ci(lint): measure where Lint & Repo Gates' 198 serial steps spend 14–37 minutes, and bring the maintainer a split plan (maintainer-directed)

Protects: the merge queue's required context Lint & Repo Gates (AGENTS.md:519), the published check-run name every PR, merge group and release of @objectstack/* passes through; this card keeps the name and cuts its wall.
Ruled: 6120820653 · letter D · 2026-10-12T08:44Z

Filing gate ③: a maintainer-directed task. In the domain:devx seat 1 session (session_012yK1ddAdAqfzwigRHZq3r2), the maintainer asked「22973 一个 ci 30 分钟还没跑完，其他主流类似的系统也是这么慢吗」. The seat answered with the readings below and offered a card to measure the job's step-time distribution and bring back a split plan. The maintainer replied, verbatim:「开卡吧」.

Reader who acts: domain:devx seat 1 dispatches the measurement (part 1). The split plan (part 2) goes to the maintainer as a decision with the four-axis analysis. ⛔ lint.yml is not restructured before that ruling. Lane and grade are triage's.

What is measured (seat 1, 2026-10-12)

PR ci: Test Core's two timing gates exit-grade on schedule alone (C2), with a 20s ceiling floor #22973 (head ef3d0764b): 31 of its 32 non-skipped checks finished within 8.5 minutes. Test Core's slowest shard was 8.5 min, Type Check 3–4 min, Build Docs 3.6 min. Lint & Repo Gates alone was still running at 32 minutes, with 128 of its 197 steps done.
Lint & Repo Gates, newest completed runs (lint.yml, jobs API, 05:06Z–07:50Z):

event runs min median p90 max

pull_request 24 6.1 14.3 20.9 35.9

merge_group 25 10.5 15.5 34.5 37.6

Every run has 198 steps in one job on one runner.
It is the job the queue waits on about half the time. In 13 of those 25 merge groups, Lint & Repo Gates finished after the whole CI workflow of the same group, by up to 23.3 minutes. The differences, lint end minus CI end in minutes, run −14.3 … +23.3.
The workflow's size: lint.yml has 7186 lines and 284 - name: steps, 199 pnpm check: invocations and 118 node scripts/ invocations. scripts/ holds 165 check-*.mjs files.

What to do

Measure (dispatchable). Read the step-level started_at / completed_at that the jobs API returns, over a window of PR and merge_group runs. Report:
- each step's median and p90;
- the top steps by share of the job;
- which steps are already path-scoped by scripts/ci/select-gate-families.sh (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) #16496), and which run on every PR;
- the needs chain, if a later step needs an earlier one's output.
  ⛔ No change to lint.yml in this part.
Propose (for the maintainer's ruling). Give options with workload and four-axis analysis, at least these:
- (a) split the job into N parallel lanes behind an aggregator that keeps the required Lint & Repo Gates context. The precedent is TypeScript Type Check and its Type Check · * lanes in the same file.
- (b) extend 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) #16496's per-family path scoping to the steps that still run on every PR.
- (c) a cache for gates whose inputs did not change.
- (d) a combination.
  Name the critical path each option is expected to reach, as a measurement-backed estimate, never a guess.

Prior work (read before proposing)

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) #16496 (closed completed, PR ci(lint): run each scoped Lint & Repo Gates family only on the paths it reads #16754): select-gate-families.sh, path-scoped families on merge groups.
ci(lint)/pm: the PM dispatch-gates self-test (7.6–13.1 min, one step) is Lint & Repo Gates' critical path on PR and merge_group runs; its family read-set selects it on most PRs — tier the battery or narrow the read-set #22076 (closed completed): the PM dispatch-gates self-test, 7.6–13.1 min, was this job's critical path; it was tiered.
[finding] Lint & Repo Gates 在 merge group 里 ≥26 分钟不结论,队列在它仍 in_progress 时踢掉 PR —— #16496 修过的关键路径形状再现,而且连着两组都命中 #19189 (closed not_planned on 2026-09-25, triage 5831394361): "Lint & Repo Gates is the merge-group critical path". It was closed because the job then measured 8–14 min at parity with the other legs. Its reopen condition: "a merge group's Lint & Repo Gates is again the job the queue waits on". The table above meets that condition in 13 of 25 groups.
Split Lint & Repo Gates into parallel lanes behind an aggregator, the way TypeScript Type Check already is: 193 serial steps are 28 minutes of every PR's critical path and one step is 41% of them #19686 (closed duplicate of [finding] Lint & Repo Gates 在 merge group 里 ≥26 分钟不结论,队列在它仍 in_progress 时踢掉 PR —— #16496 修过的关键路径形状再现,而且连着两组都命中 #19189): the earlier measured split proposal, "193 serial steps are 28 minutes … the way TypeScript Type Check already is".
ci(lint): every job in lint.yml carries a timeout-minutes derived from its measured duration — Lint & Repo Gates has none, so a hang holds the merge-queue head until GitHub's 360-minute default #21050: this job's timeout-minutes.

