Repository navigation
dogfood: activity-text-predicate.dogfood.test.ts times out at vitest's 5000 ms default under load (Rerun Safety pass 1, 10-02 and 10-14); the file sums 12.9 s unloaded with no declared budget #23977
Description
Filing gate ①: a measured CI defect at a public door. reach: the Rerun Safety nightly on main. Its pass 1 went red on this file on two readable nights. That makes it the "dogfood" half of #23880's pass-1 failures (the "cli" half was turbo stopping cli after dogfood failed).
Filed by the domain:devx seat 2 (#20163) · os-sales · session_016PVjjuVPte5XEJvaQx1E4t. It is the split card from the #23880 dev's report (6218645603, out-of-scope finding 1), filed by the seat under os-dev.md. ⛔ Not a claim; the lane and grade are triage's.
Reader who acts: the dogfood lane, for packages/qa/dogfood/test/activity-text-predicate.dogfood.test.ts.
Measured (#23880's dev)
- CI: Rerun Safety pass 1,
Error: Test timed out in 5000ms.in this file. It happened on 10-02 (run 36962761984, log line 465) and on 10-14 (run 38509291145, log line 470). - Unloaded, on an idle 4-core container at
98ba1bf1d2:- the file's 22 tests sum to 12.9 s;
- the test
control: the unmasking reader filters and groups by every column of a pinned parentalone takes 3377 ms, 68 % of vitest's 5000 ms default; - the sibling files
activity-parent-read-gate(1063 ms) andaudit-log-parent-read-gate(974 ms) pass.
- Budget: the file is over the 10 s per-file line (
AGENTS.md; ruling record6181973363), and it declares no// test-budget:. - The hourly battery: the hourly
CI(Dogfood 3 shards) stayed green in the same window. Under Rerun Safety's one-runner unsharded load, the 3.4 s test crosses 5 s.
Done when
The file runs under the 10 s line, with the fix in the test or in what it waits on. Alternatively, it carries a measured budget in a class a ruling names. ⛔ No timeout raise, no deletion without a named duplicate, and no move to a slower or skipped project (ruling 6181973363 item 5). The control test then no longer sits within 2× of its timeout under load.
Dedupe
MCP search_issues on this repository, closed included, at this filing: activity-text-predicate dogfood test timed out 5000ms file over 10 s unmasking reader control gave 7 hits. #23185 (test budget, devx, closed), #19768 and #19631 (other files' 5000 ms timeouts, closed), #22400, #22653, #15240 and #20820 are unrelated. None is this file.
Dedupe words: activity-text-predicate timeout; dogfood file over 10 s; unmasking reader control test 5000ms.
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 16, 2026
Triage: first grade,
bug·priority:p1·domain:cli·area:devpath·pm:queue· triage seat ·session_01AavokzJ5DndAwitDXvKy4U· 2026-10-16T16:01ZTriage seat (objectstack-wide, seat post #6015). ⛔ Not a claim, ⛔ not a dispatch.
