[temporary] raise `Test Core (N/6)` timeout-minutes 30 → 45 while #16173's shard balance is unfixed — and un-censor the readings that #16173 needs

[huangyiirene]

Maintainer directive, verbatim: 「临时抬墙」 (2026-09-07, in session session_01T6HeZvT9wdSJD1ZxJb5Eno), chosen from three options put to the maintainer: sanction a re-run / raise the wall / assign a devx seat.

⚠️ Seat note, stated rather than hidden: the landing point is .github/workflows/**, so this card is domain:devx by shape. It is being claimed and dispatched by the domain:spec PM seat because the maintainer directed that seat to execute it. Recorded here so the claim does not read as a lane trespass.

What to change — one line

.github/workflows/ci.yml:295 — timeout-minutes: 30 on the test job (name: Test Core (${{ matrix.shard }}/6)) → 45.

45 is not an arbitrary number: the same file already runs console-pin at timeout-minutes: 45, so this is a value the repo already tolerates on a required job.

Why 45, from measurement

Test Core (5/6) has been killed at the 30-minute wall repeatedly today — 12+ observations at 30:16 ×7 · 30:17 ×2 · 30:18 ×2 · 30:21, every other job in every one of those runs green.

Those numbers are not measurements of the shard. They are readings of timeout-minutes: the true duration is ≥ 30:16, censored and unbounded above. The only uncensored shard-5 readings are 25:32 and 27:36, and shard 6/6 — which has never been killed — has been observed as high as 28:56.

So the observed green band is 25:30–28:56 against a 30:00 wall: a margin of about a minute at the top. 45 restores real headroom without approaching console-pin's precedent from above.

⭐ The second reason, which is the more valuable one

I recorded on #16173 that the shard timings cannot be re-derived from CI history, because every observation in the tail is censored by the wall. Raising the wall fixes that: the next runs will show how long shard 5 actually takes, which is precisely the input #16173 needs to rebalance the shards properly.

⇒ This raise is an instrument as well as an unblock. Whoever fixes #16173 should read the uncensored durations this produces before touching scripts/test-shard-timings.json.

Scope fences — binding

• ⛔ Change only the timeout-minutes on the Test Core (N/6) matrix job at :295. The file has nine other timeout-minutes values (filter 10, aggregate Test Core 10, temporal-conformance 30, dogfood 30, dogfood-verify 20, dogfood-gate 10, build-core 30, build-docs 30, console-pin 45). ⛔ None of them moves.
• ⛔ Do NOT touch scripts/test-shard-timings.json. That constraint was not lifted and the rebalance is CI: the shard-timings file is stale for the CLI package — 672s predicted vs 28m46s measured against a 30-minute timeout, so Test Core shard 1/6 is one slow run from being killed on any PR touching the CLI #16173's, owned by a devx seat.
• ⛔ Do NOT re-run any cancelled shard. That option was not the one chosen.
• ⛔ Do not "also fix" the shard split, skip a test, or change any vitest config.

Temporary means it carries its own expiry

The new value must carry an inline comment naming #16173 as the reason and stating the revert condition explicitly — this returns to 30 once the shard balance is fixed. A temporary raise with no expiry written next to it becomes permanent by forgetting, which is the failure mode this repo's ratchets exist to prevent.

Revert trigger

Close this card by reverting to 30 when #16173 lands a rebalance. Until then this card stays open as the record that a wall was raised and why.

[huangyiirene]

Claim:

• session: session_01T6HeZvT9wdSJD1ZxJb5Eno (domain:spec PM dispatch seat, executing a direct maintainer instruction on a domain:devx-shaped card — see the card body's seat note)
• branch: claude/issue-16445-raise-test-core-shard-wall
• dispatched at: default tier (opus)

The assignee and this comment are one act: the PM sets both, the dev inherits both and posts no second claim.

Clause ② reads no: a CI job's timeout-minutes, no package surface, no schema, no export, no accept-set. Confirm rather than inherit; if your reading is yes, stop and report.

---

Generated by Claude Code

[huangyiirene]