Dedupe

MCP search_issues on this repo, closed included: "Lint & Repo Gates job duration serial steps shard parallel matrix split lint.yml wall time" → 31 hits, all read by title. None is open.
The nearest are [finding] Lint & Repo Gates 在 merge group 里 ≥26 分钟不结论,队列在它仍 in_progress 时踢掉 PR —— #16496 修过的关键路径形状再现,而且连着两组都命中 #19189 and Split Lint & Repo Gates into parallel lanes behind an aggregator, the way TypeScript Type Check already is: 193 serial steps are 28 minutes of every PR's critical path and one step is 41% of them #19686, both closed. This card is new because it is maintainer-directed, and it asks for a measurement and a plan, not an immediate split.
Dedupe words: Lint & Repo Gates; serial steps; critical path; split lanes; aggregator; select-gate-families.

Generated by Claude Code

Ruling: batch #326 item 1 · letter D · maintainer 「同意，立卡」 2026-10-12T08:44Z

Director seat, summon #37, session_01LqNgdPZB2Jnr5HpGbaqLeB (GitHub os-zhuang; written as objectstack-fleet[bot] via the relay). The maintainer asked for this card's analysis on the direct channel (「22991 帮我综合分析，和我讨论」); this seat measured the job (part 1 of the card, below), presented the four options of part 2 with the critical path each reaches, and recommended D. The maintainer stated the governing principle verbatim, 「我觉得能让整个车队提速的方案就是好方案，main 很多天才会发一个版本，我只要确保发版本之前做好各种测试就可以。」, and answered 「同意，立卡」. The principle is filed as the governance card #22998 (two-speed CI); this card is its first execution.

The ruling

D: option (a), lanes behind an aggregator, plus the PM dispatch-gates self-test's slow tier leaves the pull_request / merge_group path.

Lint & Repo Gates becomes an aggregator job that only needs: the lanes, name kept (the required context is matched by name, AGENTS.md:519); the lanes get new names. The roster discipline of TypeScript Type Check applies: OS_AGGREGATOR_MEMBERS in YAML, scripts/check-aggregator-roster.mjs statically, toJSON(needs) at run time.
Lane cut, from the sum of step medians: ESLint with the engine test-double gate (about 2.3 min); the 137 tree-wide gates in three lanes of about 3 min each; the scoped ratchets and corpus checks (query-options erasure 143 s, slot-lookup 75 s, comment-mask 61 s, verify-lock 99 s) in one lane; the PM dispatch-gates self-test in its own lane. Each lane pays about 1 min of checkout and install. Lane budget: 5 minutes at median; a lane over budget splits, it does not skip.
pnpm check:pm-dispatch-gates on pull_request / merge_group runs the FAST tier (--self-test --fast); the SLOW tier runs on schedule and push: main, the full battery. ci(lint)/pm: the PM dispatch-gates self-test (7.6–13.1 min, one step) is Lint & Repo Gates' critical path on PR and merge_group runs; its family read-set selects it on most PRs — tier the battery or narrow the read-set #22076's constraint stands: every case still runs at least daily. The slow tier's red is a main-red under the governance card's rule.
Not taken: (b), extending per-family path scoping to the 137 tree-wide gates (their input is the tree; a wrong declaration skips a gate silently); (c), input caches (the same declaration problem). The 50 always-on gate self-tests cost 1.8 min together and are not worth scoping.
Expected critical path: a product-only run about 4–7 min (from 12.9 median today); a PM-touching run about 5–8 min (from 29.2); the queue entry then waits on the CI workflow (13.6 min median), which the governance card addresses.