main. The maintainer's rule 「平台测试清单反复变红，肯定要修啊」 applies.domain:cli:packages/qaisdomain:cliby the domain table.6181973363. Declare a budget only where the wait is real. ⛔ Raising vitest's timeout alone does not fix it.objectstack-fleet commentedon Oct 16, 2026
Claim: PM loop round 1 (#23977:
activity-text-predicate.dogfood.test.tsruns under the 10 s per-file line, with its slowcontrolcase far from vitest's 5000 ms default under load, by fixing what the file waits on, with no timeout raise, no deletion and no move) · 2026-10-16T17:51ZSession:
session_018t6a5rvqPRvEDMCVVW9kYTAccount:
os-tesla(the seat's linked user asGET /useranswers it; the card's assignee from this act)Branch:
claude/issue-23977-activity-text-predicate-budgetWorktree:
objectstack-issue-23977Domain:
domain:cliSeat:
domain:cli#2(seat post #22648)Scope:
batch3.labelis served withlabel= its qualified name (recheck_gate.default), invented by the read-side translator, so the console view tab shows the machine name #23967. Malformed JSON request body is silently parsed as {}: a truncated POST /data/:object/query answers 200 with the whole table, and a truncated PATCH answers 200 with an empty update #23914's dev run has ended; its PR fix(plugin-hono-server,hono,cloud-connection)!: refuse an unparseable request body with 400 INVALID_REQUEST instead of reading it as {} #24005 is in its contract review.labelis served withlabel= its qualified name (recheck_gate.default), invented by the read-side translator, so the console view tab shows the machine name #23967's suites hold it; the queue is empty).Order, read at this act:
p0,pm:blockingortarget:card of this lane is inpm:queue. This card is the lane's only queuedp1.6220729537records the maintainer's word to that seat, 「当前任务处理完，后续只派p3卡」, so this card is this seat's.File surface (read on
origin/maind132140ebc; stop on a breach and explain it in the report):packages/qa/dogfood/test/activity-text-predicate.dogfood.test.tsonly (500 lines). The fix goes to what the file waits on.beforeAllboot (about:241) is clocked as a hook.controlcase (about:470) makes about 30 sequential list calls.AGENTS.md:124's convention, through what@objectstack/verifyalready exports.6181973363item 5).// test-budget:line: a singlebeforeAllboot is neither class that ruling names. If the measurement shows the wait is real and of a named class, stop and report.@objectstack/verify(boot-warm-up.testkit.tsis not exported, and exporting it widens a published package).class:andreach:instead of fixing it here..github/workflows/rerun-safety-nightly.ymlor its scripts (PR ci(rerun-safety): a truthful verdict, passes bounded by the job's time, and a card on a red night #23975, ci(rerun-safety-nightly): red three nights running on main and nobody was told — pass 1 fails @objectstack/cli and @objectstack/dogfood, pass 2 adds a different package set each night, 10-16 hit the 170-minute timeout #23880).Container & model:
S(one test file, measured A/B under load),mode:subagent,model: TIER_DEFAULT(dispatch-gates --tier: no path-derived mandate)Clause-②: no
Basis for the Clause-② line: a test file only. No published package's
files[]reaches it (@objectstack/dogfoodis private), and no accept set or public surface moves.Responsibility:
the dogfood test file in packages/qa (this lane) | the Rerun Safety nightly (#23880, PR #23975), which runs it unsharded on one runner, and the refresh lane's 10 s line | main's nightly, red on pass 1 on two readable nightsThread-read: 6219463123
Serial constraints cleared: open PRs read in this act.
packages/verify/src/index.tsorboot-warm-up.testkit.ts.Generated by Claude Code
objectstack-fleet commentedon Oct 16, 2026
os-dev-report
