ci(test-shards): the cli slices ran 1132 s against 735 s predicted (×1.54) on Test Core 3/6 — refresh scripts/test-shard-timings.json (dated 2026-10-13) and re-derive the cli slice count under pin 3c

Filing gate ③, a follow-up owed by landing work. It is option B of the os-dev-report 6358271889 on #25372 (premise_still_valid: false), which domain:spec seat 1's review 6358287660 endorses as a separate card. Filed by the triage seat (session_01AavokzJ5DndAwitDXvKy4U) as part of the #25372 re-scope. It is the CI performance programme the maintainer set on 2026-10-23: 「我希望优化的是性能和效率速度」. ⛔ Not a claim.

Reader (who acts): the domain:devx execution seat, in the same lane as #25376 (shard count). It is one os-dev and one PR.

One-sentence problem

The 24.3-minute Test Core shard that #25372 cited was set by a cli slice, not by spec. The partition predicted 735 s for it, CI measured 1132 s (×1.54), and the dataset the prediction comes from is dated 2026-10-13.

Evidence (the dev's readings on `origin/main` `c898bd8`; the triage seat did not re-measure)

Run 39893808670, Test Core (3/6): the tests step ran from 03:16:20 to 03:40:37 (24.3 min). The cli slice leg took 18m52.554s (vitest Duration 1129.45 s). The shard's drift step prints "Heaviest overshoot: @objectstack/cli, predicted 735.3s, measured 1132.4s". Spec is not among that shard's overshoots.
Merge-group run 39894014103 aggregate: cli 2138.29 s across its 3 slices.
scripts/test-shard-timings.json: last refreshed by chore(ci): refresh the Test Core shard-timings dataset #23141 on 2026-10-13. The weekly shard-timings-refresh.yml scheduled run on 2026-10-19 completed success, but no refresh landed after 2026-10-13. The dev reads why: a refresh that opened no PR, or a PR that never merged.
Pin 3c (sliceCountProblems in scripts/partition-test-shards.mjs): it judges {cli: 3} with no problems on the committed dataset. A recount has to pass it on the refreshed dataset.

Ask

Land a refreshed scripts/test-shard-timings.json through the repo's own refresh lane (shard-timings-refresh.yml, from the hourly schedule run of ci.yml, never a push run, per the workflow's header and 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). Not by hand.
Re-derive the cli slice count on the refreshed dataset under pin 3c, which may give n = 4. Change FILE_SHARDED_PACKAGES only if the pin accepts the new count.
If the refresh lane is what failed since 2026-10-13, fix that first and say so in the PR.

.github/workflows/** is Tier H. If the fix needs a workflow change, the dev stops and reports it.

Done when

The dataset is no more than 7 days old at merge.
The cli slice count is the one pin 3c accepts on that dataset.
A merge-group Test Core run on the merged head shows the cli overshoot below ×1.2. The PR quotes the drift line.

Size: S–M.

Dedupe

Method: open objectstack-ai/objectstack issues labelled area:devpath, and the last 100 PRs, matched on shard, timing and slice. Siblings on other levers are #25372 (spec's single-file floor), #25376 (shard count), #25375 (DTS in the shard build), #25373 (cancelled push runs) and #25321 (repo-scan pins). None carries the cli drift or the stale dataset. New.

Generated by Claude Code

Claim: PM loop round 3 · 2026-10-23T12:14Z
Session: session_01V7SdBWbSZCQFPJD4ofKPPc (card thread, dispatched by seat session session_01LKHDniRQSWjRJyDtacwpEd)
Account: os-elon-musk
Branch: claude/project-thread-9ujbz0 (the harness-pinned branch for the card thread; it names no issue, so this claim is the identity record)
Worktree: objectstack-issue-25390
Domain: domain:devx
Seat: domain:devx#1 (seat post #6023)
File surface: scripts/test-shard-timings.json, written by the refresh lane (shard-timings-refresh.yml, from an hourly schedule run of ci.yml), never by hand; and FILE_SHARDED_PACKAGES in scripts/partition-test-shards.mjs for the cli slice count. A .github/workflows/** need is a stop-and-report. (stop on breach; explain in the report)
Container & model: S-M, mode:thread, model: default tier, effort: high
Clause-②: no
Metadata-protocol: none
Responsibility: Test Core shard balance | scripts/test-shard-timings.json + the cli slice count (pin 3c) | every Test Core run
Thread-read: none
Serial constraints cleared: none — #25375 and #25374 edit ci.yml, not these files.

Dispatched under the maintainer's 「同意」 to the director's ruling batch 1, item 5A (devx dispatches p1–p3). This card is also covered by the maintainer's 「同意，立卡」 for the CI performance audit cards, where the card says so. Dispatch order: /mnt/project-files/pm-devx-seat/dispatch-25390.md.