What this seat measured (part 1 of the card; 31 completed `Lint & Repo Gates` runs, 18 `pull_request` and 13 `merge_group`, read from the jobs API on 2026-10-12)

reading	value
job wall min / median / p90 / max	6 / 15.5 / 32.1 / 36.4 min
steps	198, serial, one runner; 187 always-on (11.5 min of medians) and 11 path-scoped (17.1 min when all run)
step dependencies	none except `steps.gate-families.outputs.*` (9 reads); one step id in the job
PM dispatch-gates self-test	ran in 10 of 31 runs; median 534 s, p90 861 s, max 903 s; the job's wall with it 29.2 min median (max 36.4), without it 12.9 (max 20.9)
other scoped steps	query-options erasure 143 s (ran 20 of 31) · verify-lock 99 s (9) · slot-lookup 75 s (20) · comment-mask corpus 61 s (24) · workflow-command gate 48 s (11)
always-on heavies	ESLint 96 s · engine test-double 39 s · checkout 31 s · tenant-audit census 28 s · ADR anchors 25 s
tail	145 steps under 5 s = 2.3 min; 44 steps of 5–30 s = 7.6 min; 9 steps of 30 s or more
merge-group walls (success)	12.1–36.4 min, median 13.8; the `CI` workflow of the same groups 13.6 median

The per-run and per-step data sit in this seat's scratchpad (audit/lint-steps.json); the reading script uses only the jobs API and the dev can re-run it.

Prior rulings read: #19189 (closed not_planned; its reopen condition is met by the table above, and this card carries it), #19686 (the aggregator pattern and roster discipline), #22076 (the two tiers; CI runs both today per the header of scripts/pm/check-dispatch-gates.mjs), #16496 (five scoped families), #21050 (timeouts), AGENTS.md:519 (required contexts by name).

After this ruling

Labels: tooling · domain:devx · area:devpath · priority:p1 (the grade [finding] Lint & Repo Gates 在 merge group 里 ≥26 分钟不结论,队列在它仍 in_progress 时踢掉 PR —— #16496 修过的关键路径形状再现,而且连着两组都命中 #19189 carried when this job stalled the queue; triage may re-grade) · pm:queue. Body line 1 names the protected surface (the tooling queue-entry rule).
Follow-ups named, not carried: Type Check · workspace (11.8 min median on merge groups, the next bottleneck) and the Test Core shard variance (3.7 min median, 15.2 p90), on the governance card's lane.
⛔ No code by this seat; no dev dispatched by this seat.

Claim: PM loop round 1 · the director seat dispatching on the maintainer's order, verbatim 「三仓的 ci 优化卡你负责派发」 (this seat's session, 2026-10-12T08:55Z)
Session: session_01LqNgdPZB2Jnr5HpGbaqLeB
Account: os-zhuang
Branch: claude/issue-22991-lint-lanes
Worktree: objectstack-issue-22991
Domain: domain:devx
Seat: domain:devx#9
File surface: .github/workflows/lint.yml (the lint job becomes lanes plus an aggregator named Lint & Repo Gates), scripts/check-aggregator-roster.mjs (only if a second aggregator needs a parameter), scripts/pm/check-dispatch-gates.mjs and the root package.json script check:pm-dispatch-gates (the fast tier on pull_request / merge_group, both tiers on schedule / push); scripts/ci/select-gate-families.sh only if a lane needs the selector's output as a job output (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus (the default judgment tier; dispatch-gates --tier --repo objectstack-ai/objectstack on the paths answers "no path-derived mandate … floor sonnet · default opus · ceiling fable"; the lane cut is a design judgment, so the default, not the floor)
Clause-②: no
Responsibility: n/a — not a defect card
Thread-read: 6120820653
Ruling-ref: 6120820653
Serial constraints cleared: #22998 (this seat's sibling claim, same round) edits .github/workflows/ci.yml, .github/workflows/release.yml, AGENTS.md and two references/*.md files, not lint.yml; its AGENTS.md paragraph names Lint & Repo Gates, which this card keeps by name. In-flight pm:dispatched claims read in this act (14 cards): none declares lint.yml (#22979 declares SKILL.md and platform-readings.md). PR #22916 (S2, pm:blocked) does not touch lint.yml. Merge origin/main before opening the PR and again after #22998 lands.

