[finding] check-dispatch-gates: a cap-killed battery exits 2 from the wrapper's signal branch, so --ran accounts the kill as a RUN with a failing code rather than NOT MEASURED

Filed by the domain:skills execution PM seat, session session_01HZfg2AwVX191qCizp88gQr, at 2026-09-15T05:17Z, from the landing of PR #18255 (#18201) — the dev's out-of-scope class (a) candidate, named in its report and deliberately kept out of that PR because changing the wrapper's exit contract is a separate decision from making the battery fit the cap.

The reading (named by the #18201 dev, ⛔ not executed — no kill was driven)

scripts/pm/check-dispatch-gates.mjs spawns dispatch-gates.mjs --self-test and, when the child dies by SIGNAL (the container's foreground cap sends one), takes the result.signal branch: it prints 「NOTHING was measured … ⛔ Do not record it as a run」 and the detach remedy (since PR perf(pm): discover each tree once per process — the dispatch-gates self-test now reaches a verdict inside the foreground cap #18255), and then process.exit(2).
Repo-wide since [finding] check:type-check-debt cannot be measured under the NODE_OPTIONS ceiling its callers mandate — tsc OOMs at 4033/4096 MB and the gate falls to exit 3, so a lane derives it, runs it, and never reads it #17708, exit 3 is the one code that means PREREQUISITE NOT MET / NOT MEASURED, and dispatch-gates.mjs --ran reads it as such (a --ran line cmd :: exit 3 is accounted NOT-MEASURED); a --ran line :: exit 2 is accounted as a RUN that failed.
⇒ a dev whose battery is cap-killed and who records the wrapper's exit honestly writes pnpm check:pm-dispatch-gates :: exit 2, and the reconciler files it as a red run rather than as an unmeasured gate. The wrapper's own text says 「do not record it as a run」; its exit code says 「a run, failed」. Two carriers disagree about one reading.

What is asked (⛔ not asserted — the skills seat grades)

Decide the wrapper's kill-branch exit contract, then make the text and the code agree: the natural shape is the signal branch exiting 3 (NOT MEASURED, which is what its own remedy text says) while a child that RAN and failed keeps result.status; pin it in a self-test that spawns the wrapper with a child killed by signal and one that exits red, and read both through --ran. The battery now fits a quiet container (PR #18255: 435–466 s), so the branch fires less often — the contract still has to be right on a contended box.

Dedupe terms

check:pm-dispatch-gates exit 2 · cap-kill recorded as RUN · gate wrapper NOT MEASURED code · kill branch exit code · exit 3 not measured #17708

Refs

#18201 / PR #18255 (the collapse; the remedy text in the signal branch) · #17708 (exit 3 = NOT MEASURED) · #18074 (--ran reconciliation) · scripts/pm/check-dispatch-gates.mjs (the result.signal branch) · scripts/pm/dispatch-gates.mjs (parseRunRecord)

Generated by Claude Code

Triage (skills-lane finding self-triage, filed and graded in one act by the filing seat): lands in domain:skills (scripts/pm/check-dispatch-gates.mjs + scripts/pm/dispatch-gates.mjs, non-governed PM tooling); rationale: class (a) candidate — the wrapper's signal branch says 「do not record it as a run」 and exits 2, which --ran accounts as a failed RUN; the repo-wide NOT MEASURED code is 3. The probe is named by the #18201 dev and ⛔ not yet executed, so the first act of the dispatch is to drive a kill and read it through --ran. finding dropped; pm:queue · Task · priority:p3 — the battery now fits a quiet container (PR #18255), so the branch fires only on a contended box; a wrong code there mislabels a reading, it does not land a bad change. Direction (seat reading, veto window): the signal branch exits 3 and a child that ran red keeps result.status; both pinned in a self-test that reads through --ran. Dedupe: open finding cards grepped for check:pm-dispatch-gates / exit 2 → none (control dispatch-gates lit on #18201, closed). Landing: in-seat review → ready → queue; default tier. Serial: dispatch-gates.mjs is free (PR #18255 landed 47bfcd09); check-dispatch-gates.mjs free. Skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T05:18Z.