{ "issue": 23977, "status": "rework", "branch": "claude/issue-23977-activity-text-predicate-budget", "pr": "https://github.com/objectstack-ai/objectstack/pull/24026", "session": "session_018t6a5rvqPRvEDMCVVW9kYT", "premise_still_valid": true, "summary": "Part of #23977, not Fixes: one of the card's two halves is met. MET: the control case no longer sits within 2x of its timeout under load. Base timed out at 5000 ms on 2 of 3 loaded runs here (control 5028/4892/5031 ms, the nightly's red reproduced). Head splits it into 4 cases with every probe and assertion moved intact, and the slowest part took 2488/1906/1629 ms loaded. NOT MET: the file is not under the 10 s line. The module-scope warm-up (bootStack, already exported) took the hook from 5.8-6.8 s to 3.1-3.4 s, but the file's unloaded median only fell from 16022 to 12247 ms. What remains is 141 CPU-bound requests at a median of 60-74 ms each, and 62-72 % of each request is one platform cost: the REST API-exposure gate re-validates every object's metadata on every data request (finding 1). That is platform code the claim holds off. Assumption 2 is falsified: concurrent probes were no faster (2137 ms vs 2016 ms for 33), so no concurrency was added. Recommended routing: the seat files finding 1 in the rest/metadata-protocol lane, #23977 gains a Blocked-by: line naming it and stays open, and the file is re-measured after it lands (projected roughly 6-8 s by the measured ratio). Worktree removed after the PR opened.", "tests": "All at head 0c7592a22, through os-verify-lock, shared 4-vCPU box. (1) pnpm --filter @objectstack/dogfood exec vitest run --maxWorkers=2 test/activity-text-predicate.dogfood.test.ts test/activity-parent-read-gate.dogfood.test.ts test/audit-log-parent-read-gate.dogfood.test.ts => VITEST_EXIT=0, 'Test Files 3 passed (3) / Tests 39 passed (39)'; the file line read 25 tests 12514ms. (2) pnpm --filter @objectstack/dogfood typecheck (tsc --noEmit) => exit 0; tsc --listFilesOnly lists the changed file once (308 dogfood test files in the program). (3) A/B: base = untracked copy blob bb7cbee56a89 (identical to 61a680956), head = 0c7592a22. Isolated, interleaved B-H/H-B/B-H, file line of the default reporter (agent variables unset). Unloaded file ms: base 17680/16022/15338 (median 16022), head 12059/12731/12247 (median 12247). Loaded with 7 CPU hogs (about 2.2-2.7x): base 42696(1 failed)/34224/32896(1 failed) (median 34224), head 37072/29319/26863 (median 29319). Control: base unloaded 2572/2493/2398, loaded 5028/4892/5031 (both failures are 'Error: Test timed out in 5000ms.'); head slowest part unloaded 779/804/832, loaded 2488/1906/1629. (4) Ablation, one-time, prediction written first ('3 failed / 22 passed: the three per-class control parts on expect(snapHit.rows).toBe(1)'). node scripts/ablation-replace.mjs wrap mode changed the per-class snapHit filter to NOMATCH. Mutation landed: anchor x1 -> x0, blob 8b5eb87856f9 -> 2ac1463e3d0a. vitest exit=1, '25 tests | 3 failed', each 'expected +0 to be 1'. Restore: blob after restore 8b5eb87856f9 == HEAD, git diff HEAD empty, ABLATION_TOOL_EXIT=0. No build was involved: the test file is source. (5) Assertions kept: base's 12 it/describe title literals hash 887e5a912c4fe675 (sha256 prefix, file order), and head's 14 minus the 2 added part titles hash the same. expect( 40 = 40.", "gates": "Local subset at 0c7592a22. pnpm exec eslint --no-inline-config packages/qa/dogfood/test/activity-text-predicate.dogfood.test.ts => ESLINT_EXIT=0. node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no paths; 1 path vs merge base 61a680956) derived 53 commands; each was run with its exit captured before any pipe. 52 => exit 0. pnpm check:dual-build-cjs-loads => exit 3, PREREQUISITE NOT MET (no dist for studio, client-react, embedder-openai, knowledge-*, organizations, service-cluster-redis, service-knowledge), so NOT MEASURED locally. --ran with ':: exit N' per line => RAN_EXIT=0, '53 derived famil(ies) accounted for — 52 run, 1 NOT-MEASURED (1 DERIVED from a recorded exit 3)'. CI on PR #24026 at report time (head 0c7592a22240): 'No other open PR may claim the same issue' completed/success; 'Packed-tarball smoke (opt-in)' completed/skipped; 22 more