os-dev-report
{
"issue": 22991,
"status": "done",
"branch": "claude/issue-22991-lint-lanes",
"pr": "#23014",
"session": "session_01LqNgdPZB2Jnr5HpGbaqLeB (subagent: the dispatching director seat's id)",
"premise_still_valid": true,
"summary": "Ruling D is implemented in draft PR #23014 (head 79b0fd6). The 193-step lint job is now eight parallel Lint · * lanes. The aggregator keeps job id lint and the name Lint & Repo Gates byte-identical, and follows the typecheck roster discipline: OS_AGGREGATOR_MEMBERS equals needs:, toJSON(needs) is checked at run time, and the job runs under if: always(). Every gate step moved verbatim, proven by a parse-and-compare: 184 steps, same run/if/comment, same relative order, other jobs and triggers unchanged. check:pm-dispatch-gates now runs the fast tier on pull_request and merge_group, and both tiers on every other CI event. The PR's own run 38233160614 is green: all 8 lanes, the aggregate, and Lint & Repo Gates on the check list. The aggregate verdict landed 5.57 min after run start, against a 35.5 min median for comparable runs before.",
"lanes_measured_on_pr_run": {
"run": "38233160614 (pull_request, head 79b0fd6; every scoped family selected, so this is the heaviest PR shape)",
"note": "One run, so these are single-run walls, not medians. 'predicted p50' is the 70-run rebuild. Wall = job started_at to completed_at; body = wall minus the checkout step.",
"rows": [
"lint-docs-code | Lint · docs & code guards | wall 5.42 | body 2.65 | checkout 166 s (outlier; window median 32, max 42) | predicted p50 4.1",
"lint-scoped-erasure | Lint · scoped · query-options erasure | wall 4.37 | body 3.82 | erasure ratchet 167 s | predicted p50 3.5",
"lint-scoped-slot-corpus | Lint · scoped · slot lookup & corpus | wall 4.30 | body 3.82 | slot-lookup 87 s, corpus 72 s, entry-guard 34 s | predicted p50 3.5",
"lint-scoped-tooling | Lint · scoped · tooling self-tests | wall 3.92 | body 3.38 | verify-lock 99 s, workflow-cmd 52 s, bare-root 21 s | predicted p50 1.1",
"lint-pm-dispatch-gates | Lint · PM dispatch-gates | wall 3.88 | body 3.42 | PM self-test 174 s (fast tier; both-tier window n=23 min 477 / median 847 / max 903 s) | predicted p50 1.1",
"lint-eslint | Lint · ESLint | wall 3.52 | body 2.97 | ESLint 105 s, engine test-double 42 s | predicted p50 3.3",
"lint-ci-engine | Lint · CI, release & engine gates | wall 2.98 | body 2.55 | predicted p50 3.9",
"lint-repo-tooling | Lint · repo & tooling gates | wall 2.92 | body 2.52 | predicted p50 4.2",
"lint (aggregate) | Lint & Repo Gates | wall 0.07 | verdict at +5.57 min after run start"
]
},
"critical_path": {
"before": "Old single job, 70 successful runs 2026-10-12T04:40Z to 08:45Z: p50 17.8 / p90 35.9 / max 38.5 min. Runs where every heavy scoped family ran (n=22, the shape of this PR): min 22.9 / median 35.5 / max 38.5. Runs without the PM self-test (n=47): median 14.3, max 20.9.",
"after_on_pr_run": "5.57 min from run start to the Lint & Repo Gates verdict. The slowest lane is docs-code at 5.42, and that comes from its 166 s checkout. Without the outlier the slowest is the erasure lane at 4.37. Type Check · workspace (9.62) is now the longer required leg in the same run, the next bottleneck the ruling names.",
"after_predicted": "From the 70-run rebuild: p50 4.2 / p90 4.7 / max 5.1 min on pull_request and merge_group. On schedule and push the PM lane runs both tiers, about 15–16 min."
},
"files_changed": [
".github/workflows/lint.yml +1443/-615 (the lint job becomes the aggregator plus 8 lanes; comments now name the step a moved neighbour referred to)",
"scripts/pm/check-dispatch-gates.mjs +91/-28 (tier by GITHUB_EVENT_NAME under GITHUB_ACTIONS; the self-test covers each event)",
"scripts/check-required-contexts.mjs +101/-12 (the pin may live in a lane the lint aggregate declares AND needs, with no if: at step or lane level; 4 new fixtures)",
"scripts/check-aggregator-roster.mjs +23/-6 (lint registered as a required aggregate; lint positive controls; typecheck fixtures anchored)",
"PR total +1658/-661 = 2319 changed lines (under the 3000 human-merge size line)"
],
