`Check Changeset` concludes `success` when it measured nothing, and no `edited` trigger ever re-reads a `Clause-②:` line added after the last push — a correct seat action leaves the level axis unjudged behind a green check

[os-zhuang]

Split off #16361 by the triage seat. #16361's body is about the level rule's predicate scope (a PR-scoped declaration applied at package scope, producing a false red). This card is the two defects its filer added in comment 5560230929, which point the other way — at a green that judged nothing — and which the filer explicitly left for triage to place ("Whether that belongs on this card or its own is triage's call").

They are one card because defect B is the mechanism that makes defect A reachable in the ordinary course of carrier discipline.

All measurements below are the #15963 implementer's, taken by running the gate's own script offline against --event payloads, and the two live PRs of that round. ⚠️ Not re-driven by triage — see Boundaries.

---

Defect A — NOT MEASURED and PASSED are the same check-run conclusion

⭐ First, the credit, because it narrows the card: the gate is honest when blind. Measured, three states on commit 273247e56f:

event payload	exit	what the step log says
Clause-②: yes in the body, no carrier label	1	⛔ "…grades a package it grew patch" · declaration line: Clause-②: yes — fires on the body line alone
neither carrier nor body line	0	LEVEL AXIS: NOT MEASURED
(fixed head) body line, no carrier	0	✓ LEVEL AXIS: … no package … is graded patch

⇒ it prints LEVEL AXIS: NOT MEASURED rather than inventing a green verdict. ⛔ So this is not "the gate lies", and the filer corrected their own earlier wording to say so.

The defect is one layer out. The check-run conclusion is success in both the "judged and passed" and the "had nothing to judge" cases. A reader who opens the step log can tell them apart. The merge queue, the required-check set, and every summary view read the conclusion, not the log. ⇒ at the only layer that gates a merge, a gate that measured nothing is indistinguishable from a gate that passed.

That is this repo's own stated failure class, applied to its own CI: a reading that cannot fail is indistinguishable from one that passed. Here the reading did not merely not-fail — it did not happen, and said so somewhere nobody with merge authority is reading.

Defect B — the durable declaration is never re-read until someone pushes

.github/workflows/pr-automation.yml triggers on [opened, synchronize, reopened, labeled, unlabeled], its Check Changeset job skips label events, and there is no edited trigger.

⇒ A PR-body edit that adds the Clause-②: line is never read by the gate until the next synchronize — i.e. until someone pushes a commit.

Measured live on both PRs of that round:

• feat(lint): field-typed equality/membership arm for filter-preset-comparand on declared date/datetime fields #16347 — the passing run's event payload was the push at ~15:22:05Z, which predates the body edit at 15:22:26Z. CI judged on the carrier alone and had not yet seen the standalone line. (The implementer confirmed the line by running the script offline against the same head with the edited body.)

• fix(spec): every defineStack refusal carries an ADR-0112 envelope — six STACK_* codes beside STACK_CROSS_REFERENCE_INVALID #16342 — getting CI to read the line required a deliberate git merge origin/main after the body edit, purely to manufacture a synchronize. Run 34042090196 (job 101510576899, 15:23:13Z) then printed, verbatim:

  ✓ LEVEL AXIS: this PR declares clause-② `yes`, and no package whose `packages/*/src/**` it moves is graded `patch`.
  · carrier: `needs:contract-review` is not on this PR (4 label(s) read)
  · declaration line: `Clause-②: yes`
  

  — the first measured pass in that round where the gate read the body line with the carrier already correctly cleared.

⭐ Why the two are one card: the failure needs no mistake

Compose them and the failure mode is reached by a seat doing only correct things, in a defensible order:

1. at-tier review PASSes;
2. the review seat strips needs:contract-review — correct carrier discipline;
3. the seat adds the durable Clause-②: line to the PR body so the declaration outlives the carrier — exactly the remedy [finding] Check Changeset's clause-② level rule is PR-scoped but the fact it judges is package-scoped — a PR that widens package A is refused for grading package B patch, where B only received a comment #16361 proposes;
4. nothing else is pushed.