in_progress, among them Lint & Repo Gates' lanes (Lint · ESLint, Lint · repo & tooling gates, Lint · docs & code guards, Lint · CI, release & engine gates, Lint · PM dispatch-gates, Lint · scoped ×3), Type Check · workspace/source gates/consumer gates/debt ledger, Check Changeset, Governed Surface Queue Guard, 'Part-of PR must not also close its card', 'The card this PR closes must claim this branch'. Test Core and Build Core were not yet listed. Dogfood is not a PR-CI family; it runs on main's full battery.", "deviations": [ "PR first line is 'Part of #23977', not the closing-keyword line the dispatch asked for: the 10 s half is unmet, and os-dev.md's DoD says a merge that must not close the card uses Part of.", "Assumption 2 is falsified, so no concurrency was added. The dispatch's 'concurrency ablation' became an ablation of the split: one moved probe returns a wrong answer and the case goes red.", "The control case is split into 4 its, as the dispatch allows for a timeout. So the head title-line set has 2 added titles; every base title is kept verbatim.", "Local scope: the whole dogfood suite was not run (the file plus its 2 named siblings only, per the dispatch). test:repo was not run (no repo-project file touched). check:dual-build-cjs-loads is NOT MEASURED locally (exit 3).", "Load emulation was 7 CPU hogs, about 2.2-2.7x slower, which is harsher than the card's about 1.5x. The box is shared, so other agents' unlocked work ran too.", "Environment: about an hour into the run the container's disk hit 0 bytes free. Two probe runs died on ENOSPC (one at module load, one on the tool's own output); they were re-run and are not counted. The dominant consumer then was the shared scratchpad's issue-23805 directory (21G, last written 08:49Z). I deleted nothing outside my own scratch; space came back (22G free) without my action.", "Commit trailers use AGENTS.md's model-free pair (Claude-Session + 'Co-authored-by: Claude'), not the harness reminder's model-named Co-Authored-By.", "The branch is 7 commits behind origin/main and was not merged: none of those commits touch the file, armed.ts, catalog-permission-set.ts, packages/qa/dogfood/vitest.config.ts or packages/verify/src/harness.ts, and git merge-tree is clean.", "No labels written (the dispatch said label-write: none). The seat applies skip-changeset (Clause-②: no, private package)." ], "files_changed": [ "packages/qa/dogfood/test/activity-text-predicate.dogfood.test.ts" ], "mcp_calls": "0 — none", "api_writes": "2 — both through the fleet-write relay (repository_dispatch POST /repos/objectstack-ai/objectstack/dispatches): (1) pr_create, run 38953684601 = POST /repos/objectstack-ai/objectstack/pulls (draft) + POST /repos/objectstack-ai/objectstack/issues/24026/assignees ['os-tesla'], read-back 9411 bytes sent = 9411 stored; (2) this os-dev-report comment = POST /repos/objectstack-ai/objectstack/issues/23977/comments via scripts/pm/post-stamped.mjs. Plus 2 git pushes of the branch (not REST).", "open_questions": [], "out_of_scope_findings": [ "class: a · reach: a measured red at a CI door, the Rerun Safety nightly on main (this card's own reach), plus every authenticated REST data request in production paying the same cost · evidence: every /data/OBJECT read and query request enters enforceApiAccess (packages/rest/src/rest-server.ts) to read ONE object's enable flags. That calls loadObjectItems (packages/rest/src/meta-item-read-gate.ts:409) -> protocol getMetaItems({type:'object'}) -> readFlattenedMetaItems(..., 'served') (packages/metadata-protocol/src/protocol.ts:9723) -> decorateMetadataItems -> computeMetadataDiagnostics (packages/metadata-protocol/src/metadata-diagnostics.ts:55/168), a full Zod safeParse of every registered object's schema, on every request. Measured at 0c7592a22 on this file's boot: an inspector CPU profile (200 us sampling) of the control's 33 list calls put 1311 of 2841 ms sampled inside readFlattenedMetaItems, 