"gates": [
"node scripts/check-aggregator-roster.mjs :: exit 0",
"node scripts/check-aggregator-roster.mjs --self-test :: exit 0",
"node scripts/check-ci-filter-parity.mjs :: exit 0",
"node scripts/check-closing-keyword-parity.mjs :: exit 0",
"node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0",
"node scripts/check-comment-mask-corpus.mjs :: exit 0",
"node scripts/check-declaration-mirrors.mjs :: exit 0",
"node scripts/check-declaration-mirrors.mjs --self-test :: exit 0",
"node scripts/check-merged-branch-reaper-outcome.mjs :: exit 0",
"node scripts/check-merged-branch-reaper-outcome.mjs --self-test :: exit 0",
"node scripts/check-position-name-fold-loaders.mjs :: exit 0",
"node scripts/check-position-name-fold-loaders.mjs --self-test :: exit 0",
"node scripts/check-scripts-symbol-anchors.mjs :: exit 0",
"node scripts/check-scripts-symbol-anchors.mjs --self-test :: exit 0",
"node scripts/check-self-test-wired.mjs :: exit 0",
"node scripts/check-self-test-wired.mjs --self-test :: exit 0",
"node scripts/check-self-test-workflow-commands.mjs :: exit 0",
"node scripts/check-self-test-workflow-commands.mjs --self-test :: exit 0",
"node scripts/check-step-collectors.mjs :: exit 0",
"node scripts/check-step-collectors.mjs --self-test :: exit 0",
"node scripts/check-whole-set-label-write.mjs :: exit 0",
"node scripts/check-whole-set-label-write.mjs --self-test :: exit 0",
"node scripts/ci/scheduled-full-run.mjs --self-test :: exit 0",
"node scripts/docs-audit/check-drift-comment.mjs :: exit 0",
"node scripts/pm/bare-root-worklist.mjs --self-test :: exit 0",
"node scripts/pm/ci-failure.mjs --self-test :: exit 0",
"pnpm check:agent-test-spelling :: exit 0",
"pnpm check:bash32-floor :: exit 0",
"pnpm check:cli-command-ids :: exit 0",
"pnpm check:cross-package-test-inputs :: exit 0",
"pnpm check:declared-population-live :: exit 0",
"pnpm check:driver-memory-census :: exit 0",
"pnpm check:entry-guard :: exit 0",
"pnpm check:gitlink-declared :: exit 0",
"pnpm check:node-version :: exit 0",
"pnpm check:nul-bytes :: exit 0",
"pnpm check:parse-guard :: exit 0",
"pnpm check:pm-expected-skips :: exit 0",
"pnpm check:pm-post-stamped :: exit 0",
"pnpm check:pnpm-acquisition :: exit 0",
"pnpm check:pnpm-filter-targets :: exit 0",
"pnpm check:ratchet-remedy-authority :: exit 0",
"pnpm check:refd-timer-probe :: exit 0",
"pnpm check:required-contexts :: exit 0",
"pnpm check:select-gate-families :: exit 0",
"pnpm check:shard-attestation :: exit 0",
"pnpm check:stall-guard-budget :: exit 0",
"pnpm check:stall-guard-headroom :: exit 0",
"pnpm check:type-check-coverage :: exit 0",
"pnpm check:watch-hint-literal :: exit 0",
"pnpm check:workflow-status-functions :: exit 0",
"pnpm check:workflow-step-name-quoting :: exit 0",
"pnpm check:pm-dispatch-gates :: exit 0",
"NOT-MEASURED pnpm check:type-check-debt :: its --self-test half passed and its coverage verdict printed OK on f9c1ba0; the --re-measure half runs tsc over the ledgered package and was cut at this container's 500 s timeout (exit 124) before a verdict, and needs the build CI's Type Check · debt ledger lane does first. This diff touches no TypeScript program; the family is matched only because the script reads lint.yml.",
"node scripts/pm/dispatch-gates.mjs --ran ran.txt --repo objectstack-ai/objectstack :: exit 0 (54 derived, 53 run, 1 NOT-MEASURED, 0 UNRUN; at 79b0fd6)"
],
"tests": "The local union was run at 79b0fd6, the final head after merging origin/main. The set came from node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack: 54 commands. 52 ran in the foreground, all exit 0 (482 s). pnpm check:pm-dispatch-gates ran detached, exit 0: fast tier, 1828 cases, 264.9 s. pnpm check:type-check-debt is NOT MEASURED: its --self-test passed, but --re-measure is tsc over the ledgered package, needs the lane's build, and was cut at 500 s (exit 124); this diff touches no TS program. --ran reconciled 54/53/1/0. YAML was parsed by both yaml and PyYAML; actionlint is not installed. Move proof: a one-off parse-and-compare of old vs new lint.yml reported 0 problems over 184 gate steps. Ablations, each through scripts/ablation-replace.mjs, mutation confirmed on disk and restored blob == HEAD:\n(1) FAST_TIER_EVENTS emptied: first run EXIT 0, because the self-test looped over the constant under test. Fixed by naming the two events literally; rerun red, 2 cases.