⇒ Defect B means the gate never re-reads, so it still sees neither carrier nor line ⇒ LEVEL AXIS: NOT MEASURED; Defect A means that concludes success. The level rule — the maintainer ruling of 2026-09-04, decision batch #35, on #15294 — is simply never applied to that PR, and every merge-authority surface shows green.

⇒ ⭐ This also means #16361's proposed remedy does not work on its own. Requiring a body Clause-②: line is defeated by defect B whenever the line is added after the last push. The remedy needs either an edited trigger, or the line to be present from the PR's first push.

Directions (⛔ none chosen — recording what the measurements narrow to)

For A:

• NOT MEASURED should not conclude success (neutral, or a failure requiring an explicit opt-out); or
• it must be visible in the check-run's own output/summary, not only in step logs, so the conclusion carries the distinction to the surfaces that read it.

For B:

• add an edited trigger (⚠️ note it fires on every body edit, including title/description churn — cost and noise want measuring); or
• the filer's own cheapest reading: the Clause-②: line belongs in the PR body at creation time, written by the dev opening the PR rather than patched in later by the seat clearing the carrier. That also matches where the declaration comes from — the claim comment — rather than where it was being retrofitted.

⛔ No tolerance, no allowlist, no comment-only heuristic — carried over from #16361's body and it applies here too.

Priority reasoning (triage)

p2, not p1, and the capping fact comes from the round's own reviewer: all 69 packages sit in one fixed group, with 120 minor / 125 patch pending, so the released bump is set by the highest pending changeset regardless. ⇒ a level axis that goes unjudged on one PR does not, today, reliably ship a widened public surface inside a patch release.

Above p3 because the thing that goes unmeasured is a maintainer ruling, the surface that hides it is the one the merge queue reads, and reaching it requires no mistake by anyone.

Boundaries

• ⚠️ Every reading here is the finding(spec): six of defineStack's seven refusals are still bare Errors — no ADR-0112 code/status — leaving one envelope among neighbours that have none #15963 implementer's, relayed through [finding] Check Changeset's clause-② level rule is PR-scoped but the fact it judges is package-scoped — a PR that widens package A is refused for grading package B patch, where B only received a comment #16361's filer, who states plainly they were taken by running the gate's script offline with --event payloads. ⛔ Not re-driven by triage. Re-run before acting — the three-row table in defect A is the whole of defect A and it is one script invocation per row.
• ⚠️ Run/job ids and timestamps belong to 2026-09-06; the workflow file may have moved since. Re-derive the trigger list from .github/workflows/pr-automation.yml rather than inheriting it.
• ⛔ Nothing was routed around: both PRs of that round complied with the gate as written.

Related

• [finding] Check Changeset's clause-② level rule is PR-scoped but the fact it judges is package-scoped — a PR that widens package A is refused for grading package B patch, where B only received a comment #16361 — the predicate-scope half (false red). Same gate, same file, opposite direction; ⛔ neither subsumes the other, but they will collide on .github/workflows/pr-automation.yml ⇒ schedule in one window.
• finding(changeset): two independent contract reviews read the repo's own history to opposite bumps for "add an exported symbol to a published index" #15294 / decision batch [WIP] Add query enhancements and advanced validation features #35 (2026-09-04) — the level ruling being enforced.
• PRs fix(spec): every defineStack refusal carries an ADR-0112 envelope — six STACK_* codes beside STACK_CROSS_REFERENCE_INVALID #16342, feat(lint): field-typed equality/membership arm for filter-preset-comparand on declared date/datetime fields #16347 — where all of this was measured.

https://claude.ai/code/session_01SwJQDFKe8tVit3BXQ9EfR5

[claude]

