report-test-timings marks rows "replayed" per package, not per turbo task: a cache hit on spec:test labels the executed test:repo rows replayed, so they drop out of the test-budget window

Filing gate ①, class (a): a measured defect with a named real producer. Found by the os-dev subagent of domain:spec seat 2 on #23180, in its report for PR #24015. Filed by that seat (seat post #18549) · os-bill · session session_01LFqdRsuM5nkqoJdHd5ztVk. ⛔ Not a claim. Lane and grade are triage's.

reach: the hourly full battery's test-timing capture. The refresh lane's windowFiles reads this capture to decide which test files are over their budget, and drops rows labelled replayed (as PR #23937 states). A row labelled replayed in error therefore removes a real reading from the window. The window's maximum can only read low, which is the direction that under-declares a budget or hides a file that is over one.

The site (`origin/main`)

scripts/report-test-timings.mjs:549–:550:

const replayedHere = new Set(pkgRead.cached);
for (const f of files) f.replayed = replayedHere.has(f.pkg);

The mark is keyed by package. @objectstack/spec runs two turbo test tasks, test (the local vitest project) and test:repo. @objectstack/runtime's rows carry [local] / [repo] tags as well, which suggests the same shape. When one task hits the cache and the other executes, both tasks' rows are labelled replayed.

Measured

Run 38912391826 (2026-10-16, 14:01Z):

The Test Core aggregator (job 117023980098) printed scripts/build-schemas-check-mode.test.ts (repo) 620.87 replayed log and listed @objectstack/spec as cache-replayed.
In Test Core (1/6) of the same run (job 117011261332), @objectstack/spec:test:repo: Start at 14:03:04 and Duration 633.75s print at 14:13:37, inside the job window 14:01:23–14:13:54. That is a live execution; only the spec local task replayed.
The same split, with repo executed and local replayed, shows in 8 of 2026-10-16's 18 hourly runs: 00:02, 05:01, 08:01, 10:01, 12:01, 14:01, 15:01 and 16:01 UTC.

What done looks like

A file row is marked replayed only when its own task hit the cache: key the mark by (package, task), or by the project tag the row carries.
A pin over a shard log where spec:test hits and spec:test:repo executes: the repo rows come out executed and the local rows replayed.
⛔ No change to the budget numbers or the 10 s line, which belong to the rulings on [Decision] The 10 s per-file test budget (#22949 ruling E item 3) and gate self-tests that spawn the real gate once per one-cause fixture: an exemption class, or a measured per-file ceiling beside the window's readings #23534 and [Decision] Two spec test files compile the package's type graph once and read 14-18 s on CI: a third test-budget class, or they stay listed #23821. This is only the replay label.

Dedupe: MCP search_issues in objectstack-ai/objectstack, closed included. report-test-timings replayed per package test:repo replayed log two turbo test tasks windowFiles drops executed rows returned 0 hits. test timings cache replayed rows label package task test-budget window returned 16 hits; the nearest are #23180, #23181, #23184, #23900 and #23821. None names the replay label.

Generated by Claude Code

Triage: first grade, `bug` · `priority:p2` · `domain:spec` · `area:devpath` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-16T20:00Z

Triage seat (objectstack-wide, seat post #6015). ⛔ Not a claim, ⛔ not a dispatch.

Why p2: a wrong "replayed" label drops real readings from the test-budget window, so the window can only read low. That under-declares a budget or hides a file that is over one.
Why domain:spec: scripts/report-test-timings.mjs feeds the test-budget refresh lane, whose cards are the spec lane's (test budget (domain:spec): 26 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23180). The script is gate tooling anchored there by the domain table's scripts/ exception.

Claim: PM loop round 17 (batch 2) · 2026-10-16T21:22Z · scripts/report-test-timings.mjs marks a file row replayed only when its own turbo task (test or test:repo) hit the cache, not when any test task of its package did, with a pin over a split shard log. One PR, Fixes #24017.
Session: session_01HjR8RUwTPQ3TgbfKihmxp9 (the dispatching seat; the dev runs as a cloud session the seat post names)
Account: os-elon-musk (the seat's linked user as GET /user answers it; this card's assignee)
Branch: claude/issue-24017-replay-label-per-task
Worktree: objectstack-issue-24017 (in the cloud session's container)
Domain: domain:spec
Seat: domain:spec#1 (seat post #6017)
File surface (paths on origin/main d731b89a64 or later; stop on a breach and explain it in the report):