\n(2) Event branch forced to always-fast: red, 5 cases.\n(3) lint-ci-engine deleted from the aggregate's needs: check:required-contexts red, naming the unwired pin.\n(4) The needs: read in jobsCarriedByLint deleted: red on the declared-only fixture.\n(5) lint dropped from REQUIRED_AGGREGATORS: roster self-test red on the lint refusal fixture.\nTwo earlier ablation attempts exited 1 from the tool refusing an empty --replacement; no mutation landed there, and both were rerun with --delete. CI: run 38233160614 is green on every lane and the aggregate.",
"deviations": [
"H3 variant: the fast tier is a CLOSED list (pull_request, merge_group) rather than 'fast unless schedule or push'. workflow_dispatch and any unlisted event keep both tiers. The ruling moved the slow tier off those two events only, and this keeps the safe failure direction.",
"H2 cut refined: H2's single scoped lane rebuilds to p50 6.4 / max 11.5 min, over the ruled 5-minute budget. The scoped families are split over three lanes plus the PM lane, and query-options erasure and slot-lookup are kept apart.",
"File surface: scripts/check-required-contexts.mjs is outside the claim's declared surface. It was edited because its wiring pin read jobs.lint.steps literally and went red when the pin step moved into a lane. The edit is an honest re-pin, not a loosening. check-aggregator-roster.mjs got a registry row and fixtures, not a parameter.",
"The root package.json check:pm-dispatch-gates script is unchanged: the event decision lives inside the .mjs, so no workflow or manifest edit was needed.",
"New keys beyond pure moves: each lane's checkout step has timeout-minutes: 20, the Type Check lanes' checkout-budget scheme. Lane walls are 30, and 50 for the PM lane. The aggregate wall dropped from 55 to 10. The tail-reporter pair is repeated in every lane and the selector in the four scoped lanes.",
"The PR body carries the PREDICTED lane table: the body is written once, at create, before its own run existed. The measured walls are in this report; the seat may fold them into the body.",
"PM fast tier on CI is evidenced by duration only: 174 s, against a both-tier minimum of 477 s (n=23). The job log is blob-hosted and not reachable over REST from this container, and no MCP read was made.",
"Commit trailers use the model-free pair AGENTS.md prescribes, not the harness reminder's model-named trailer.",
"origin/main was merged once, before the PR opened (79b0fd6). main has moved at least one commit since, and #22998 has not landed, so the claim's 'merge again after #22998 lands' is still owed."
],
"mcp_calls": "0",
"api_writes": "3 REST writes, each one relay dispatch (POST /repos/objectstack-ai/objectstack/dispatches) that the board workflow executed as objectstack-fleet[bot]: (1) pr_create: POST /pulls (draft) + POST /issues/23014/assignees, run 38233112121; (2) label-write skip-changeset: POST /issues/23014/labels, run 38233208427; (3) this os-dev-report comment: POST /issues/22991/comments. Separately, 6 git push to claude/issue-22991-lint-lanes (not REST).",
"open_questions": [],
"out_of_scope_findings": [
"carrier: 承接者:无 · noted, not filed. check:select-gate-families pins the nine scoped if: spellings as text, but does not pair each with a selector step in the SAME job. A lane that loses its selector fails OPEN (the step runs on every PR), never silent-skip. Recorded in the PR's Acceptance notes.",
"carrier: 承接者:无 · noted, not filed. Every lane keeps fetch-depth: 0, because which of the ~190 gates read history was not re-derived. A per-lane shallow checkout would save about 20 s a lane. Recorded in Acceptance notes.",
"carrier: 承接者:无 · noted, not filed. The PM step's comment says 'Same family as the Changeset-family gate self-tests step above', but that step was already BELOW it before this split. Pre-existing prose drift; left as is."
]
}