1244 ms of it inside computeMetadataDiagnostics. A measurement-only probe memoizing protocol.getMetaItems for type 'object' cut the same 33 calls from 2376/2255 ms to 865/668 ms (-62 to -72 %). By that ratio it is roughly 5-7 s of this file's 9-11 s of tests, and the remaining half of #23977. File in the rest / metadata-protocol lane; #23977 gets a 'Blocked-by:' line naming it, with a back-link · dedupe words: enforceApiAccess per-request metadata diagnostics; loadObjectItems listMetaItems object every request; computeMetadataDiagnostics safeParse REST data request latency; api-exposure gate object list cost" ] }Generated by Claude Code
objectstack-fleet commentedon Oct 16, 2026
Blocked-by: #24031
Release:
domain:cliseat 2 ·os-tesla· sessionsession_018t6a5rvqPRvEDMCVVW9kYT· 2026-10-16T19:55Z · cause: PR #24026 landed (Part of #23977) · destination:pm:blockedon #24031, unassignedLanded: PR #24026 →
2b274600b1(Part of #23977)merged: true, merge commit2b274600b1, merged 2026-10-16T19:53Z through the merge queue (enqueued 2026-10-16T19:34Z)origin/main2b274600b1is an ancestor oforigin/main: 1 file, +66 / −26, as reviewed at0c7592a222gh-readonly-queue/main/*pr-24026*head remains6223151728(the seat's record; test-only, private package).skip-changesetwas applied by this seat.controlcase no longer sits within 2× of its timeout under load. Base ran 5028 / 4892 / 5031 ms loaded and timed out on 2 of 3 runs; the head's slowest part ran 2488 / 1906 / 1629 ms. A module-scope warm-up pays the first boot outside the clock. Every assertion moved intact; there was no timeout raise and no skip./datarequest. That is filed as rest: every /data request's API-exposure gate reads EVERY object's metadata (with a full schema diagnostic) to check ONE object's enable flags — 62-72% of a data request's time, measured #24031, and this card waits on it.main's first full battery for this file, which does not run on a PR.pm:dispatchedbecomespm:blockedand the assignee is cleared.Generated by Claude Code
objectstack-fleet commentedon Oct 17, 2026
Unlock:
pm:blocked→pm:queue·domain:cliseat 2 ·os-tesla· sessionsession_018t6a5rvqPRvEDMCVVW9kYT· 2026-10-17T02:05ZBlocked-by: #24031is met. PR fix(rest): the API-exposure gate reads the one object it checks, by name, instead of every object's metadata on each /data request #24093 landed asc1c0a90942at 2026-10-17T02:03Z. On every/datarequest, the REST API-exposure gate now reads the one object it judges instead of the whole decorated object list. Measured on the showcase: 33 list calls 2421 ms → 647 ms.6223543949:packages/qa/dogfood/test/activity-text-predicate.dogfood.test.tsonmainat or afterc1c0a90942;pm:queue, unassigned, in the stroke after this comment.Generated by Claude Code
objectstack-fleet commentedon Oct 17, 2026
Claim: PM loop round 1, serial, p1/p2 only (the maintainer to this seat, verbatim 「串行派发，只处理 p1 p2」). #24040's PR #24169 waits only in the merge queue (
added_to_merge_queue, read in this act). Label order read in this act: notarget:board item is left inpm:queuefor this lane, and #23977 is its only queued p1. #23977: re-measureactivity-text-predicate.dogfood.test.tsonmainafter #24031 (c1c0a90942); close on a reading under the 10 s line, or take the next slice · 2026-10-17T13:39ZSession:
session_01BVm1pAXVXQ3PjC42j1kPFfAccount:
os-sales(the seat's linked user asGET /useranswers it; always the card's assignee)Branch:
claude/issue-23977-activity-text-predicate-remeasureWorktree:
objectstack-issue-23977Domain:
domain:cliSeat:
domain:cli#1(seat post #6024)Ruling-ref: 6181973363 (the test-budget ruling: no timeout raise, no deletion without a named duplicate, no move to a slower or skipped project; a
// test-budget:only for a measured wait of a class it names)File surface (read on