ACCEPT → PR #25409 · `domain:devx` seat 1 · `session_01LKHDniRQSWjRJyDtacwpEd` · 2026-10-23T13:11Z

This card's PR is the refresh lane's own bot PR, following the chore(ci): refresh the Test Core shard-timings dataset #23141 precedent. shard-timings-refresh.yml was dispatched at main through the relay (run 39951977591, success). It opened chore(ci): refresh the Test Core shard-timings dataset #25409 from 8 hourly schedule runs, the newest being 39943147947, with the dataset dated 2026-10-23. Nothing was written by hand.
Reviewed at head a3d5300ce7: one file, scripts/test-shard-timings.json (+255 / -434). The head is the bot commit 8114b394 plus a relay pr_update_branch merge of main, which adds nothing to the dataset.
Why no refresh landed after 2026-10-13: the 2026-10-19 refresh opened chore(ci): refresh the Test Core shard-timings dataset #24467. A GITHUB_TOKEN PR starts no CI, and nobody landed it. No workflow change is needed. chore(ci): refresh the Test Core shard-timings dataset #24467 is superseded by this PR and is to be closed once the maintainer gives the word.
Pin 3c on the new dataset: cli n=3 passes (966 s ≤ 1.3 × 803 s), n=2 is refused, and n=4 is refused. FILE_SHARDED_PACKAGES stays {cli: 3}, so there is no code change.
Prediction: the cli slice goes from 735.3 s to 966.3 s. The last 10 green merge-group readings (989.7–1121.0 s) drop from ×1.35–×1.52 to ×1.02–×1.16.
Gates: on the PR, 40 check runs; each one is success or an expected skipped (31 / 9). The skips are the path-filtered or opt-in lanes for a data-only diff.
Not governed: check-governed-merges.mjs --pr 25409 reports NOT governed.
Close-out: after landing, the card thread reads the next merge-group drift line for the cli slice (it needs < ×1.2) before this card closes.
Finding: an off-grid test-budget: disabled the refresh's regression diff. Filed as test-budget annotations off the 10 s grid (cli audit-mail-bodies archive test: 85) make the weekly shard-timings refresh refuse its readings; only the weekly generator checks the grid #25411.

Closing: Done-when met after #25409 (the 2026-10-23 shard-timings refresh) merged at 14:44Z. The new cli prediction is 966.3 s per slice.

Post-merge merge-group runs whose Test Core shard ran a cli slice. The cli figure is the slice leg's own vitest duration:

run · shard	cli slice	vs prediction	shard drift
39974273343 · 5/6	651.70 s	×0.67	OK 0.62×
39974273343 · 3/6	902.00 s	×0.93	OK 0.86×
39970969793 · 3/6	915.75 s	×0.95	OK 0.90×
39979035157 · 5/6	991.35 s	×1.03	OK 0.90×
39974536894 · 5/6	1022.97 s	×1.06	OK 1.08×
39974536894 · 4/6	1105.01 s	×1.14	OK 1.11×

Before, on the 10-13 dataset, the cli slice ran ×1.35 to ×1.52, with a drift WARN on 6 of 10 merge-group runs. After, every slice is under ×1.2, and no Test Core shard on these runs has a cli WARN. Pin 3c holds {cli: 3}. As a cross-check, the 15:01Z schedule run 39974593515 is red only on Console Pin Gate (#25438), not on Test Core. #24467 was closed as superseded by #25409. The off-grid budget finding stays open as #25411.

Generated by Claude Code

Navigation Menu

Repository navigation

ci(test-shards): the cli slices ran 1132 s against 735 s predicted (×1.54) on Test Core 3/6 — refresh `scripts/test-shard-timings.json` (dated 2026-10-13) and re-derive the cli slice count under pin 3c #25390

One-sentence problem

Evidence (the dev's readings on `origin/main` `c898bd8`; the triage seat did not re-measure)

Ask

Done when

Dedupe

ACCEPT → PR #25409 · `domain:devx` seat 1 · `session_01LKHDniRQSWjRJyDtacwpEd` · 2026-10-23T13:11Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

ci(test-shards): the cli slices ran 1132 s against 735 s predicted (×1.54) on Test Core 3/6 — refresh scripts/test-shard-timings.json (dated 2026-10-13) and re-derive the cli slice count under pin 3c #25390

Description

One-sentence problem

Evidence (the dev's readings on origin/main c898bd8; the triage seat did not re-measure)

Ask

Done when

Dedupe

Activity

objectstack-fleet commented on Oct 23, 2026

objectstack-fleet commented on Oct 23, 2026

ACCEPT → PR #25409 · domain:devx seat 1 · session_01LKHDniRQSWjRJyDtacwpEd · 2026-10-23T13:11Z

objectstack-fleet commented on Oct 23, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

ci(test-shards): the cli slices ran 1132 s against 735 s predicted (×1.54) on Test Core 3/6 — refresh `scripts/test-shard-timings.json` (dated 2026-10-13) and re-derive the cli slice count under pin 3c #25390

Evidence (the dev's readings on `origin/main` `c898bd8`; the triage seat did not re-measure)

ACCEPT → PR #25409 · `domain:devx` seat 1 · `session_01LKHDniRQSWjRJyDtacwpEd` · 2026-10-23T13:11Z