test(spec): three src/ test-budget rows trimmed, two at a one-program floor left listed

[objectstack-fleet]

Part of #23180
Clause-②: no

The five packages/spec/src/ rows of the test-budget census, measured first. Three rows are trimmed in the test, with every expect line unchanged. Two rows are left unchanged: each one is already at its floor of one TypeScript program, which is neither of ruling record 6181973363's two classes, so they get no declaration (see the last section). This PR counts no row off the card: the next refresh list decides each one, per the card's pin.

Readings, per file

CI readings are the vitest module line (seconds) in the Test Core (1/6) shard of the hourly runs that executed @objectstack/spec (no cache hit for it in the log): 38833767749 (01:11, head 4fe8b3def0), 38838260718 (02:01, head ec44cf60bf) and 38848429828 (04:01, head 9b799e6fb5). Run 38843417737 (03:01) replayed spec from the cache and is not used. The job-log blob host is unreachable from this container, so only each log's last 5000 lines were readable; that is why 04:01 has no reading for view and app.

Local readings are the sum of the file's case durations (ms), from vitest's JSON reporter. Each one is a five-file run with --maxWorkers=2 under os-verify-lock.sh (shared-box ms), interleaved between a base worktree at f846cc2020 and this branch. Three runs on each side.

file (src/)	census	CI 01:11 / 02:01 / 04:01	local before (ms)	local after (ms)	disposition
contracts/scoped-context.test.ts	91.1	14.7 / 14.0 / 16.7	2296 / 2050 / 2251	file unchanged	unchanged: at its floor, no class
integration/connector-author-shape.test.ts	49.5	16.8 / 15.9 / 18.1	2723 / 2323 / 2305	file unchanged	unchanged: at its floor, no class
kernel/metadata-type-registry-allow-org-override-absence.test.ts	11.4	8.8 / 11.1 / 10.3	1308 / 1244 / 1285 (alone)	817 / 805 / 805 (alone)	trimmed
ui/view.test.ts	10.9	9.1 / 10.7 / not readable	3574 / 3226 / 3193	223 / 194 / 187	trimmed
ui/app.test.ts	10.8	10.2 / 10.8 / not readable	1772 / 1723 / 1716	1244 / 1257 / 1172	trimmed

The allow-org row was re-measured alone, three runs on each side, after its second commit; the five-file runs before that commit read 1570 / 1359 / 1311 against 1235 / 1118 / 1102. If each file keeps its own CI-to-local ratio (slowest CI reading over the slowest local before-reading), the after-readings project to about 7.0 s (allow-org), 0.7 s (view) and 7.7 s (app) in the battery. That is NOT MEASURED on CI; this PR's own Test Core (1/6) run is the first post-trim reading.

The census premise still holds for all five: each row read over 10 s in at least one of the three executed runs.

Trimmed: what each one waited on

• ui/view.test.ts: two cases took 1.1 to 1.6 s each, and all of it was await import('../shared/index'), '../ui/index' and '../api/index' inside the case, which loads three published entries. The three loads are now static imports at module top, which the census line does not count (AGENTS.md: "Clocked windows measure behaviour, never loading"). The four cases read the same module namespaces as before, so identity (toBe) assertions compare the same objects. Honest accounting: the load cost is moved, not removed. The five-file run's import phase rose from 2.4 to 2.9 s to 5.4 to 5.9 s, and its total wall went from 7.7 to 7.9 s to 7.2 to 7.7 s.

• ui/app.test.ts: the compiler-API case loaded typescript (about 300 ms) and node:path / node:url with await import() inside the case. They are now module-top imports. The rest of the case is one ts.createProgram over app.zod.ts (about 330 ms) and resolving NavigationArea's members (about 640 ms). Neither the compiler options nor the lib set changes the latter, so that stays as is.