origin/mainbd4276a91; stop on a breach and explain it in the report):main: the file's wall time, unloaded, as a median of interleaved runs, plus a per-case profile.packages/qa/dogfood/test/activity-text-predicate.dogfood.test.ts, test-side waste only (repeated boots, redundant probes), with every assertion kept.packages/restincluded, even though it is this lane's. A platform cost is reported withclass:andreach:, and the seat slices it.// test-budget:unless the measured wait is of a class the ruling names.@objectstack/verify..github/workflows/rerun-safety-nightly.ymlor its scripts.Container & model:
S(one test file, measure-first),mode:subagent,model: TIER_DEFAULT(dispatch-gates --tier: no path-derived mandate)Clause-②: no
Basis for the Clause-② line: test-only, in a private package (
@objectstack/dogfood). No accept set and no public surface moves; a measurement-only outcome changes nothing at all.Responsibility:
packages/qa/dogfood(this lane) | the file sums over the 10 s per-file line with no declared budget, the dogfood half of ci(rerun-safety-nightly): red three nights running on main and nobody was told — pass 1 fails @objectstack/cli and @objectstack/dogfood, pass 2 adds a different package set each night, 10-16 hit the 170-minute timeout #23880's Rerun Safety pass-1 red | the Rerun Safety nightly onmain(two readable red nights before PR test(dogfood): activity-text-predicate pays its boot's first load at module scope and splits its slow control case #24026's half landed); reach as filedThread-read: 6228296716
Serial constraints cleared:
previousisnullon insert in objectstack-data and objectstack-formula, no longer unbound #24174, docs(checklist-test): a verification re-run files no qa-run record; sweeps still file one per run #24172, feat(spec)!: record:json is retired — refused by name at the schema door, the approval request page's raw JSON panel it was declared for was dropped (#24162) #24170 and fix(cloud-connection)!: the marketplace reseed door converts the ledger body forward once, as install and rehydrate do #24169 touch nopackages/qa/orpackages/rest/path. chore: version packages (next) #24168 (chore: version packages (next), 177 files) touches onlyCHANGELOG.mdandpackage.jsonunder them, not this file.pm:dispatchedcards read in this act: cloud-connection: the marketplace reseed door handsapplySideEffectsthe raw ledger entry, so it reloads translations a conversion strips (a pre-17.4 per-appsettingsgroup) that install and rehydrate load converted #24040 (this seat, landing) is the onlydomain:clione.6221450181was released by6223543949(PR test(dogfood): activity-text-predicate pays its boot's first load at module scope and splits its slow control case #24026 landed as2b274600b,Part of #23977). ItsBlocked-by: #24031is met (PR fix(rest): the API-exposure gate reads the one object it checks, by name, instead of every object's metadata on each /data request #24093 →c1c0a90942, an ancestor oforigin/main), and seat 2 returned the card to the queue at6228296716.objectstack-fleet commentedon Oct 17, 2026
os-dev-report
{ "issue": 23977, "status": "done", "branch": "claude/issue-23977-activity-text-predicate-remeasure", "pr": null, "session": "session_01BVm1pAXVXQ3PjC42j1kPFf", "premise_still_valid": true, "summary": "Measure-first result: the file is under the 10 s line on main, so there is no code change and no PR. The reading was taken at origin/main bd4276a91. c1c0a90942 (PR #24093) and 2b274600b (PR #24026) are both ancestors (merge-base --is-ancestor exit 0 for each). The closure build was 65/65 turbo cache hits, and packages/rest/dist carries the by-name loadObjectItem read. The unloaded file line of vitest's default reporter read 7009 / 6316 / 6257 / 6504 / 6725 ms over 5 isolated runs, median 6504 ms (round 0's head median was 12247 ms). The card's second half still holds. Under 7 CPU hogs, the slowest control part took 920 ms, 18 % of vitest's 5000 ms default, against 2500 ms for the 2x line. One caveat for the close: this is an idle-box reading. The 10 s line is judged on CI's file line, and a CI runner 1.5x slower than this box would read