[objectstack-fleet]

Requested by Steedos · project thread

Part of #23180. This PR covers the packages/spec and packages/lint rows of refresh PR #25960's per-file list that are still over 10 s or over their declared budget (12 rows, listed below). #23180 stays open for the other packages' rows.

Clause-②: no
Metadata-protocol: none

Summary

Tests only. Each row is either made cheaper (three files), or given a budget re-measured from #25960's window, or both. Nothing is deleted, no vitest timeout is raised, and no file moves to a slower or skipped project.

• packages/lint/src/non-record-object-entry.test.ts: the junk-value sweeps ran every rule once per assertion. They now run each rule once per distinct fixture and memoize the outcome (findings count, or the thrown message), and every assertion reads that record. 348/348 tests, 2.11 s to 1.00 s locally.
• packages/spec/src/ui/app.test.ts: the compiler-API case built its program with the ES5 default lib (DOM declarations) and every ambient @types package. Neither is an input of app.zod.ts, so the program now uses lib.es2022.d.ts and types: []. The guard still reads the full member list off the resolved type. 1.49 s to 1.25 to 1.37 s locally.
• packages/spec/scripts/build-schemas-check-mode.manifest.test.ts: the [finding] the published bundle json-schema/objectstack.json declares x-schema-count: 1596 but carries 1585 definitions — and a docs page publishes that field as "the total number of definitions" #12588 block ran the generator twice on one fixture for two read-only cases. It now runs it once per block (beforeAll plus a lazy memo, the same shape as openapi-self-consistency.test.ts's shared run). That is one generator spawn (about 7 s) fewer per run. 136.3 s to 125.1 s locally (shared box).

Rows covered, with their source lines

Source: refresh workflow run 40275747937, job 121418010419 (the #25960 refresh). Line numbers are that job log's.

Row	Log line	Slowest CI reading (run)	Declared before	Now
spec scripts/build-schemas-check-mode.retired-keys.test.ts	1339	241.05 s (40204927516)	190	budget 310
spec scripts/build-schemas-check-mode.manifest.test.ts	1340	226.64 s (40204927516)	180	one spawn fewer, budget 290
spec scripts/build-schemas-check-mode.anchor.test.ts	1346	144.97 s (40204927516)	140	budget 190
spec src/eager-entry-import.test.ts	1394	41.61 s (40266979145)	30	budget 60
spec scripts/upgrade-guide-docs.test.ts	1408	30.07 s (40214410982)	30	budget 40
lint src/lazy-deps.test.ts	1442	13.91 s (40200279306)	none	budget 20
lint src/runtime-lazy-deps.test.ts	1452	13.05 s (40200279306)	none	budget 20
spec scripts/root-entry-type-nameability.pin.test.ts	1453	12.92 s (40214410982)	none	budget 20
lint src/non-record-object-entry.test.ts	1456	12.49 s (40214410982)	none	cheaper, about half the local time
spec src/ui/app.test.ts	1459	11.90 s (40204927516)	none	cheaper, budget 20
spec scripts/pure-schema-construction.test.ts	1460	11.86 s (40214410982)	none	budget 20
spec scripts/openapi-self-consistency.test.ts	1473	10.48 s (40204927516)	none	budget 20

The window is the hourly runs 40200279306 to 40272142430. Over the window's executed readings, the declared budgets were exceeded in: retired-keys 6 of 10, manifest 6 of 10, anchor 1 of 10, eager-entry 7 of 8, upgrade-guide 1 of 8.

Re-measure arithmetic

Every budget follows rulings 6181973363 item 2: the slowest CI reading in the window, x 1.25, rounded up to the next 10 s. Each header line names its reading and run, and each was parsed with readTestBudget from scripts/measure-test-shard-timings.mjs.

• retired-keys: 241.05 x 1.25 = 301.3, so 310
• manifest: 226.64 x 1.25 = 283.3, so 290
• anchor: 144.97 x 1.25 = 181.2, so 190
• eager-entry-import: 41.61 x 1.25 = 52.0, so 60
• upgrade-guide-docs: 30.07 x 1.25 = 37.6, so 40
• lazy-deps: 13.91 x 1.25 = 17.4, so 20
• runtime-lazy-deps: 13.05 x 1.25 = 16.3, so 20
• root-entry-type-nameability.pin: 12.92 x 1.25 = 16.2, so 20
• ui/app: 11.90 x 1.25 = 14.9, so 20
• pure-schema-construction: 11.86 x 1.25 = 14.8, so 20
• openapi-self-consistency: 10.48 x 1.25 = 13.1, so 20

non-record-object-entry.test.ts gets no budget. Its local time halved, which puts its CI reading at about 6 s by the window's CI-to-local ratio (12.49 / 2.11). Per c91af8e, a later CI reading over 10 s would be the measurement a declaration needs.

