omit the PR's own comment thread read UNJUDGED (exit 2) on the record read — by design; the acceptance runs above supply it.

---

Generated by Claude Code

[os-tesla]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 4cb58bcb01161e5eb13c9b49b9c2e3e2e1a173ec

In-seat review by the domain:skills seat at the contract-review tier (served tier read at seating, marker 5738863635) — the record this lane owes on every round it delivers, Clause-②: no included; one record for the fold's three cards (#18862 anchor · #18829 whole · #18773 reader half), which the pair reading names as C6 on all three. Verified against GitHub and the fetched branch (claude/issue-18862-claim-reader-one-spelling at 4cb58bc, merge-base 5d0ee8f), ⛔ not against the report: two files, scripts/pm/check-clause2-carriers.mjs and scripts/pm/check-half-states.mjs, +803 / −250, four commits, every trailer pair model-free. Every hunk in the sibling file is the one roster member leading-sigil, its docblock and its BATTERY68 pins — the surface the claim amendment 5738916776 allowed (no predicate, no H-row, no other export). Every hunk in the arbiter lies in the retraction section, the new C9 section, pairRows / pairNotes / the input record, and the self-test. check-governed-merges.mjs --pr 19169: NOT governed, 1053 ≤ 5000. No governed file is on the branch (git diff origin/main...4cb58bc --stat names the two scripts/pm/** files only).

① Derived judgments

• The retraction channel NARROWS to one. retractionChannel(row) reads Release: through the imported markerMatches; RETRACTION_PROSE_ANCHORS, COMMENT_ID_SCAN and undecorateRetractionLine are deleted, not disabled (tree-wide at the head, seat's own greps: undecorateRetractionLine 0 against control undecorateProseLine 20; RETRACTION_PROSE_ANCHORS 0 against control RELEASE_COMMENT_MARKER 64; on origin/main the two read 5 and 4). Correct per [finding] the protocol names Release: as the act that takes a card out of a seat's hands, but a claim RETRACTED before work starts is written as prose — the arbiter now reads that prose (PR #18770), and whether the protocol should REQUIRE the Release: spelling for a retraction is a governed-text question nobody has ruled #18773 A (one declared spelling, ⛔ B) and [finding] check-clause2-carriers keeps two undecoration paths that disagree on 5 of 12 spellings — the retraction line strips a leading sigil / list marker / heading / underscore and the shared claim reading refuses them — both directions carry a live specimen, so converging is a protocol decision, not a cleanup #18829 A (the shared reading is the definition). The one widening inside it — a decorated **Release:** or backticked release now retracts — is the ruling's own text, and the raw-constant control is pinned beside it. No comment-id requirement is added; none existed.
• A new READING beside C8, not a second selector. claimHandovers walks the live, attributable claims (claimRetractions is the membership authority, as for C8 and governance) and names every author change; c9CrossAuthorLiveClaims is a row at exit 4 only when a taking claim is dated strictly after the frozen CROSS_AUTHOR_CLAIM_ROW_EFFECTIVE_AT = '2026-09-19T03:45Z'; c9HandoverNote lists everything at or before it (and an unreadable stamp) as C9-BEFORE-EFFECTIVE, never the exit; claim.handover joins INPUT_RECORD_PAIR_FIELDS. CLAIM_SELECTION_RULE and claimCarrierSelection's return shape are untouched; C8's direction-(b) pin still asserts C8 silent on the cross-author shape and the new pin asserts C9 is the reader that speaks. Correct per [finding] the #18828 triage's p1 escalation condition is MET — 12 open cards across both boards carry LIVE Claim: comments from two or more DIFFERENT authors with no retraction between them, and every reader prints the newer one green as a supersession; a cross-session ownership transfer has no reader #18862 b: own row, own remedy sentence (holder posts Release:; taker posts nothing until then; a yielding taker posts its own with 去向 「让先到者」), exit 4, ⛔ not a widening of C8, history listed never red; the instant is a constant, ⛔ not derived from git.
• The sibling's vocabulary gains leading-sigil between separator and the two [finding] four open cards carry a real claim in a spelling the identity record's own contract does not permit — a ## Claim heading, Claiming the producer half, ## Claim + dispatch — so every marker-reading tool reads them as UNCLAIMED; the writers drifted off AGENTS.md :394–:395, not the reader #18831 additions; the exact-roster pin moves with it; markerMatches itself is untouched, so H20's - Claim: / * Claim: refusals hold and are pinned on both sides. Correct per [finding] check-clause2-carriers keeps two undecoration paths that disagree on 5 of 12 spellings — the retraction line strips a leading sigil / list marker / heading / underscore and the shared claim reading refuses them — both directions carry a live specimen, so converging is a protocol decision, not a cleanup #18829 A.
• Fixtures: the prose-channel fixtures are deleted and re-pinned in the ruled direction with the ruling cited (a line opening with 撤回 and naming the id retracts nothing; retract / withdraw are prose); the [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 specimen is replayed both ways with the sigil-only and bold-only controls; the twelve measured pairs are replayed from REST rows (read 2026-09-19T03:30Z) — none judged today, every one judged when re-dated sixty days forward; the two live decorated releases (finding(spec): zod z.record() SILENTLY DROPS a __proto__ key from its parse OUTPUT while reporting success — ObjectSchema accepts the document and hands back a different one #17852, objectui#7848) are pinned as read. Battery roster 33 → 34 with its floor; self-tests 978 → 1046 and 5043 → 5063 per the report; ablations 21 / 1046 and 6 / 5063 driven red and restored byte-identical.
• The dev's mechanism correction is accepted: the [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 PROSE line cannot be named by a near-miss form without reading the verb — the channel [finding] the protocol names Release: as the act that takes a card out of a seat's hands, but a claim RETRACTED before work starts is written as prose — the arbiter now reads that prose (PR #18770), and whether the protocol should REQUIRE the Release: spelling for a retraction is a governed-text question nobody has ruled #18773 A retires (⛔ B). The named row is the sigil-led spelling of the declared act, the prose specimen is pinned as named by no form with its reason, and its thread is loud through C9's note. The ruling's two intents — named not stripped; not silent — both hold, and its words 「the loud, declared direction keeps reading it」 name the declared direction.

