fix(pm): H26 claims no SCHEDULED or DECLARED release, not "no mechanism" — counted, not asserted

[os-steve]

Fixes #18017
Clause-②: no

H26 told every reader that a pm:blocked card whose Blocked-by: target carries pm:on-hold or
needs-user-decision has "NO MECHANISM THAT WILL EVER RELEASE IT", on the grounds that those are
"by definition states a card sits in WHILE OPEN". The grounds were an assertion. Counting refutes it.
The row is miscalibrated, not wrong, so the population is untouched — every card it fired on before
it fires on now — and only what it CLAIMS has changed: it now claims what it actually checked.

The count, and its control

Re-run of the filing card's own instrument, on this board, objectstack-ai/objectstack, taken
2026-09-20T19:06Z. REST /issues?labels=LABEL&state=STATE&per_page=100 walked to exhaustion, the
population being every ISSUE carrying the label, pull requests excluded.

label	open issues	closed issues	closed state_reason
pm:on-hold	101	14	7 completed, 7 not_planned
needs-user-decision	8	15	9 completed, 6 not_planned
pm:blocked (control)	69	21	17 completed, 4 not_planned

The control has a known direction and it hit: all three CLOSED legs came back non-zero, so a
state=closed leg that silently answered nothing would have read as three zeros, and it did not.
⇒ cards do close out of both indefinite states, 16 of those 29 closures completed — the hold
lifted and the work finished, which is exactly the release the retired wording said could not exist.

⛔ Excluding pull requests is not tidying: 96 of the 111 rows the closed needs-user-decision leg
returns are PRs. A census that skipped that filter would be measuring a different population, so the
filter is stated in the rationale block beside the numbers rather than left implicit.

Two readings agree on direction. The lead commit's 2026-09-18 reading (carried on this branch, see
below) reproduces row for row on pm:on-hold — 101 open / 14 closed, 7 completed + 7 not_planned —
while needs-user-decision moved 2/5 → 8/15 over the two days between the readings.

Before and after

BEFORE (origin/main a0e62e6, h26BlockOnIndefiniteTarget):

pm:blocked on N target(s) that can never CLOSE: … The unlock predicate is "the Blocked-by:
target closed", and pm:on-hold / needs-user-decision are by definition states a card sits in
WHILE OPEN — so this block has NO MECHANISM THAT WILL EVER RELEASE IT. …

AFTER:

pm:blocked on N target(s) with no scheduled or declared release: … The unlock predicate is
"the Blocked-by: target closed"; no timer, sweep or queue in this loop closes pm:on-hold or
needs-user-decision, and this card declares no label-transition exit for them — so this block has
NO SCHEDULED OR DECLARED RELEASE. ⛔ Deliberately weaker than the categorical claim this row used
to make, which counting refutes: on this board 2026-09-20, pm:on-hold closed 14 and
needs-user-decision closed 15, 16 of the 29 completed (control: pm:blocked closed 21, so the
closed enumerator fires) — these targets DO close. …

⭐ H26's final sentence — the half the filing card calls the useful one — survives verbatim, and is
now pinned as such: "the release has to come from the target's own state changing (a ruling answered,
a hold restarted) and someone has to want that."

The same retirement is carried through every H26-owned site in this file, including the one a
line-oriented grep cannot see: the resolution-map rationale in sweepInto, whose "can never / close"
is split across two lines. Sites moved: the header docblock's H26 row, the region banner and rationale,
the INDEFINITE_TARGET_LABELS jsdoc, the emitted string, the sweepInto note, and the self-test.

The pin this card flips, on purpose

origin/main scripts/pm/check-half-states.mjs:30799 pinned the retired sentence POSITIVELY:

t('H26: …and the row says the block has no releasing mechanism',
  h26row(...).includes('NO MECHANISM THAT WILL EVER RELEASE IT'), true);

That case is replaced, not deleted, and three more that named the same phrase (bothLegs,
bothOnOne, the chain-leg negative) move to the new one. What stands in its place is a POSITIVE pin on
the replacement plus a retirement pin per retired phrase — because an absence-only check would stay
green if a later edit deleted the assertion along with the sentence it named, which is the exact
failure mode this card exists to describe.