os-dev-report

{
  "issue": 16445,
  "status": "done",
  "branch": "claude/issue-16445-raise-test-core-shard-wall",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/16451",
  "premise_still_valid": true,
  "summary": "Executed the maintainer directive 「临时抬墙」 verbatim. `.github/workflows/ci.yml`, job `test` (`name: Test Core (${{ matrix.shard }}/6)`): `timeout-minutes` 30 -> 45. The line was at :295 on base bf45e62ce as the card said, and now sits at :316 — the 21 added comment lines account for the whole shift. Those comment lines are the expiry: they name #16173 as the reason, state the raise is temporary, and state the revert condition explicitly (back to 30 once #16173 lands its shard rebalance). The pre-existing rationale paragraph above the value is left byte-identical — it argues for 30 and is exactly what the revert must satisfy — so the new block says in words that the paragraph above describes the value this line must RETURN to, not the value it now carries. Scope fences held: scripts/test-shard-timings.json untouched, no matrix/shard-count/vitest change, no test skipped, no cancelled shard re-run, content/docs/releases untouched. GOVERNED SURFACE: NO, measured not assumed — check-governed-queue-guard.mjs restates no list of its own (it imports GOVERNED_SURFACES from check-governed-merges.mjs; its --self-test pins case `this-file-restates-no-surface-list`), and `node scripts/pm/check-governed-merges.mjs --test .github/workflows/ci.yml` printed 'NOT governed — ordinary queue landing applies' at EXIT=0 (EXIT_TEST_NOT_GOVERNED=0, EXIT_TEST_GOVERNED would be 3), re-run on the FINAL one-path file list; control path .claude/x.md DOES hit the same predicate, so the negative is not a dead matcher; today's register printed from the module is `docs/adr/** · .claude/** · skills/** · AGENTS.md · CLAUDE.md`; and the live `Governed Surface Queue Guard` check on the PR concluded success. So the ordinary merge-queue path applies — but the PR is left DRAFT, unflipped and unqueued, as instructed. CLAUSE ②: NO, confirmed independently — a CI job wall-clock backstop touches no packages/spec/src/**, no *.zod.ts, no error-code ledger, no export, no accept-set; nothing published accepts or rejects differently and no public surface widens. CHANGESET: not owed; nothing here publishes, so `skip-changeset` is right and it is a real mechanism in this repo (changeset-check in pr-automation.yml is label-exempted, with no path-based auto-exemption); label applied by me at PR-open time as an authoring step, union-written and read back as [size/s, skip-changeset] with nothing stripped. Session note: repo-scoped REST is 403 for this seat ('GitHub access is not enabled for this session') while /rate_limit answers 15000/hr, so card body came from the zero-quota public-repo payload channel and every write went via MCP.",
  "tests": "dispatch-gates derived 38 families for this change set and --ran reconciled clean: '✓ dispatch-gates --ran: 38 derived famil(ies) accounted for — 38 run, 0 NOT-MEASURED' (script stderr confirms repo objectstack-ai/objectstack at commit daeeb32211, merge base bf45e62ce, three-dot). 36 families EXIT=0, each exit code captured by redirect-then-capture, never through a pipe. TWO ARE 'PREREQUISITE NOT MET' AND ARE REPORTED AS NOT MEASURED, NEITHER GREEN NOR RED: `pnpm check:dts-closure` EXIT=3 ('Not one of the 79 workspace package(s) under . has a dist/ ... This is NOT a pass and NOT a finding') and `pnpm check:dual-build-cjs-loads` EXIT=3 ('this gate reads built output, and some package has no dist/. Nothing was measured'). Both read dist/ and this diff changes no package source, so a closure build would have them measure something no path here can move; CI builds the closure and runs them — declared narrowing, not a silent skip. LOAD-BEARING AND GENUINELY MEASURED: `pnpm check:stall-guard-budget` EXIT=0 printed its own verdict line 'check-stall-guard-budget: OK (32 workflow file(s), 56 job(s), 611 step(s), 7 guard-wrapped step(s); every effective cap clears its job budget by at least one stall window)' and read the new value back — 'ci.yml:658 job `test` step Run this shard's tests · window 10m · cap 20m · budget 45m (job timeout-minutes) · slack 25m', slack up from 10m. `pnpm check:stall-guard-headroom` EXIT=0 but runs --self-test only ⇒ checker-health, NOT a PR verdict. Also EXIT=0 on the real diff: check:required-contexts, check:shard-attestation, check-ci-filter-parity, check:nul-bytes, check:pm-dispatch-gates, check:workflow-status-functions. CENSUS PROOF, keyed on job so line drift cannot hide a second change: 10 timeout-minutes occurrences before and 10 after; diffing the two job-value censuses yields exactly one changed line, `test: 30` becoming `test: 45`; filter 10, test-gate 10, temporal-conformance 30, dogfood 30, dogfood-verify 20, dogfood-gate 10, build-core 30, build-docs 30, console-pin 45 all unchanged. Whole diff is 1 file changed, 22 insertions(+), 1 deletion(-). YAML re-parses: jobs.test['timeout-minutes'] == 45 as int, matrix still shard [1,2,3,4,5,6], 10 jobs. Control-byte self-scan on the changed file found none (grep -naP over the control ranges, exit 1). No ablation is owed: this is a one-value config change with no new guard to prove failable.",
  "mcp_calls": "10 — MCP GitHub calls for the whole run (1 claim-comment read, 1 PR create, 1 PR body edit, 1 label write, 2 PR reads, 1 check-runs read, 2 job-log reads, 1 report comment). Card body and all git facts came from the zero-quota payload channel and local git; repo-scoped REST was 403 all run.",
  "open_questions": [],
  "out_of_scope_findings": []
}

One reading the card asked for, and it did not come out the way the card hoped — flagged rather than buried.

The PR is self-testing in principle, and all six Test Core shards on it went green. But this PR un-censors nothing, and no shard-timing number should be taken from it. Test Core (5/6) completed in 54s (02:05:18Z to 02:06:12Z) and its own log says why:

shard 5/6: 0/0 items (0 of them file-level slices of 0 package(s)), 0.0s predicted
No packages on this shard — nothing to test.
No test log — the test step did not get far enough to produce one.
Attested: test-5-of-6 ran to completion with every step green

A .github/workflows/**-only diff affects zero packages, so the partitioner scheduled zero items onto every shard. The 54s is fixed per-shard overhead — checkout, pnpm/Turbo cache restore, install, attestation — and nothing else. Reporting it as shard 5's duration would be the same category error the card is about: a reading of the harness rather than of the shard.

So the raise is still the instrument, but it only fires on a PR whose diff actually schedules packages onto shard 5. The first uncensored shard-5 duration will come from the next such PR to run against this workflow — that is where the rebalance work should take its input, and it is the reason this raise wants to stay in place until then.

CI state at report time: Test Core (aggregate), Lint & Repo Gates, three Type Check legs, Console Pin Gate, Temporal Conformance, Build Docs and Dogfood Verify CLI were still in_progress. Governed Surface Queue Guard, Check Changeset, Check PR Size, Build Core, all six Test Core (N/6) shards and all three Dogfood Regression Gate (N/3) shards had concluded success. Nothing red at report time. Convergence is the PM's read, not mine.

---

Generated by Claude Code