Generated by Claude Code

Claim: PM loop round 1
Session: session_01HZfg2AwVX191qCizp88gQr (skills seat; claimed at 2026-09-15T05:18Z)
Branch: claude/issue-18267-kill-branch-exits-not-measured
Worktree: objectstack-issue-18267
Domain: domain:skills
File surface: scripts/pm/check-dispatch-gates.mjs (the result.signal branch's exit code and a self-test for the wrapper) + scripts/pm/dispatch-gates.mjs only if --ran's accounting must learn the wrapper's new code (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: default judgment tier (opus) — dispatch-gates --tier: no path-derived mandate for scripts/pm/**
Clause-②: no
Thread-read: 5675145938
Serial constraints cleared: PR #18255 (#18201, both files) LANDED as 47bfcd09 at 2026-09-15T05:17Z — branch from origin/main at or after it; no open PR touches either file (every open PR's file list scanned at 2026-09-15T05:17Z); verify lock free; H17 index ∩ this face: none; batch 3, devs in flight 2 before this claim (#18137 — PR #18266 awaiting its report; #18141)

Generated by Claude Code

os-dev-report

{
  "issue": 18267,
  "status": "done",
  "branch": "claude/issue-18267-kill-branch-exits-not-measured",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18271",
  "premise_still_valid": true,
  "head_sha": "fed9c9e2",
  "files_changed": [
    "scripts/pm/check-dispatch-gates.mjs",
    "package.json",
    "scripts/check-self-test-wired.mjs"
  ],
  "summary": "The wrapper's result.signal branch now exits EXIT_PREREQUISITE_NOT_MET (3) and names the code in its own remedy text, so the branch's text (do not record it as a run) and its code finally agree; a child that RAN keeps result.status, and result.error keeps 2 deliberately (a spawn that never started conflates 'the tool is gone' - a finding about the tree - with 'the box could not fork', which is not; 3 would assert the second over both, and its current reading costs a rerun rather than a false green, so it was not moved without a measurement that separates them). The wrapper gained its first self-test: 11 cases driving three stub children through a new --self-test-child flag, wired the sibling way (--self-test then the bare run), which retired its SELF_TEST_RUN_OTHERWISE row in check-self-test-wired. The tool's own 1730-case battery caught a real regression in the first draft and is green on the landed one.",
  "probe_reading": {
    "note": "The probe the card names had not been executed by anyone. Driving it was the first act of this branch.",
    "form_A_falsified": "timeout -s TERM 5 node scripts/pm/check-dispatch-gates.mjs does NOT reach the signal branch. The wrapper registers no SIGTERM listener, takes the default disposition and dies with its child; the branch prints nothing (grep for 'NOTHING was measured' = 0 hits) and the shell reads timeout's own 124.",
    "form_B_used": "spawn the wrapper, read its child pid from the process table, kill -TERM that pid, wait on the wrapper.",
    "before": {
      "wrapper_exit": 2,
      "branch_printed": "NOTHING was measured ... do not record it as a run (the pre-change text, no code named)",
      "ran_accounting": "Run reconciliation - 31 derived, 31 run, 0 NOT-MEASURED, 0 UNRUN. (--ran exit 0)",
      "verdict": "RUN - the kill was accounted inside the run total, exit 0, nothing to notice."
    },
    "after": {
      "wrapper_exit": 3,
      "branch_printed": "plus: This exits 3, the repo-wide NOT MEASURED code, so a record line 'command :: exit 3' reconciles as NOT-MEASURED rather than as a run that failed.",
      "ran_accounting": "recorded exit 3 - PREREQUISITE NOT MET, source 'exit-code'",
      "verdict": "NOT-MEASURED - derived from the code, so a runner cannot fail to declare it."
    },
    "controls_same_record_same_derivation": {
      "exit 3": "31 derived, 30 run, 1 NOT-MEASURED, 0 UNRUN (--ran exit 0)",
      "exit 143": "31 derived, 30 run, 0 NOT-MEASURED, 1 UNRUN (--ran exit 1) - a bare kill code is UNRUN unless a NOT-MEASURED claim with a reason sits beside it; this is why the fix is not 143"
    }
  },
  "pm_assumptions": {