Generated by Claude Code

Review: ACCEPT · PR #23014 (draft, head `79b0fd6d`) · director seat dispatching on the maintainer's order, 2026-10-12T10:12Z

Checked on GitHub, not on the report's say-so:

PR shape: exists, draft, base main, body line 1 Fixes #22991, assignee os-zhuang, labels ci/cd · size/xl · skip-changeset (no package changes; the classifier's label is right for a workflow-and-scripts diff). 4 files, +1658/−661 = 2,319 changed lines, under the 3,000 human-merge line: .github/workflows/lint.yml (+1443/−615), scripts/pm/check-dispatch-gates.mjs (+91/−28), scripts/check-required-contexts.mjs (+101/−12), scripts/check-aggregator-roster.mjs (+23/−6). No content/docs/releases/.
Claim file surface amended in this review (the in-scope adjacency rule: same defect class, mechanical re-pin, no other claim on the file, same gate family): scripts/check-required-contexts.mjs, whose wiring pin read jobs.lint.steps literally and went red when the pin step moved into a lane; the edit is a re-pin with four new fixtures, not a loosening (ablation 3 in the report: deleting lint-ci-engine from the aggregate's needs: turns it red).
Ruling D, held: the lint job id and the name Lint & Repo Gates are byte-identical on the PR's check list; the aggregate declares OS_AGGREGATOR_MEMBERS, equals needs:, re-checks toJSON(needs) at run time under if: always(); 184 gate steps moved verbatim (the dev's parse-and-compare: 0 problems); the fast tier runs on pull_request and merge_group as a closed list, both tiers on every other event (a stricter variant of H3, accepted: the safe failure direction). Lane cut refined from H2 to eight lanes so no lane rebuilds over the 5-minute budget.
Checks on head 79b0fd6d: Lint & Repo Gates success; the eight Lint · * lanes success; TypeScript Type Check and its four lanes success (Type Check · debt ledger covers the one locally NOT MEASURED half of check:type-check-debt, cut at the container's 500 s cap); Governed Surface Queue Guard success; Build Core, Temporal Conformance skipped by the path filter (no package touched).
Gates: 54 derived by dispatch-gates --commands, 53 run exit 0 (the PM battery detached: fast tier, 1,828 cases, 264.9 s), 1 NOT MEASURED with its reason, --ran reconciled 54/53/1/0 at the head. Five ablations through scripts/ablation-replace.mjs, mutation confirmed on disk and restored blob == HEAD; the first ablation's false green (the self-test looping over the constant under test) was fixed by naming the events literally and re-run red.
Measured on the PR's own run 38233160614 (the heaviest PR shape: every scoped family selected): verdict of Lint & Repo Gates at 5.57 min from run start. Lanes (wall min): docs & code guards 5.42 (a 166 s checkout outlier; window median 32 s), scoped · query-options erasure 4.37, scoped · slot lookup & corpus 4.30, scoped · tooling self-tests 3.92, PM dispatch-gates 3.88 (fast tier 174 s against a both-tier minimum of 477 s), ESLint 3.52, CI/release/engine gates 2.98, repo & tooling gates 2.92. Before, on 70 runs the same morning: p50 17.8 / p90 35.9 / max 38.5; runs of this shape (n=22) median 35.5. Predicted from the 70-run rebuild: p50 4.2 / p90 4.7 / max 5.1 on pull_request and merge_group; 15–16 min on schedule and push, where the PM lane runs both tiers. Type Check · workspace (9.62 min in the same run) is now the longer required leg: objectstack#23017.
Deviations accepted: per-lane checkout timeout-minutes: 20 (the Type Check lanes' scheme), lane walls 30 (PM lane 50), aggregate wall 10; the selector repeated in the four scoped lanes and the tail-reporter pair in every lane; the PR body carries the predicted table, the measured one is above; the fast tier on CI is evidenced by duration (the job log host is unreachable from a container). Writes: 3 REST (pr_create, label-write, the report); mcp_calls 0.
Still owed before landing: merge origin/main again after skills(ci): two-speed CI — fast set on PR and merge-queue runs, full battery hourly on main and before a release, main-red stops the lane, four required contexts plus the guard, timing gates report-only, test budgets not deletions #22998 lands (the claim's serial line; main has moved since 79b0fd6d). The dev or this seat does it at landing time; a bare base merge carries no other change.

Landing: Tier H; the maintainer approves and merges by hand. This card stays pm:dispatched until the PR lands; the landing record follows the merge.

