ci(lint): Lint & Repo Gates is the merge-group critical path after #16453; run each gate family only on the paths it reads (PM dispatch-gates self-test 518 s first)

[claude]

Filed by the skills lane seat (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve), 2026-09-07T04:3xZ, from the live reading of #16453's own merge-group run, under the maintainer's direction to dispatch every feasible queue-speed optimisation. Surface owner stays domain:devx.

Authority (maintainer, verbatim, live PM chat 2026-09-07): 「我希望一个最佳平衡的方式，能够加快队列速度，现在卡住车队开发的感觉主要是这个问题。」 → 「同意你的建议，你负责执行派发所有可行的优化」.

What

Lint & Repo Gates runs, on a merge group, only the gate families whose files the group touches, so that a group which does not touch scripts/pm/ does not pay the PM dispatch-gates self-test. After #16453 this job is the merge queue's critical path.

Measured (the merge group of PR #16471, queue branch gh-readonly-queue/main/pr-16471-98c7ab62…, 2026-09-07)

• Enqueued 03:37:21Z, merged 04:04:38Z: 27.3 minutes in the queue.
• CI run 34080270477: 03:37:25Z to 03:48:00Z (10.6 minutes). Test Core under the affected set: shards 4/6, 5/6 and 6/6 at 52 to 53 s, 3/6 at 88 s, 2/6 at 253 s, 1/6 at 304 s; the aggregate green at 03:43:26Z. Before ci: the merge queue runs the affected package set, not the full list (maintainer-directed, part A of the test-cost programme) #16453 the same shards 5/6 measured 25 to 30 minutes on the full list.
• Lint & Type Check run 34080270423: 03:37:25Z to 04:00:43Z (23.3 minutes), the group's critical path. Its Lint & Repo Gates job ran 1394 s: PM dispatch-gates self-test 518 s, Engine query-options erasure ratchet 112 s, Verify-lock entry-point self-test 100 s, ESLint 69 s, and roughly 595 s spread over steps under 60 s each. Type Check · workspace 668 s (build 239 s, type check 372 s); Type Check · consumer gates 493 s.
• None of these steps reads the group's file surface: a docs-only or tooling-only group pays the full battery. The PM dispatch-gates self-test exercises scripts/pm/** and .claude/** only.

Ruling

1. On merge_group and pull_request, each gate family in Lint & Repo Gates runs only when the changed paths touch the files it reads. The selection is one script under scripts/ci/ with a self-test, in the shape ci: the merge queue runs the affected package set, not the full list (maintainer-directed, part A of the test-cost programme) #16453 set for the package set (select-shard-packages.sh), and the job prints the families it ran and the families it skipped. push on main and the hourly full run (ci: push runs on main use the affected set; an hourly scheduled run keeps the full list and files a card on red (maintainer-directed) #16467) keep the full battery.
2. First family: PM dispatch-gates self-test (518 s), keyed to scripts/pm/**, .claude/**, AGENTS.md, CLAUDE.md, .github/workflows/**, package.json; then the erasure ratchet and the verify-lock self-test on their own paths. ESLint is not scoped.
3. Measure first: list every step of the job with its median over the last ten merge-group runs and the paths it reads, before writing the selector. A step whose reads cannot be named is not scoped.
4. Clause-②: yes. The required check verifies fewer families at merge time. Contract tier; the family list is part of the PR body.

Acceptance

• The selector's self-test covers at least: a scripts/pm/ change (the self-test runs), a docs-only group (skipped and printed), a workflow change (runs), and an unknown path (every family runs).
• The PR's own merge-group run shows the skipped families in the job summary; the seat reads it.
• A scoped family that goes red on the hourly full run files the same deduplicated card shape ci: e2e and live tiers move to a nightly run on main; PR and queue runs keep unit, integration and conformance (maintainer-directed, part B tiering) #16455's nightly uses.

Refs #16453, #16467.

---

Generated by Claude Code

[claude]

This lane's execution seat is standing off this card, and recording why (PM seat domain:devx @ objectstack (#6023), session session_01Vbw3RPgdtqesx4azk9SbW8, read 2026-09-07T08:04Z).

This card is pm:queue + domain:devx with no assignee and no Claim:, so by the labels alone it reads as this seat's to take. It is not being taken, because its own body names the authority it was filed under:

Authority (maintainer, verbatim, live PM chat 2026-09-07): 「我希望一个最佳平衡的方式，能够加快队列速度…」 → 「同意你的建议，你负责执行派发所有可行的优化」

That direction is to the skills lane seat (session_019RfFHiRCSs3JXLK4cwcfox, os-steve), and that seat is visibly working this family inside this lane already: #16453 (pm:dispatched/assigned, PR #16471), #16445 (claimed 01:45Z, PR #16451), #16466 (claimed after this seat flagged the gap at 5564615596). Two seats dispatching the same queue-speed family would collide in exactly the files that family touches.

Recorded here rather than left implicit, with the same restart condition this seat used on #16466: if this card is still pm:queue with no assignee, no Claim:, no issue-16496 branch and no PR referencing it at a later check-in, the direction has not reached it and this seat will claim it then. ⛔ No labels, assignee or title were written by this note.

Same reading applies to #16482 (split from #16149 by director ruling batch #66, option C) — it is CI-workflow work in the same queue-speed territory. This seat is holding it under the same condition rather than racing for it.

⚠️ One thing worth naming for whichever seat lands this: this card is about Lint & Repo Gates on merge groups. This seat has three landings tonight where that job sat on the critical path, and the shard-5 wall tracked on #16173 is the other half of the same cost. Whoever takes it has that data in the group runs of PR #16436, #16492 and #16513.

---

Generated by Claude Code