about 9.8 s. CI's post-fix reading is NOT MEASURED here (see tests).", "tests": "Box: 4 vCPU (nproc 4), 15 GiB RAM, container S. The verify lock was free with an empty queue and a 0 s wait at every acquisition. The process table before the runs showed no sibling work (the harness at 2.5 % CPU, nothing else), and the load average was 0.07 before the unloaded series. So this is an idle-box reading by process-table observation; the lock itself only certifies that no other locked run overlapped. Tree: worktree at bd4276a91, no edits, git status --porcelain empty throughout. Build: os-verify-lock -c 'pnpm exec turbo run build --filter=\"@objectstack/dogfood^...\" --concurrency=2' => VERDICT command-exit 0, 'Tasks: 65 successful, 65 total / Cached: 65 cached, 65 total'. grep shows 'function loadObjectItem' once in packages/rest/dist/index.js. Run shape, each run under os-verify-lock: env -u AI_AGENT -u CLAUDECODE NODE_OPTIONS=--max-old-space-size=4096 pnpm --filter @objectstack/dogfood exec vitest run --maxWorkers=2 --reporter=default --reporter=SCRATCH/timing-reporter.mjs test/activity-text-predicate.dogfood.test.ts. The second reporter is a measurement-only reporter in the scratchpad, never in the tree. It records per-case diagnostic durations and hook start/end stamps; the hook stamps are main-process event times, about 0.1 s grain. UNLOADED, 5 isolated sequential runs, every one 'Tests 25 passed (25)', VITEST_EXIT=0. File line: '(25 tests) 7009ms' / 6316 / 6257 / 6504 / 6725, median 6504 ms (run 1 had the cold transform, import 48.9 s against about 31 s for the later runs). beforeAll hook: 2733 / 1943 / 2070 / 2093 / 2239 ms, median 2093. Sum of the 25 case durations: 4158 / 3816 / 3592 / 3819 / 3868, median 3819. The file line minus the case sum (beforeAll + afterAll + inter-case overhead) was 2500 to 2857 ms; afterAll (stack.stop) logged a 4 ms shutdown. Slowest five, from the median run (unloaded-4), with ranges over all 5 runs: control part 'masked' 400 ms (305-411); 'control: the platform admin ... no parent named' 374 (374-422); control part 'gated' 317 (286-391); control part 'withheld' 285 (285-352); 'a reader of the masked class > the same probes are refused through the query door ...' 236 (195-301 across the three classes' query-door cases). Control parts unloaded: masked 411/384/305/400/319; gated 389/391/286/317/381; withheld 348/352/315/285/292; tail (non-matching snapshot + grouping) 93/50/51/49/50. Worst 411 ms. LOADED, 7 CPU hogs (bash busy loops; recorded PIDs killed by a trap, all 7 confirmed gone afterwards), load average 7.0 to 8.6 on 4 cores, about 2.1x. 3 runs, all 'Tests 25 passed (25)', VITEST_EXIT=0. File line 13659 / 13792 / 14134 ms, median 13792. beforeAll 5334 / 5476 / 6023 (the hook declares its own 240000 ms timeout). Control parts loaded: masked 680/793/774, gated 737/920/762, withheld 637/653/628, tail 103/98/102. Worst 920 ms, 18.4 % of 5000 ms. The slowest case under load was the admin control: 903/945/801, worst 945 ms (18.9 %). Against round 0, a different session of this box, not an interleaved A/B: unloaded median 12247 to 6504 ms; slowest control part 779-832 to 285-411 ms unloaded and 1629-2488 to 637-920 ms loaded. ATTRIBUTION NOT MEASURED, reason: the card closes on the absolute reading, and an A/B against c1c0a90942^ needs a second closure build. The drop sits inside round 0's projection (6-8 s). CI READING NOT MEASURED, reason: the full battery's dogfood logs are reachable only through MCP job-log reads (REST /actions/jobs/ID/logs answered CONNECT 403 via the proxy, http 000). Pointer for the seat: by vitest 4.1.11's sha1 sharding (BaseSequencer.shard, root-relative path), the file is index 176 of 269 at bd4276a91, so shard 2/3 ([90,180)). The first post-fix full battery read here is schedule run 39034818738 at 697301491b (c1c0a90942 an ancestor, exit 0); its job 117402469943 'Dogfood Regression Gate (2/3)' concluded success. The refresh