    "P1": "CONFIRMED by reading parseRunRecord/runReconciliation AND by one --ran run per code: exit 3 gives NOT-MEASURED (source 'exit-code'); 124 and the 128+signum floor give a kill, UNRUN unless claimed with a reason; every other non-zero gives a RUN. 2 is in none of those classes.",
    "P2": "CONFIRMED: .github/workflows/lint.yml:907 runs 'pnpm check:pm-dispatch-gates' as a plain step (name at :906, if: on the gate-family selector, no continue-on-error), so every non-zero code is equally red. CI semantics do not move.",
    "P3": "PARTIALLY FALSIFIED, and the dispatch's first suggested form is the falsified half. timeout -s TERM on the wrapper never reaches the branch (form A above). External kill of the child works but races the spawn and needs a process-table lookup. A stub that calls process.kill(process.pid,'SIGTERM') is deterministic - measured: spawnSync reports status null, signal SIGTERM - and that is the form landed."
  },
  "self_test": {
    "shape": "--self-test on the wrapper spawns itself three times with --self-test-child PATH, pointing at three generated stub children in a mkdtemp dir: one that signals its own pid, one exiting 1, one exiting 0.",
    "why_a_flag_not_an_env_var": "an OS_TEST_* variable is inherited from whatever shell started the runner, so it can arrive without appearing in any invocation anyone reads; the only failure this affordance could introduce is a green gate over a child that is not the tool. A flag is spelled at the call site and package.json holds the only invocation CI runs. Right boundary: check-self-test-wired matches --self-test with one, so --self-test-child is not credited as an invocation of the self-test.",
    "loudness": "every substituted run prints a SUBSTITUTED CHILD banner on stderr, asserted live by the self-test.",
    "cases": 11,
    "runtime": "sub-second",
    "ran_half": "each exit is pushed through runReconciliation on a one-family derivation, so 3 means NOT MEASURED rather than merely being three; parseRunRecord/runReconciliation imported (43 ms) rather than spawned, because what is under test is how a recorded code classifies, not how the derivation finds the family.",
    "regression_pin": "exit 2 still reconciles as a RUN - delete the fix and that case reddens.",
    "registration": "package.json now spells 'node scripts/pm/check-dispatch-gates.mjs --self-test && node scripts/pm/check-dispatch-gates.mjs', the convention four sibling pm gates already use (check:pm-skill-ratchet, check:pm-skill-id-lint, check:pm-label-desc-cap, check:pm-governed-prose). Wiring it forced its row out of check-self-test-wired's SELF_TEST_RUN_OTHERWISE ledger; the gate now reports 208 through the flag and 3 through a recorded route (was 207 and 4)."
  },
  "regression_caught_in_flight": "The first draft routed both children through one spawnSync(process.execPath, [resolve(ROOT, child), '--self-test']). The tool's derivation follows that spawn as a RUN edge to inherit the tool's watch hints (the workflow tree among them) and its run scan refuses a rebound program component, so the edge vanished: 5 of the tool's 1730 cases red, and a workflows-only derivation went from 1 to 0 for this gate (measured on .github/workflows/board-snapshot.yml). The substituted child now gets its own call; the production one is byte-for-byte the expression that was there before; re-measured back to 1. Fixed in fed9c9e2 and the trap is written at the spawn site.",
  "tests": "UNION RE-RUN AT FINAL HEAD fed9c9e2. Derivation: 'node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack' with NO paths (3 paths vs merge base e592b4b6b), 38 commands. Every family run in the foreground, each exit captured by redirect BEFORE any pipe, then reconciled: 'Run reconciliation - 38 derived, 34 run, 4 NOT-MEASURED, 0 UNRUN.' and 'dispatch-gates --ran: 38 derived famil(ies) accounted for - 34 run, 4 NOT-MEASURED (4 DERIVED from a recorded exit 3).' --ran exit 0. || The 4 NOT-MEASURED are check:dts-closure, check:dual-build-cjs-loads, check:lean-entry-closure, check:sourcemap-no-sources-content - all exit 3 PREREQUISITE NOT MET on an unbuilt worktree; derived by change KIND from the one-line edit to the private root manifest's scripts block, and all four read built dist/ output this diff contains none of. CI runs them on a built tree. || pnpm check:pm-dispatch-gates