Claim: PM loop round 2
Session: session_012GKcPZbMoGq7WPzKLfRBTU
Branch: claude/issue-16776-changeset-level-axis-unmeasured
Worktree: objectstack-issue-16776
Domain: domain:devx
File surface: .github/workflows/pr-automation.yml + scripts/check-changeset-no-major.mjs + its self-test/tests + .changeset/ (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus — default judgement tier (TIER_DEFAULT). ⛔ Not the floor tier: the card records two directions per defect and ⛔ chooses neither; picking between neutral and failure-with-opt-out, and between an edited trigger and a declare-at-creation rule, is the work.
Clause-②: no
Reason for no: the diff is a CI workflow plus the gate script that implements a level rule. packages/spec/src/** is untouched and no published surface moves. ⚠️ Note the irony and do not let it confuse the axis: this card is about the clause-② level gate, it does not make a clause-② change. Judged from content.
Thread-read: the card body in full, and #16361's comment 5560230929, which is where these two defects were first written down.
Serial constraints cleared: all 19 open PRs' changed-file lists fetched paged to exhaustion at 2026-09-08T14:1xZ and filtered for .github/workflows/pr-automation.yml and scripts/check-changeset-no-major.mjs — zero hits; control, 19 of 19 returned a non-empty list.

⛔ #16361 is held behind this card by this seat

The card says it in terms: #16361 is the same gate, same file, opposite direction (a false red from predicate scope), ⛔ neither subsumes the other, and they "collide on .github/workflows/pr-automation.yml ⇒ schedule in one window". #16361 stays pm:queue until this PR is MERGED or closed. ⇒ ⛔ Do not fix #16361's predicate-scope defect here, however tempting it is with the file open. If your reading shows the two fixes are genuinely one edit, ⛔ do not merge them on your own judgement — say so in the report and this seat will answer fold-or-serial.

⭐ Re-drive the measurements first — the card orders it and it is cheap

The card's Boundaries block is explicit: every reading in it is second-hand (the #15963 implementer's, relayed through #16361's filer) and ⛔ not re-driven by triage. Also ⚠️ the run ids and timestamps are from 2026-09-06 and the workflow may have moved.

1. Re-derive the trigger list from .github/workflows/pr-automation.yml on your worktree's base, not from the card. (For your orientation, this seat read :5 as types: [opened, synchronize, reopened, labeled, unlabeled] at 14:1xZ — ⛔ treat that as a hint to check, not as a reading you may inherit.)
2. Re-run defect A's three-row table — it is one script invocation per row against an --event payload, and it is the whole of defect A. Carry a firing control and a nonsense control; a table of three rows with no control proves nothing.
3. Only then choose a fix.

What "fixed" has to mean here

⭐ The card's sharpest point is the one to build against: the merge queue and the required-check set read the check-run conclusion, not the step log. A gate that prints LEVEL AXIS: NOT MEASURED into a log nobody with merge authority reads, and concludes success, is this repo's own stated failure class — a reading that cannot fail is indistinguishable from one that passed — turned on its own CI.

⇒ Whatever you choose, the distinction must reach the layer that gates a merge. If you take the output/summary route rather than the conclusion route, you owe an argument for why that surface is actually read by merge authority, and ⛔ "a careful reviewer could see it" is not that argument.

⛔ No tolerance, no allowlist, no comment-only heuristic — carried from #16361's body, and the card restates it. ⛔ Do not weaken the gate to make a case go green.

⚠️ If you take the edited trigger for defect B, the card's own warning is an acceptance obligation: it fires on every body edit including title/description churn ⇒ measure and state the added run volume. If you take the cheaper route (the Clause-②: line must be present from the PR's first push), then the thing you are changing is partly os-dev's own contract, which lives in .claude/agents/os-dev.md — ⛔ that path is a governed surface this seat cannot land. Stop and report rather than editing it; the gate half may still be landable on its own.

---

Generated by Claude Code