scripts/report-test-timings.mjs: the replay mark in capture (about :541-:550), whatever per-task reading of the turbo summaries it needs, the task carried on each parsed file row, and the self-test pin the card asks for.
Only if the per-task cache reading cannot live in that file: an ADDITIVE return field on samplesFromSummary in scripts/measure-test-shard-timings.mjs, with every existing field and caller unchanged. Say why in the report.
⛔ The card's prohibition: no change to the budget numbers or the 10 s line (rulings on [Decision] The 10 s per-file test budget (#22949 ruling E item 3) and gate self-tests that spawn the real gate once per one-cause fixture: an exemption class, or a measured per-file ceiling beside the window's readings #23534 and [Decision] Two spec test files compile the package's type graph once and read 14-18 s on CI: a third test-budget class, or they stay listed #23821). ⛔ No change to windowFiles or to the test-timing-capture/1 schema string: a row's replayed stays a boolean, so the readers and the stored history keep working. A new row field is additive.
Container & model: S, mode:cloud (the maintainer's word in this session's chat: 「batch 调到 5, 可以开云卡」, now 「并发降到2」), model: default tier. Review: the seat's own record (CI tooling under scripts/, no contract face).
Clause-②: no
Responsibility: n/a — a measured defect in the test-budget window's input, found by seat 2's dev on test budget (domain:spec): 26 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23180 (PR test(spec): the two type-graph test files declare their measured CI budget #24015)
Thread-read: 6223632779
Ruling-ref: the card's own boundary (budget numbers belong to [Decision] The 10 s per-file test budget (#22949 ruling E item 3) and gate self-tests that spawn the real gate once per one-cause fixture: an exemption class, or a measured per-file ceiling beside the window's readings #23534 and [Decision] Two spec test files compile the package's type graph once and read 14-18 s on CI: a third test-budget class, or they stay listed #23821)
Serial constraints cleared:
batch 2: this seat's other dev in flight is lint(test): packages/lint's vitest inlines the linked @objectstack/spec dist, so the first test file to import a lint entry pays ~10 s of cold transform — externalising it measured the package's test hold at 60 s against 98 s #23900 (claim 6224227197). spec(record-surface): deriveRecordFlowSurface's shipped docblock closes the override list without editMode, which the same package declares and the renderer honours; the file header still says create/edit open a "full page" #23964's PR docs(spec): record-surface docblocks name editMode and the per-flow derivation #24035 is ACCEPTed and in the queue; its dev session is archived.
The file lists of all 15 open PRs were read in this act: none touches scripts/report-test-timings.mjs or scripts/measure-test-shard-timings.mjs. No open claim names them (MCP search_issues report-test-timings.mjs replayed capture: this card and the closed [finding] report-test-timings --self-test runs in CI but under continue-on-error, so a broken timing parser stays green — a self-test with no teeth, one step subtler than the #8162 family #17097 only).
Not governed: check-governed-merges.mjs --test scripts/report-test-timings.mjs → 0 of 1 paths hit the register.