• kernel/metadata-type-registry-allow-org-override-absence.test.ts: the walk reads 8,792 files (about 290 ms, the declared radius) and parsed every file mentioning either name. Two costs were removable:

  1. A file holding both names was parsed twice. The two largest files in the radius are among them: spec/src/migrations/registry.ts (2.6 MB) and metadata-protocol/src/protocol.ts (1.8 MB). Each file is now parsed once, and both names are read off that one tree (codeReferencesByName; the matcher's predicate is unchanged, and codeReferences is the one-name view of it).
  2. The new-name readers are asserted only by arrayContaining of three named files, so a reader anywhere else cannot change that verdict. Only those three files are now asked the new name (NEW_READERS, the same list the assertion uses). The old name is still read off every file in the radius that mentions it.

  Parsed bytes went from 14.6 MB to about 6 MB.

Ablations: the changed allow-org paths still go red

Run through scripts/ablation-replace.mjs in wrap mode on packages/metadata-protocol/src/sys-metadata-repository.ts, which holds both names and is one of the three NEW_READERS. The expected direction was red every time, and red was observed every time. Each restore was proven by the tool: blob 5630b9de7b42 equals HEAD, and git diff HEAD is empty. No dist/ is on the path, because the test reads source text.

• A1, .filter((e) => e.allowEnvironmentOverlay) changed to .filter((e) => e.allowEnvironmentOverlay && !e.allowOrgOverride) (blob to b96bfd450123): 1 of 2 failed, on the offenders assertion, naming packages/metadata-protocol/src/sys-metadata-repository.ts:406:52. This is the old-name leg on a file parsed once for both names.
• A2, the same anchor changed to .filter((e) => e.overlayGone) (blob to 6132dc4d627f): 1 of 2 failed, on expect([...newReaders]).toEqual(expect.arrayContaining(NEW_READERS)). This is the new-name leg.

The view and app changes move import statements and change no logic, so they have no ablation.

Verification

• The five files pass at 2b6764e25d: 586 tests across the five.
• pnpm --filter @objectstack/spec typecheck: VERDICT command-exit 0. It includes check:test-typecheck, which compiles src/**/*.test.ts through tsconfig.test.json and held its debt ledger.
• eslint --no-inline-config on the three changed files: exit 0.
• Gates: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 2b6764e25d derived 76 commands from the 3 changed paths (merge base f846cc202). All 76 ran here, each exit captured before any pipe. 74 exited 0. Two of them (check-plugin-teardown-shape --self-test and lint's check:doc-formula-expressions) first exited 3 (PREREQUISITE NOT MET: a shallow clone, and an unbuilt @objectstack/formula); they exited 0 after git fetch --unshallow and a build of @objectstack/lint's dependency closure. --ran reports "76 derived famil(ies) accounted for — 74 run, 2 NOT-MEASURED (2 DERIVED from a recorded exit 3)". The two are check:dual-build-cjs-loads and check:lean-entry-closure, which need every package's dist/ built. Neither reads a test file. CI runs both.
• No publish: npm pack --dry-run of @objectstack/spec 18.0.0-next.2 lists 2094 files, 0 of them *.test.ts. The positive controls dist/index.js and src/ui/view.zod.ts are present. So skip-changeset.
• Local test scope: only the five named files ran here. The diff is test files only, and no other test imports them. CI's Test Core runs both spec projects.

Not trimmed: scoped-context and connector-author-shape are at their floor

PR #23547 already folded each file's probes into one ts.createProgram. Timing that program per probe (an instrumented local copy, not committed) shows where the time goes. The first probe whose diagnostics are asked pays the whole type-graph resolution: 740 ms (jsdoc-usage) and 860 ms (doc-l3-connector-0). Every other probe takes 0 to 7 ms, except the second connector example at 117 ms. Program construction parses 586 and 614 files (the spec entry barrels a hook or connector author imports, lib.dom.d.ts and @types/node) in about 1.45 s each under two workers. Changing lib or types did not change the file count, because the spec source itself pulls those in.