LANDED — PR #23014 is MERGED (2026-10-12T11:29Z, merge commit 5dda6002) and read in place on origin/main 5dda6002. Director seat, summon #37, session_01LqNgdPZB2Jnr5HpGbaqLeB, 2026-10-12T11:33Z.

probe on `origin/main`	want	got
`lint.yml` lanes named `Lint · *`	8	8
`OS_AGGREGATOR_MEMBERS` in `lint.yml` (the lint aggregate joins the typecheck one)	>1	11 hits
the job named `Lint & Repo Gates`	exactly 1	1
`GITHUB_EVENT_NAME` read in `scripts/pm/check-dispatch-gates.mjs` (the fast/slow tier by event)	>0	9 hits

The card was closed by the PR's Fixes line; pm:dispatched stripped and the assignee released in this act. The queue closed no other card (Fixes #22991 alone).
The claim's serial line ("merge origin/main again after skills(ci): two-speed CI — fast set on PR and merge-queue runs, full battery hourly on main and before a release, main-red stops the lane, four required contexts plus the guard, timing gates report-only, test budgets not deletions #22998 lands") is moot: the maintainer queued ci(lint): Lint & Repo Gates becomes an aggregate over eight parallel lanes; PM dispatch-gates runs its fast tier on PR and queue (#22991) #23014 first; ci: two-speed CI — PR and queue runs carry the fast set, the full battery runs on main and gates the publish lane, required set 7 → 5 #23020 (skills(ci): two-speed CI — fast set on PR and merge-queue runs, full battery hourly on main and before a release, main-red stops the lane, four required contexts plus the guard, timing gates report-only, test budgets not deletions #22998) builds on this main in the queue and carries the second edit to scripts/check-required-contexts.mjs (disjoint hunks).
Owed, read from the next merge groups: the lint verdict time on pull_request and merge_group runs of main's new shape (the PR's own run gave 5.57 min against a 35.5 median for its shape; predicted p50 4.2 / p90 4.7), and the slow-tier wall on the next schedule run (predicted 15–16 min). Recorded here after ten groups.
Unblocked by this landing: objectstack#23017 (Type Check · workspace, the same file), claimed and dispatched next.

Navigation Menu

ci(lint): measure where `Lint & Repo Gates`' 198 serial steps spend 14–37 minutes, and bring the maintainer a split plan (maintainer-directed) #22991

What is measured (seat 1, 2026-10-12)

What to do

Prior work (read before proposing)

Dedupe

The ruling

What this seat measured (part 1 of the card; 31 completed `Lint & Repo Gates` runs, 18 `pull_request` and 13 `merge_group`, read from the jobs API on 2026-10-12)

After this ruling

Review: ACCEPT · PR #23014 (draft, head `79b0fd6d`) · director seat dispatching on the maintainer's order, 2026-10-12T10:12Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

event	runs	min	median	p90	max
`pull_request`	24	6.1	14.3	20.9	35.9
`merge_group`	25	10.5	15.5	34.5	37.6

ci(lint): measure where Lint & Repo Gates' 198 serial steps spend 14–37 minutes, and bring the maintainer a split plan (maintainer-directed) #22991

Description

What is measured (seat 1, 2026-10-12)

What to do

Prior work (read before proposing)

Dedupe

Activity

objectstack-fleet commented on Oct 12, 2026

The ruling

What this seat measured (part 1 of the card; 31 completed Lint & Repo Gates runs, 18 pull_request and 13 merge_group, read from the jobs API on 2026-10-12)

After this ruling

objectstack-fleet commented on Oct 12, 2026

objectstack-fleet commented on Oct 12, 2026

objectstack-fleet commented on Oct 12, 2026

Review: ACCEPT · PR #23014 (draft, head 79b0fd6d) · director seat dispatching on the maintainer's order, 2026-10-12T10:12Z

objectstack-fleet commented on Oct 12, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

ci(lint): measure where `Lint & Repo Gates`' 198 serial steps spend 14–37 minutes, and bring the maintainer a split plan (maintainer-directed) #22991

What this seat measured (part 1 of the card; 31 completed `Lint & Repo Gates` runs, 18 `pull_request` and 13 `merge_group`, read from the jobs API on 2026-10-12)

Review: ACCEPT · PR #23014 (draft, head `79b0fd6d`) · director seat dispatching on the maintainer's order, 2026-10-12T10:12Z