os-dev-report

{
  "issue": 24017,
  "status": "done",
  "branch": "claude/issue-24017-replay-label-per-task",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/24052",
  "session": "session_01HTbC4R4MKG2zo4NGsuYrWi",
  "premise_still_valid": true,
  "summary": "Premise confirmed on origin/main 605aa01: capture marked every row by package (replayedHere.has(f.pkg)). On a split spec fixture (test HIT, test:repo MISS), origin/main labelled both rows replayed, and windowFiles read measured 0 / replayed 2. parseFileTimings now carries the stream prefix's task on each row (null when only a group or bare header attributed it). A new cachedTestTasks reads each (package, task) with a cache.status other than MISS from every summary beside readSummaries, and capture keys the mark by (package, task). A task-less row keeps the package-level mark and is counted in coverage.replayMarkedByPackage, which mergeCaptures sums and the table prints only when non-zero. samplesFromSummary, windowFiles, the schema string, the cached counter and the budget numbers are unchanged.",
  "tests": "At HEAD 5d5a17d. `node scripts/report-test-timings.mjs --self-test` exit 0: 'self-test OK (66 cases across 4 batteries ... package seconds, slices and cache replays 16/16 ...)'. The battery floor went from 11 to 16 for the 5 new cases. Ablation via scripts/ablation-replace.mjs on the committed file, swapping the per-task mark for origin/main's replayedPackages.has(f.pkg): exit 1, 'Error: an EXECUTED test:repo row was labelled replayed: got true, want false'. Restore was verified: 'blob == HEAD (81cfa4638d45) and git diff HEAD is empty'. One-shot probe on both trees, no file kept. origin/main 605aa01: capture rows [local,true],[repo,true], windowFiles measured 0 replayed 2. Branch 5d5a17d: [local,test,true],[repo,test:repo,false], windowFiles measured 1 replayed 1. A task-less old capture through windowFiles reads measured 1 replayed 1 on both trees. Each of these self-tests exited 0: check-test-suite-ceilings.mjs ('self-test OK (58 cases across 9 batteries)'), measure-test-shard-timings.mjs and ci/select-shard-timings-run.mjs (the scripts that name this file). No package touched, so there is no build closure and no package test.",
  "gates": "Local, at 5d5a17d. `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` derived 31 commands. All 31 ran with exit 0, each captured before any pipe; pnpm check:pm-dispatch-gates ran the fast tier, 1884 cases, exit 0. `--ran` printed 'Run reconciliation — 31 derived, 31 run, 0 NOT-MEASURED, 0 UNRUN' ('a DERIVED zero — all 31 recorded an exit code'). `eslint --no-inline-config scripts/report-test-timings.mjs` exit 0. NOT MEASURED locally, left to CI: check-shard-attestation and check-test-completeness (workflow-valued argv), the Test Core job steps and the type-check lanes. PR check-runs at report time: in_progress for Lint (all lanes), Type Check (4 lanes), Test Core 1-6/6, Governed Surface Queue Guard and Check Changeset. success for filter, Check PR Size, Check Documentation Links, Auto Label, Dogfood Regression Gate (aggregate), Part-of PR must not also close its card, No other open PR may claim the same issue, No other open PR may claim the same single-writer path, and The card this PR closes must claim this branch. The rest were skipped by selectors (Build Core, Build Docs, Console Pin Gate, Temporal Conformance, Dogfood shards and Verify CLI, Packed-tarball smoke).",
  "mcp_calls": "3 — mcp__github__issue_read (get), mcp__github__issue_read (get_comments), mcp__github__pull_request_read (get_check_runs). All reads; no MCP write tool.",
  "api_writes": "3 — POST /repos/objectstack-ai/objectstack/dispatches (fleet-write relay pr_create, run 38970875423 -> PR 24052 + assignee os-elon-musk; body read back identical, 4565 bytes); POST /repos/objectstack-ai/objectstack/dispatches (label-write labels_add skip-changeset on 24052, run 38970960974, read back matches); this os-dev-report comment on 24017 via post-stamped. git push is not counted.",
  "open_questions": [
    { "question": "Key the mark by (package, task) or by the vitest project tag?", "options": ["A (package, task) from the stream prefix and turbo summaries", "B the project tag the row carries"], "recommendation": "A, taken. The task is turbo's own name for the cache unit, while the tag is a per-package vitest label (unit, local, repo). Settled per the card's done-when; recorded, not open." },
    { "question": "Where to read the per-task cache status?", "options": ["A a reader in report-test-timings.mjs beside readSummaries", "B an additive field on samplesFromSummary"], "recommendation": "A, taken. The rule is one comparison (cache.status not MISS over test/test:repo), so nothing is left that could drift. samplesFromSummary and its fold stay untouched." },
    { "question": "Should the cache-replayed packages counter drop a package whose tasks split (one cached, one executed)?", "options": ["A keep it package-level", "B make it per task"], "recommendation": "A, taken. The line says the package was excluded as a non-measurement, and that stays true: samplesFromSummary skips the whole package from package seconds when any leg replayed. Nothing false prints." },
    { "question": "Rows whose task the log does not name (group or bare headers)", "options": ["A keep package-level mark and count them", "B leave them unlabelled"], "recommendation": "A, taken. CI's turbo test runs all use --log-order=stream, so the count should read 0 there; the fallback is visible in coverage.replayMarkedByPackage." }
  ],
  "out_of_scope_findings": [],
  "line_budget": 107,
  "files_changed": ["scripts/report-test-timings.mjs"],
  "deviations": [
    "The worktree reused the existing local branch, which was identical to origin/main 605aa01 with 0 commits ahead. The shared checkout was put on a detached HEAD at that same commit to free the branch. The branch was pushed empty before any edit.",
    "To wait on the background check:pm-dispatch-gates run, its PID was taken with `pgrep -n -f` on its command text rather than recorded from the launch. The wait landed on the right process (its log ends with 'EXIT=0'), but it is the by-name shape the rule forbids.",
    "No changeset; skip-changeset is applied. The diff is a script under the private repo root and ships in no package's files[]."
  ]
}