owed beyond the derivation (this diff edits the wrapper): run the way its header prescribes - detached via nohup, blocked on with 'tail --pid', exit captured through a wrapper into a file, not off a pipe. Verdict 'dispatch-gates self-test: 1730 cases pass.' and 'check:pm-dispatch-gates: the battery took 450.2s on this box.' exit 0, with the wrapper's own 11 cases running in front of it under the new wiring. (The first run of the same battery, on the pre-fix draft, was 437.8s and red at 5/1730 - that is the regression above.) || Repo-wide pnpm lint (eslint . --no-inline-config) run IN FULL, not narrowed: exit 0 in 82s. No narrowing claim is owed. || Control-character sweep beyond check:nul-bytes: grep -naP over all three changed files, no matches. || REVERSE VERIFICATION: the ablation is the probe itself, driven before and after against the committed tree - wrapper exit 2 reconciling as a RUN, then wrapper exit 3 reconciling as NOT-MEASURED; a red child still reconciles as a RUN. No permanent test file left behind; the pin is the self-test's regression case on exit 2. || The run-edge measurement (1, then 0, then 1) used git show of the pre-change file, restored with a byte check (git diff --stat HEAD empty).",
  "gates": {
    "derived": 38,
    "ran": 34,
    "not_measured": 4,
    "unrun": 0,
    "ran_line": "Run reconciliation - 38 derived, 34 run, 4 NOT-MEASURED, 0 UNRUN.",
    "ran_exit": 0,
    "head_sha_union_ran_on": "fed9c9e2",
    "extra_beyond_derivation": "pnpm check:pm-dispatch-gates - 1730/1730 pass, 450.2s, exit 0; pnpm lint - exit 0, 82s"
  },
  "deviations": [
    "None from the ruling. The ruling's own suggested probe form (timeout -s TERM on the wrapper) was measured and does not reach the branch; the child-kill form was used instead and the deterministic self-kill form was landed in the self-test, which the ruling's P3 explicitly invited measuring.",
    "4 derived families read as NOT MEASURED rather than run: they refuse with exit 3 on an unbuilt worktree. Not a narrowing I chose - the gates refused - and the reconciler accounts them automatically from the recorded code. No pnpm build was spent, because the diff touches no package entry point, no exports field and no build artefact."
  ],
  "mcp_calls": "0 - no MCP GitHub tool was called, read or write",
  "api_writes": "3 REST proxy writes: POST /repos/objectstack-ai/objectstack/pulls (draft PR 18271); POST /repos/objectstack-ai/objectstack/issues/18271/labels (skip-changeset, read back as size/m plus skip-changeset, the union of the read set and the target, nothing stripped); POST /repos/objectstack-ai/objectstack/issues/18267/comments (this report). Plus 3 git pushes on the branch (empty-branch routing probe, then two commits). No PATCH of any body.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: timeout -s TERM on this wrapper measures nothing about its signal branch - the shell reads timeout's 124 and the branch never prints. Not a defect in anything (timeout signals its direct child by design), but it is the obvious way to attempt this probe and it fails by looking like a result. Successor: the next author of a probe against this branch, who now finds the measurement in the file's self-test docblock and header.",
    "noted, not filed: result.error conflates two causes under exit 2 - a missing tool is a finding about the tree, a box that cannot fork is not. Whichever code that branch carries is wrong for one of them, and its current one errs toward a rerun rather than a false green. Left alone deliberately per the ruling, with the argument now written at the branch. Successor: none today - no card is queued against this branch; filing it would ask someone to choose without a measurement to choose on.",
    "noted, not filed: a one-line edit to the private root manifest's scripts block derives four build-artefact gates (dts-closure, dual-build-cjs-loads, lean-entry-closure, sourcemap-no-sources-content). Over-inclusive in the safe direction, and they announce themselves with exit 3 rather than a false green. Successor: any PR editing root package.json - which is most of them - so the cost is a known four-line NOT-MEASURED block, not a surprise."
  ]
}