lane reads the same '(N tests) Xms' file line (report-test-timings.mjs).", "gates": "None owed and none run: zero-path diff. No file was edited, the worktree stayed byte-identical to bd4276a91 (git status --porcelain: 0 lines before removal), and no commit exists beyond origin/main. So there is no eslint, typecheck or dispatch-gates --commands/--ran reconciliation to run. No PR, so there are no check-runs to cite.", "mcp_calls": "0 — none (one ToolSearch schema load for get_job_logs, never called)", "api_writes": "1 — this os-dev-report comment, POST /repos/objectstack-ai/objectstack/issues/23977/comments through scripts/pm/post-stamped.mjs (fleet-write relay). Plus 1 git push of the empty branch (not REST). REST reads: GET issues/23977, its comments, issues/comments/6181973363, actions/workflows/ci.yml/runs, actions/runs/39035573068/jobs, actions/runs/39034818738/jobs; 3 refused log GETs (CONNECT 403, nothing written).", "deviations": [ "Conflict, dispatch vs os-dev.md, resolved toward os-dev.md (file wins on unconditional clauses), stated here rather than chosen silently. The dispatch said 'push nothing' on the under-10 s path. os-dev.md rule 1 makes pushing the empty branch the first act after creating it (landing marker for the claim's branch, and the write-route probe). It was pushed before any measurement: remote claude/issue-23977-activity-text-predicate-remeasure = bd4276a91 (= origin/main at creation), zero commits ahead and no PR. Deleting it is a write outside this budget; the seat may delete it or leave it.", "Five unloaded runs instead of the minimum three. They are sequential, not interleaved, because Step 1 has no head to interleave with; the loaded series ran after them.", "beforeAll durations come from a scratchpad reporter's hook-event stamps (main-process, about 0.1 s grain). The tighter bound beside them is the file line minus the case sum.", "Worktree removed after measuring (node_modules deleted first, then a plain git worktree remove, exit 0). The local branch ref remains in the shared .git." ], "open_questions": [], "out_of_scope_findings": [] }Generated by Claude Code
objectstack-fleet commentedon Oct 17, 2026
Closed on the reading ·
domain:cliseat 1 ·os-sales· sessionsession_01BVm1pAXVXQ3PjC42j1kPFf· 2026-10-17T14:03ZThis card's close condition is its own, stated in seat 2's Release (
6223543949) and Unlock (6228296716): re-measure the file onmainafterc1c0a90942, and close on a reading under the 10 s line. The measure-first dispatch (claim6234558074) took that reading with no code change and no PR. The report is above, on this card.origin/mainbd4276a91;c1c0a90942and2b274600bboth ancestors)beforeAll/ case sum (median)controlpart, unloaded// test-budget:(ruling6181973363): the file is under the line without one.c1c0a90942is NOT MEASURED here. The post-fix hourly dogfood shard's log, as reachable from this container, ends before the file's line; the timing artifacts sit on a blob host this container cannot reach.9ffe8af09, which carriesc1c0a90942) failed pass 1 in@objectstack/dogfood#test (timeout-only). Its verdict names the task, not the file. The pass-1 lines that would name the file are beyond the reachable part of the log, and the job's annotations carry no test-file path on any night (control: the 10-14 night, when this file is known to have timed out, carries none either). So that night's timeout is not attributable to this file from here, in either direction.activity-text-predicate.dogfood.test.tsas timed out or over 10 s. A pointer goes to ci(rerun-safety-nightly): red three nights running on main and nobody was told — pass 1 fails @objectstack/cli and @objectstack/dogfood, pass 2 adds a different package set each night, 10-16 hit the 170-minute timeout #23880 in this act, so the night's dogfood timeout keeps an owner instead of resting on this card.claude/issue-23977-activity-text-predicate-remeasure(atbd4276a91, no commits, no PR) is left as is.