⚠️ One correction to the lead commit, and it is the same class of defect. The lead pinned the
retired opening as .includes('can never close') — lower case. The row never wrote it that way; it
wrote can never CLOSE. Measured against origin/main's own module, before any change:

includes('can never CLOSE')  = true    ⇐ the spelling the row really used
includes('can never close')  = false   ⇐ the lead's pin: already false on main, so it pinned nothing

A negative pin that was already satisfied before the change is a phantom check: it evaluates, it passes,
and deleting it leaves every gate just as green. This PR pins can never CLOSE.

Ablations — both legs, on disk, restored

Run through scripts/ablation-replace.mjs, which proves the mutation reached the file (anchor count and
git hash-object blob) and proves the restore against HEAD rather than against an exit code.

Run on head 2e6ee9e (file blob c49d47a7484d), in an isolated worktree so no other measurement
shared a tree with the mutation.

ablation	mutation	result
the replacement phrase is really asserted	has NO SCHEDULED OR DECLARED RELEASE. ' + → has no release anyone has scheduled. ' +	blob c49d47a7484d → d50b37d13bb1; 3 cases RED, self-test exit 1
the counted census clause is really asserted	closed 15, 16 of the 29 …closed 21 → the same sentence without the counts	blob c49d47a7484d → 7052be9a1f12; 2 cases RED, self-test exit 1

Both restored: blob == HEAD (c49d47a7484d) and git diff HEAD empty, git status --porcelain empty
after each leg. ⛔ No ablation artefact is left in the tree; neither leg is a permanent test.

What was carried forward from the lead commit, and what was dropped

The branch already existed on origin at 9180b90 — the domain:devx seat's aborted dispatch of
2026-09-18, one commit, +92 / −30, no PR, base 86b9000. It is merged, never re-derived and never
force-pushed: 3746268 is the merge commit, and its history is intact.

Merge conflict, measured rather than assumed. Exactly one textual conflict, and it is a boundary
rather than a rewrite: #19255 inserted the label-transition exit block (UNLOCK_LABEL_EXIT_RE,
unlockLabelExits, h19DeclaredExitFired) directly above H26's banner comment, so both sides edited
that one banner line. Everything else merged textually with no conflict, and the merge reproduces the
lead's delta exactly (+92 / −30 against origin/main).

Kept from the lead: the direction itself (claims no SCHEDULED release, not "no mechanism" — counted,
not asserted); the census-with-control shape in the rationale block; the retirement carried to every
H26-owned site including the two-line-split one; the requirement that the self-test pin the new wording
POSITIVELY; H26's useful final sentence preserved verbatim.

Dropped or rewritten, because #19255 landed in between:

• The lead's "no timer, no sweep and no queue in this loop closes either" is no longer true as a flat
  statement — since [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 the sweep can fire a declared label-transition exit on exactly these
  targets. The row now says what it tests: nothing schedules the close and this card declared no
  exit. NO SCHEDULED RELEASE became NO SCHEDULED OR DECLARED RELEASE, and the four pins naming the
  old phrase moved with it.
• The lead's remedy direction is superseded: [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 already added the ⭐ The exit: paragraph naming
  the declarable exit (Unlock-action: re-check #N when label needs-user-decision absent). That
  paragraph and the exitKeys stand-down are left byte-for-byte untouched.
• The lead's 2026-09-18 numbers are replaced by a reading taken for this PR (above), because the row
  now carries them permanently and they should be a reading this branch took.
• The lead's can never close pin is corrected to can never CLOSE (see above).
• Trimmed to the net-line budget: +94 / −35, net +59 against origin/main at a0e62e6 (the lead's
  own delta was +92 / −30, net +62).