Why the bscm, eager-entry and upgrade-guide floors stay

One build-schemas.ts spawn costs about 7.3 s on a profile: zod to JSON Schema 3.14 s, module load and script 2.73 s, writes 0.97 s, GC 0.40 s, child_process 0.24 s. Each bscm case is one spawn on its own single-cause fixture. retired-keys and anchor have no two cases on an identical fixture, so no run can be shared. In manifest, the #12588 pair was the one identical-fixture pair inside a block. Sharing across blocks (for example a #16431 control with another block's run) would couple fixtures that each block seeds itself, so it is left out. Batching causes or redesigning the producer is the option seat ruling 6176284384 declined.

eager-entry-import spawns one fresh tsx child per published entry by design, and upgrade-guide-docs spawns the real producer once per mode or usage error it pins (about 0.7 s each). Both are kept and re-declared.

Measurements

Local runs use the CI shape: CI=true pnpm exec vitest run --project P --maxWorkers=1 with the agent environment variables unset. They ran on a shared box, so ratios carry more weight than absolutes.

File	Before	After
lint non-record-object-entry	2111 ms	999 ms
lint lazy-deps	2312 ms	2305 ms
lint runtime-lazy-deps	1952 ms	2030 ms
spec ui/app	1490 ms	1250 to 1372 ms
spec manifest (bscm)	136.3 s	125.1 s
spec retired-keys (bscm)	124.0 s	header only
spec anchor (bscm)	121.6 s	header only
spec eager-entry-import	12.3 s	12.1 s
spec upgrade-guide-docs	6.7 s	7.3 s
spec pure-schema-construction	3.2 to 4.2 s	3.7 s
spec openapi-self-consistency	3.4 s	3.6 s
spec root-entry-type-nameability.pin	3.14 s	3.11 s

Mutation proof (non-record-object-entry)

The memo must not hide a rule regression. I mutated recordsOf in packages/lint/src/object-graph.ts to return v as AnyRec[] with scripts/ablation-replace.mjs, then ran both the new file and the BASE copy against the same mutant. Both failed the identical set, 121 of 348. The restore was proven: the blob equals HEAD's (693c48ac1308) and git diff HEAD was empty.

Acceptance notes

• Header-only files (retired-keys, anchor, eager-entry, upgrade-guide, pure-schema, openapi, root-entry pin, both lazy-deps files) change only their budget comment line.
• The bscm files run in the vitest repo project, which CI's Test Core owns. Locally I ran only the manifest file after the change, 18/18 green.
• test budget (domain:spec): 26 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23180 remains open. Rows outside packages/spec and packages/lint stay with it.

Verification

All runs below are on head 1a44227 (this branch merged with main bbf7061), unless noted.

• pnpm --filter @objectstack/lint test: 164 files, 7356 tests passed (exit 0).
• The six changed spec local files, run together after the merge: 6 files, 169 tests passed (exit 0).
• Before the second merge, on ea742f5: pnpm --filter @objectstack/spec test (723 files, 21902 tests passed, 1 todo), pnpm --filter @objectstack/spec typecheck, pnpm --filter @objectstack/lint typecheck. All exit 0.
• Derived gates: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack gives 84 commands. All 84 exit 0, and --ran prints: "84 derived famil(ies) accounted for, 84 run, 0 NOT-MEASURED". One note: check:dts-closure first exited 1. After the merge, @object-ui/plugin-editor could not build because main's new monaco-editor dependency was not installed yet. After pnpm install --frozen-lockfile and a rebuild of the two named UI packages, it exited 0 ("231/231 declared declaration file(s) present").
• The bscm files belong to the vitest repo project, which CI's Test Core runs. Locally, the manifest file ran after its change, 18/18 passed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LpCAbWQuprShJmC2GJXTp7

[github-actions]

📓 Docs Drift Check

Nothing in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 0 changed package(s)), so this run has no opinion about the docs.

What this run could not see

• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 0 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 8a990d8b4175a3da65c8f7690ac0b0da641654e5 → packageMentionDocs.

[objectstack-fleet]

席位核对

ACCEPT · domain:spec seat 2 · 2026-10-26T11:56Z · head 1a44227437fdc73694b27bd4e602f9d69eb45110

1. Shape. Draft, base main, Part of #23180 (the packages/spec and packages/lint rows of chore(ci): refresh the Test Core shard-timings dataset #25960's list; test budget (domain:spec): 26 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23180 stays open for the other packages). 12 test files, +85/−37, skip-changeset. check-governed-merges --pr 26089: NOT governed.
2. Review owed. None of the four contract faces is touched: test files only, Clause-②: no, Metadata-protocol: none, no published schema, no governed path, no narrowing declared. Per contract-review.md this is CI plus the seat's own read; no second reviewer.
3. Gates (check-runs pinned to this head). 53 runs: 40 success, 13 skipped, 0 failed. check-expected-skips --pr 26089: OK.
4. Seat read.
   • non-record-object-entry.test.ts: one sweep per distinct stack, each rule still given a freshly built stack. A rule that throws still fails the count with its own message instead of counting as zero, so the assertions are as strong as before.
   • build-schemas-check-mode.manifest.test.ts: one generator run shared by the block (beforeAll); both cases only read what that run printed and wrote. Budget 180 → 290, re-measured from the slowest of 10 readings × 1.25.
   • app.test.ts: the compiler program reads lib.es2022.d.ts with types: []; guard ✨ Set up Copilot instructions #2 still reads the full member list off the resolved type.
   • The remaining files move a test-budget: line only, each citing its measured reading.
5. Landing. Ready plus auto-merge.