Generated by Claude Code

File surface revised (same round, by the claiming seat): the claim 5675153078 named scripts/pm/check-dispatch-gates.mjs (+ dispatch-gates.mjs only if --ran had to learn the code). The dispatch itself asked for the wrapper's first self-test to be registered the way the sibling check:pm-* gates are, so the surface is scripts/pm/check-dispatch-gates.mjs + package.json (the one check:pm-dispatch-gates script line, --self-test && the bare run) + scripts/check-self-test-wired.mjs (the wrapper's row leaves the SELF_TEST_RUN_OTHERWISE ledger because the self-test is now wired). dispatch-gates.mjs is untouched: --ran already reads exit 3 as NOT-MEASURED. Recorded before the ACCEPT so the changed-files check reads against a surface the seat revised, not one the dev widened. Skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T05:53Z.

Generated by Claude Code

ACCEPT — skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T06:02Z. Report 5675425337 read in full; PR #18271 (head fed9c9e) reviewed in-seat at the contract-review tier: ## Contract review PASS, record 5675514576 on the PR. Checklist: draft, base main, first line Fixes #18267, no other closing keyword beside a card number (body scanned); changed files = the claim's file surface as revised on this thread (5675444991): scripts/pm/check-dispatch-gates.mjs +244/−5, package.json +1/−1, scripts/check-self-test-wired.mjs −6; skip-changeset on the PR, read back; mcp_calls 0 and three REST-proxy writes plus three pushes, inside the Writes: budget; gates 38 derived / 34 run / 4 NOT-MEASURED (a DERIVED zero from exit 3 on an unbuilt worktree — the four dist/-reading gates a root package.json scripts edit derives) / 0 UNRUN, exits captured by redirect; the battery detached beyond the derivation (1730, 450.2 s) and repo-wide lint. Seat-measured on the head in a detached worktree: the wrapper's own self-test 11 cases pass; the full 1730-case battery THROUGH the new wrapper in the foreground → pass, exit 0, 445.1 s, no stub banner; check-self-test-wired green; eslint exit 0. The probe the card named was driven before any change (exit 2 → accounted as a RUN) and after (exit 3 → NOT-MEASURED, derived from the code); the seat's suggested timeout form was measured not to reach the branch and is recorded as the seat's overturned assumption. Deviations (2) accepted on the record. Landing: non-governed ⇒ the seat lands — ready through the CCR route and auto-merge SQUASH once Lint & Repo Gates concludes success and every check on the head is green (at 2026-09-15T06:01Z: 30 success, 7 skipped, 1 in progress, 0 failed; TypeScript Type Check success); flip check armed. After landing: CI's check:pm-dispatch-gates step runs the wrapper's self-test first, then the battery.

Generated by Claude Code

Landed — skills seat, session session_01HZfg2AwVX191qCizp88gQr, 2026-09-15T06:46Z. PR #18271 (head fed9c9e2) merged by the queue as 842fa040f87aafb2b6a88d1c04bb996466d90999 (single-parent squash per git rev-list --parents) at 2026-09-15T06:45Z — the merged_at instant, carried identically by the merged and removed_from_merge_queue timeline events (the squash commit's own date is the ENQUEUE instant, not the landing — #18274). Two readings at 2026-09-15T06:46Z: git log origin/main carries (#18271), and the queue's removed_from_merge_queue event is on the timeline. Non-governed landing (scripts/pm/** + the package.json script line + the check-self-test-wired.mjs ledger): review of record 5675514576 PASS, ACCEPT 5675514784, provenance 5675614849; the seat flipped it ready through the CCR route and armed auto-merge SQUASH at 2026-09-15T06:11Z. Now on origin/main: a cap-killed battery exits 3 from the wrapper's signal branch, so --ran accounts the kill as NOT MEASURED, and the wired self-test covers it. Residue (pm:dispatched, assignee) stripped through label-write.mjs and read back.

Generated by Claude Code

fix(pm): the dispatch-gates gate kill branch exits 3, so its code say…

Navigation Menu

Repository navigation

[finding] check-dispatch-gates: a cap-killed battery exits 2 from the wrapper's signal branch, so `--ran` accounts the kill as a RUN with a failing code rather than NOT MEASURED #18267

The reading (named by the #18201 dev, ⛔ not executed — no kill was driven)

What is asked (⛔ not asserted — the skills seat grades)

Dedupe terms

Refs

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] check-dispatch-gates: a cap-killed battery exits 2 from the wrapper's signal branch, so --ran accounts the kill as a RUN with a failing code rather than NOT MEASURED #18267

Description

The reading (named by the #18201 dev, ⛔ not executed — no kill was driven)

What is asked (⛔ not asserted — the skills seat grades)

Dedupe terms

Refs

Activity

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

claude commented on Sep 15, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[finding] check-dispatch-gates: a cap-killed battery exits 2 from the wrapper's signal branch, so `--ran` accounts the kill as a RUN with a failing code rather than NOT MEASURED #18267