② Semver level

None owed — scripts/pm/** ships in no package; skip-changeset applied and read back (labels on the head: size/xl, skip-changeset).

③ Boundary flags

• Deviations (4) read: the mechanism correction above (accepted); 42 derived gate commands plus check:pm-governed-merges instead of the dispatch's 34 (the sibling touch pulls in eight more — all exit 0); 26 GETs for real-byte fixtures (the file's convention); same-day facts moved (spec: the evaluated-slot rule of #15430 reaches only the flow-node ledger — every other ExpressionInputSchema slot an engine evaluates (formula expression, validation / hook / sharing condition, visibleWhen…) still accepts an ast-only or blank-source envelope #15811 and [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373 closed; finding(spec): zod z.record() SILENTLY DROPS a __proto__ key from its parse OUTPUT while reporting success — ObjectSchema accepts the document and hands back a different one #17852's third claimant) and are carried into the fixtures with their read time.
• out_of_scope_findings (3, 「noted, not filed」): the reconciliation list — reconciled against the seat's own sweep of 03:31Z, which posted on all twelve cards (objectui#7848 / feat(driver-memory)!: refuse to boot into a multi-tenant deployment (#6915) #7924 clear on landing by the one-reading change; finding(spec): zod z.record() SILENTLY DROPS a __proto__ key from its parse OUTPUT while reporting success — ObjectSchema accepts the document and hands back a different one #17852's live pair is os-litant → os-elon-musk at 2026-09-18T22:04Z, before the instant ⇒ listed; objectui#7804 is os-tesla ×4 beside os-sam; [finding] check-type-source-resolution's (via …) provenance annotation is FALSE for an include entry naming a bare directory — and that annotation is the doc-block's OWN test for which remedy limb an author is on #18373, closed, joins the informational population); the --use-env-proxy hint is node's flag and the file prints it; a --pair-json document without the PR thread reads UNJUDGED by design. None is one of the three classes; not filed.
• open_questions: none. Gates: 43 run, 43 exit 0 per the report's table with --ran reconciled. CI on the head at this record's writing: 26 success + 11 rostered skips, Lint & Repo Gates and TypeScript Type Check both success on 4cb58bcb01.
• The governed-text half of [finding] the protocol names Release: as the act that takes a card out of a seat's hands, but a claim RETRACTED before work starts is written as prose — the arbiter now reads that prose (PR #18770), and whether the protocol should REQUIRE the Release: spelling for a retraction is a governed-text question nobody has ruled #18773 (AGENTS.md's release clause, SKILL.md's yielding line, the core-rules.md twin) stays on that card as Tier H; the PR's Part of #18773 line carries no closing keyword (seat read the first lines and scanned the whole body).

Implemented-by: claude/issue-18862-claim-reader-one-spelling
Reviewed-by: session_01W5y9kRg1YtYaMQYExVLRc2

VERDICT: PASS

---

Generated by Claude Code