fix(pm): derive check-half-states' H32 seat specimens from the resolved sweep repo

[os-litant]

Fixes #12994

seatLane() judges a seat title's at-repo suffix against the live resolved sweep repo — at[2] !== SWEEP_REPO.repo.split('/')[1] — never a constant. Six H32 self-test specimens spelled the board names literally, so four of them asserted this board rather than the property, and invert wholesale in any install that is not objectstack. That is the one thing a file adopted BY COPY (the whole design of resolveSweepRepo) may not do.

What changed

Both board names are now derived, in the H32 block of the self-test:

const OWN_BOARD = SWEEP_REPO.repo.split('/')[1];
const SIBLING_BOARD = OWN_BOARD === 'objectui' ? 'objectstack' : 'objectui';
const ownLaneSeat = (status = '🟢 os-x') => seat(`[PM seat] domain:devx @ ${OWN_BOARD} — ${status}`);
const siblingLaneSeat = (status = '🟢 os-x') => seat(`[PM seat] domain:devx @ ${SIBLING_BOARD} — ${status}`);

OWN_BOARD is the resolved repo's name half — the same half the predicate compares, re-derived here rather than shared through a helper, so a predicate that switched to the OWNER half still goes red. SIBLING_BOARD is a real neighbouring board picked by a ternary that makes the two provably distinct on every board, which is what stops the FOREIGN rows from going vacuously green on an install named after the specimen.

Specimen table

self-test row	before	after	verdict on a foreign board, before
H32 lane: an `@ sibling` suffix is FOREIGN	@ objectui	siblingLaneSeat()	inverted — failed
H32 lane: an `@ own-repo` suffix is NOT foreign	@ objectstack	ownLaneSeat()	inverted — failed
H32: a FOREIGN lane is out of scope	@ objectui	siblingLaneSeat()	inverted — failed
H32 gate: a foreign-lane seat buys no fetch	@ objectui	siblingLaneSeat()	inverted — failed
H32 lane: …and keeps the bare lane label	@ objectstack	ownLaneSeat()	passed, but measured the foreign path under an own-board name
H32 held: ⏳ vacant is NOT held	@ objectui	siblingLaneSeat('⏳ vacant')	passed; spelling only (seatIsHeld reads no repo)

The last two never failed, so they are not part of the measured four; they are rebuilt because the same grep catches them and because on the objectui board their titles read as the opposite of what the surrounding prose calls them.

The two-board proof

The resolver reads PM_SWEEP_REPO first, then GITHUB_REPOSITORY, at module load — so the suite runs against a chosen board with no code change. --self-test is deliberately exempt from the malformed-repo exit-2 refusal, so every leg below really ran.

leg	before this PR	after
own board, no env	✓ 1551 cases pass. (exit 0)	✓ 1551 cases pass. (exit 0)
PM_SWEEP_REPO=objectstack-ai/objectui	✗ 4 of 1551 case(s) failed. (exit 1)	✓ 1551 cases pass. (exit 0)
PM_SWEEP_REPO=objectstack-ai/cloud	not measured	✓ 1551 cases pass. (exit 0)

4 failures on the foreign board, 0 after. The four were exactly the rows named above. Suite count is unchanged at 1551 before and 1551 after — same rows, board-independent spelling. The third leg is a board neither literal anticipates, which is the actual claim being made.

Ablation — the rebuilt rows can still fail

A derived specimen could be green everywhere because it stopped asserting anything. Inverting the predicate's comparison (!== to ===) on the committed tree, mutation confirmed on disk by anchor counts (removed=1 injected=0 before, removed=0 injected=1 after) and by a blob hash differing from the HEAD blob:

ABLATED, own board                       ✗ 4 of 1551 case(s) failed.
ABLATED, PM_SWEEP_REPO=.../objectui      ✗ 4 of 1551 case(s) failed.

Both legs name the same four rows. Restoration was proved byte-identical to the HEAD blob (9ffcdb76…), with git diff HEAD and git status --porcelain both empty and the restored tree re-run green at 1551. There is no build or dist leg to rebuild here: this file is executed directly by node as source, with no package exports resolution.

Gates

Derived union from dispatch-gates.mjs --repo objectstack-ai/objectstack (change set taken from the merge base by the tool itself), run under os-verify-lock.sh with slot issue-12994-h32; exit codes captured before any pipe, verdict lines quoted from each gate's own output.

gate	exit	verdict
check:pm-half-states	0	✓ check-half-states self-test: 1551 cases pass.
check:pm-dispatch-gates	0	✓ dispatch-gates self-test: 834 cases pass.
bare-root-worklist.mjs --self-test	0	OK self-test: 47 live row(s) … none stale, none missing, none contradicted.
check:partof-closing-keyword	0	✓ check-partof-closing-keyword self-test: 28 cases pass.
check:entry-guard	0	✓ 175 scripts/ file(s) — every entry guard goes through invoked-as.mjs
check:cross-package-test-inputs	0	OK: 23 package(s) read outside themselves, all declared
check-closing-keyword-parity.mjs	0	OK (3 parsers agree on all 9 keywords …)
check-ci-filter-parity.mjs	0	OK: all 117 declared cross-package glob(s) … covered
check:agent-test-spelling, check:bash32-floor, check:cli-command-ids, check:parse-guard, check:pnpm-filter-targets, check:watch-hint-literal, check-cross-package-test-inputs.mjs	0	all green
check-governed-merges.mjs --self-test	0	✓ 206 assertions
ci-failure.mjs --self-test	0	green (it imports this file's transport classifier)

The two convention-triggered obligations are included because this diff edits a gate script; the last two are the suites of scripts that read this file, which no path derivation names.

One entry needs its reading stated: bare node scripts/check-partof-closing-keyword.mjs with no PR_BODY exits 2 as a usage refusal, not a gate failure — its CI form takes the PR body. Re-run with this very body supplied, it is green.

Scope and observations

• No predicate changed; no edit outside the self-test specimens. Diff is +24/-6 in one file, entirely inside the H32 block of selfTest().
• Observed, not widened: the two sweep repo: rows that pin the objectstack leg against DEFAULT_SWEEP_REPO are coupled to that constant, not to the resolved repo — they were green in every leg above, including the foreign-board one. Different divergence class; untouched here, as the card directs.
• Observed, unchanged: with a malformed PM_SWEEP_REPO (a bare name), the own-board row fails — identically before and after this PR. The property has no witness in that state: the name half is undefined and no title suffix can equal it. Parity, not a regression, and not something this card's shape can fix without changing the case count.
• No changeset: scripts/pm/** publishes nothing user-visible. skip-changeset applied at PR-open.
• Once this lands, objectui's parity pin still carries three name-swap-only divergence entries (h32-lane-own-board, h32-foreign-out-of-scope, h32-foreign-no-fetch). Deleting them is that repo's standard --resync act on the parity gate — owed there after merge, deliberately not a rider on this PR.

Draft on purpose: the PM self-reviews at collection and lands it.

Generated by Claude Code

---

Generated by Claude Code