Verification

• node scripts/pm/check-half-states.mjs --self-test → exit 0, 5125 cases (5118 at origin/main
a0e62e6, measured on a throwaway worktree at that commit; +7 here).
• SELF_TEST_BATTERY_FLOOR unchanged at 8, roster unchanged, every battery's registered count
  unchanged (H66 183/172, H19 37/34, H65 46/42, H67 152/142, H2/H47/H66 213/184, ISSUE_BODY_LIMIT 55/52, H19/H26 11/10, H68 26/24). The new cases are plain t(...) cases beside H26's existing ones,
  ⛔ not at selfTest()'s tail and ⛔ not a new battery.
• H26 cases: 26 → 32.
• #19255 regression case added: a card whose body carries
  Unlock-action: re-check #987 when label pm:on-hold absent still stands the row down for that
  target after the rewrite.
• Tree-wide grep before rewording (assumption 2, verified): on origin/main, NO MECHANISM has 5
  occurrences and can never CLOSE 2, and every one of them is inside this single file. Nothing
  under .claude/, scripts/ elsewhere, or .github/ pins either phrase, so no other file has to move
  in this PR.
• The derived gate union (node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack,
  41 commands, change set derived by the tool itself) is run with each exit captured before any pipe;
  results are in the os-dev-report comment on finding(pm-gate): H26's "can never CLOSE" premise is asserted, not counted — counting refutes it, and objectui's ported copy cannot be fixed without this one moving first #18017.

skip-changeset: the diff publishes nothing. scripts/pm/** is on the no-publish fast path and appears
in no released package's files[].

Acceptance notes

• Cost, disclosed: the rendered row grows from 1370 to 1726 characters for a single indefinite
  target. ISSUE_BODY_LIMIT is 262144 bytes and the renderer trims low-band rows first, so this is a
  real if small charge against the anchor body — paid because the sentence is quoted outward and the
  count has to travel with the quote. The full census lives in the rationale block, not in the row.
• Not edited, deliberately: the remedy paragraph [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 added quotes a maintainer ruling in Chinese,
  「无机制可唤醒的卡 ⛔ 不 hold」. That is the maintainer's wording about when not to hold a card, not this
  row's structural verdict about a target, and AGENTS.md forbids rewriting a quoted ruling — so it stays
  verbatim.
• Not edited, out of surface: objectui's byte-pinned port of this file follows through
  scripts/upstream-port-pin.json on the consumer card (objectui#9317 / objectui PR The datasource-admin HTTP family is reachable unauthenticated, read and write — mounts outside the enforceAuth seam #9391); ⛔ no
  divergence is declared here and the pin is not touched.
• Observation, not filed: the phantom-pin shape corrected above (a negative .includes(...) that was
  already false before the change) is a self-test hazard with no mechanical guard — a case that cannot
  fail reads identically to one that passes. Noted rather than filed because no such pin exists on
  origin/main; the one instance lived only on an unmerged branch and is fixed here. Successor: the next
  PR touching this suite's negative pins.

---

Generated by Claude Code

[os-steve]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: e0e0be006e41e021117d35f92d846209247093d5