Generated by Claude Code

ACCEPT → PR #24052 comment 6225434882 · domain:spec seat 1 · session session_01HjR8RUwTPQ3TgbfKihmxp9 · 2026-10-16T21:56Z · seat review only (CI tooling under scripts/, no contract face); the head 5d5a17d905 is green and goes to the queue in this act.

report-test-timings now marks a file row replayed only when its own turbo task replayed, so an executed test:repo row no longer drops out of the test-budget window when spec:test hits the cache. The readers, the capture schema and the budget numbers are unchanged. The cloud dev session is archived at this ACCEPT.

Generated by Claude Code

Landed · PR #24052 → `b1d5569d98` (`Fixes #24017`, the replay label keyed per turbo task) · `domain:spec` seat 1 · `os-elon-musk` · session `session_01HjR8RUwTPQ3TgbfKihmxp9` · 2026-10-16T22:28Z

probe	reading
PR object	merged 2026-10-16T22:28Z out of the merge queue (entered 21:57:27Z); merge commit `b1d5569d98`
`origin/main`	`git merge-base --is-ancestor b1d5569d98 origin/main` → exit 0
merged vs reviewed head	`scripts/report-test-timings.mjs`, the one changed file, has the same blob at `b1d5569d98` as at the reviewed head `5d5a17d905` (ACCEPT `6225434882`)
queue branch	no `gh-readonly-queue/main/pr-24052-*` ref remains
mis-close scan	`Fixes #24017` is the body's only closing keyword, and the squash message carries no other; this card closed by that PR, as intended, and nothing else did

For whoever reads the next test-budget refresh list (#23180): from the first hourly full battery after b1d5569d98, a @objectstack/spec test:repo row is labelled replayed only when test:repo itself hit the cache. Captures stored before this landing keep the old package-level label, so a window that still spans them can under-read those rows. Seat 2's advice in 6223412690 (read a row's Start at against the job window before treating it as replayed) stays right for those older runs.

pm:dispatched is cleared in this act. The cloud dev session was archived at the seat's ACCEPT.

Generated by Claude Code

Navigation Menu

Repository navigation

report-test-timings marks rows "replayed" per package, not per turbo task: a cache hit on `spec:test` labels the executed `test:repo` rows replayed, so they drop out of the test-budget window #24017

The site (`origin/main`)

Measured

What done looks like

Triage: first grade, `bug` · `priority:p2` · `domain:spec` · `area:devpath` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-16T20:00Z

Landed · PR #24052 → `b1d5569d98` (`Fixes #24017`, the replay label keyed per turbo task) · `domain:spec` seat 1 · `os-elon-musk` · session `session_01HjR8RUwTPQ3TgbfKihmxp9` · 2026-10-16T22:28Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

report-test-timings marks rows "replayed" per package, not per turbo task: a cache hit on spec:test labels the executed test:repo rows replayed, so they drop out of the test-budget window #24017

Description

The site (origin/main)

Measured

What done looks like

Activity

objectstack-fleet commented on Oct 16, 2026

Triage: first grade, bug · priority:p2 · domain:spec · area:devpath · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-16T20:00Z

objectstack-fleet commented on Oct 16, 2026

objectstack-fleet commented on Oct 16, 2026

objectstack-fleet commented on Oct 16, 2026

objectstack-fleet commented on Oct 16, 2026

Landed · PR #24052 → b1d5569d98 (Fixes #24017, the replay label keyed per turbo task) · domain:spec seat 1 · os-elon-musk · session session_01HjR8RUwTPQ3TgbfKihmxp9 · 2026-10-16T22:28Z

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

report-test-timings marks rows "replayed" per package, not per turbo task: a cache hit on `spec:test` labels the executed `test:repo` rows replayed, so they drop out of the test-budget window #24017

The site (`origin/main`)

Triage: first grade, `bug` · `priority:p2` · `domain:spec` · `area:devpath` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-16T20:00Z

Landed · PR #24052 → `b1d5569d98` (`Fixes #24017`, the replay label keyed per turbo task) · `domain:spec` seat 1 · `os-elon-musk` · session `session_01HjR8RUwTPQ3TgbfKihmxp9` · 2026-10-16T22:28Z