① Derived judgments

• Published accept set / public surface: none moves — one report-only PM patrol reader (scripts/pm/check-half-states.mjs, +94 / −35, net +59 against the ≤ +60 budget); no published package, no schema, no label written, zero new requests; the POPULATION of H26 is unchanged (every card it fired on before, it fires on now) and SELF_TEST_BATTERY_FLOOR stays 8 with every battery count unchanged (H26's plain cases 26 → 32). check-governed-merges --pr 19426: NOT governed — ordinary queue landing. check-widening-tells --declaration no: 1 file NOT MEASURED — the no stands on the seat's reading. Judged correct.
• The fix, read line by line: H26's categorical claim (「target(s) that can never CLOSE … NO MECHANISM THAT WILL EVER RELEASE IT」) is retired at all six sites the card named — the header docblock's H26 row, the region banner and rationale, the INDEFINITE_TARGET_LABELS jsdoc, the emitted string, the sweepInto note and the self-test — and replaced by what the predicate actually tests: 「with no scheduled or declared release … no timer, sweep or queue in this loop closes pm:on-hold or needs-user-decision, and this card declares no label-transition exit for them — so this block has NO SCHEDULED OR DECLARED RELEASE」, followed by the counted census that refutes the old claim (this board, 2026-09-20: pm:on-hold closed 14, needs-user-decision closed 15, 16 of 29 completed; control pm:blocked closed 21, so the closed enumerator fires) and the useful half preserved verbatim (「the release has to come from the target's own state changing … and someone has to want that」). [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255's stand-down for a declared label-transition exit and its remedy paragraph are byte-for-byte untouched, with a new case pinning that a pm:on-hold target under a declared Unlock-action: exit still stands down. The four-axis choice (C over the lead's B: after [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 the sweep CAN fire a declared exit on these targets, so 「no scheduled release」 alone under-describes) is sound. Judged correct.
• The devx seat's released lead 9180b90 was merged forward (merge commit, ⛔ no force-push, no re-derivation): exactly one textual conflict — the region banner line [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255 had edited from above — resolved by keeping [finding] a pm:blocked card whose Blocked-by target is a needs-user-decision card can never be released — the unlock predicate is "closed", and Unlock-action recognizes exactly one PR-shaped spelling #19255's block whole; the lead's 2026-09-18 census replaced by a reading this branch took (reproducing the lead row for row on pm:on-hold); the lead's negative pin .includes('can never close') (lower case — a phantom, already false on unmodified main, measured) corrected to the spelling the row wrote. Judged correct.
• Evidence, seat-run on this head in a detached worktree: check-half-states.mjs --self-test exit 0 — 5131 cases (+7 over the base, the invariant across both merges); the only remaining occurrences of the retired phrases in the file are the negative pins asserting their absence; merge-tree vs origin/main 488f4f5: 0 conflicts, and vs the head of this seat's queued PR fix(scripts/pm): H59 reads state_reason so a deliberately REOPENED card is not called a platform failure #19427 (H59 region): 0 conflicts; CI on e0e0be0: 33 check runs, all complete, 0 red. The dev's proof the pins can fail: two on-disk ablations on the final head (the replacement phrase → 3 red; the census clause → 2 red), blob-hash-proven both ways; the 41-family derived union 41 / 41 exit 0 with --ran reconciled (a DERIVED zero), identical on all three derivations.

② Semver level

None — nothing published; skip-changeset on the PR is the correct declaration. Judged correct.

③ Boundary flags

open_questions empty. Declared file-surface extension accepted: two further H26-owned sites in the same file (the header docblock's H26 row :586–:614 and the sweepInto note) carried the retired premise and moved in the same PR — assumption 2's own rule. Mechanism assumptions: 1 refined (one conflict hunk, the banner), 2 verified and counted (5 + 2 occurrences, all in this file), 3 verified with a new pin, 4 holds with no edit. The PR body quotes the first ablation run's hashes (pre-merge head) and is written once; the final-head hashes are in the report — recorded, not a defect. Disclosed cost: the rendered row grows 1370 → 1726 chars per indefinite target on the anchor body (ISSUE_BODY_LIMIT 262,144 B). Two noted, not filed items with carriers (the phantom-negative-pin shape has no mechanical guard; #19255's quoted maintainer ruling 「无机制可唤醒的卡 ⛔ 不 hold」 stays verbatim by AGENTS.md's rule — it is about when not to HOLD, not this row's verdict).

Implemented-by: claude/issue-18017-h26-counted-not-asserted
Reviewed-by: session_017ETYWqMQD4qMtZzAGovWNi

VERDICT: PASS

---

Generated by Claude Code