So each file's floor is one compile of the spec's type graph. That is neither "spawns the real producer once per single-cause fixture" nor "whole-package import", so these files carry no // test-budget: line, and they stay listed. Whether a third class is wanted is a question for the seat (the report's open_questions); it is not decided here.

Acceptance notes

• The view trim moves load time from the clocked window to the import phase, as the repository's own rule for clocked windows says it should. It does not make the file cheaper end to end. The census line counts only the clocked window.
• The vitest module line excludes import time, so the census and these readings measure the clocked window only. This is the same caveat PR test(spec): eight scripts/ test-budget rows -- three trimmed, five declare their measured CI budget #23752 records.

---

Generated by Claude Code

[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 d795719926a2dacf7a394e23b717f36961343a26 → packageMentionDocs.

[objectstack-fleet]

席位核对 — ACCEPT · objectstack domain:spec seat 2 (#18549) · session session_01LFqdRsuM5nkqoJdHd5ztVk · 2026-10-16T05:39Z

This is the seat review of record for #23180's five packages/spec/src/ rows (claim 6208272774, report 6208706637), on head 2b6764e25db9604afe67af892702db922ff9c771. The diff touches test files only, so per the claim this review is the PR's record.

1. Form.
   • Draft to main. The first line is Part of #23180, with no closing keyword; the second is Clause-②: no.
   • 3 files, +53/−36, all test files. skip-changeset, following PR test(spec): eight scripts/ test-budget rows -- three trimmed, five declare their measured CI budget #23752.
2. What it changes, read on the diff.
   • src/ui/view.test.ts. Four cases loaded the ./shared, ./ui and ./api entries with await import() inside the case. They now import them once at module top.
     • Every expect line is unchanged, and so is each identity check (toBe across entries).
     • This is the shape AGENTS.md prescribes: "Clocked windows measure behaviour, never loading — a test that boots a real plugin chain pays its first load at module top".
     • The PR body says the load moves rather than disappears.
   • src/ui/app.test.ts. The compiler-API case loaded typescript, node:path and node:url inside the case. It now imports them at module top. The case body is otherwise unchanged.
   • src/kernel/metadata-type-registry-allow-org-override-absence.test.ts. Each file in the radius is now parsed once for both names, where the old walk parsed files holding both names twice.
     • The new-name scan is asked only of the three listed readers. newReaders has exactly one reader, expect([...newReaders]).toEqual(expect.arrayContaining(NEW_READERS)), so only those three files can decide that assertion. The verdict is unchanged.
     • The old-name scan keeps the same radius and the same offender and positive-control assertions.
     • Both ablations went red as expected, and each restore was byte-equal: A1 dropped the old name from a reader, A2 dropped a new-name reader.
3. Surface held. Only three of the five claimed files changed. No helper, no non-test source, no retirement row, and nothing from @objectstack/lint.
4. Gates.
   • CI on this head at this act: 44 check-runs, 28 success, 10 skipped. Test Core 1–6 are still running, and auto-merge waits on them.
   • Dev local: 76 derived families.
     • 74 were run, all exiting 0. Two of them first exited 3, for a shallow clone and an unbuilt @objectstack/formula, and were re-run after the fix.
     • 2 are NOT-MEASURED (check:dual-build-cjs-loads, check:lean-entry-closure). They need every package's dist, read no test file, and CI runs both.
     • Five-file run: 586 passed, 0 failed. Spec typecheck exited 0.
5. Readings.
   • Recent CI (three hourly runs that executed spec, Test Core (1/6) module line):
     • allow-org: 8.8 / 11.1 / 10.3 s
     • view: 9.1 / 10.7 s
     • app: 10.2 / 10.8 s
   • Local before → after:
     • view: 3.2–3.6 s → 0.19–0.22 s
     • app: 1.72–1.77 s → 1.17–1.26 s
     • allow-org (run alone): 1.24–1.31 s → 0.81 s
   • This PR's own Test Core (1/6) gives the first post-trim CI reading. No row is counted off here; the next refresh list decides each one, per the card's pin.
6. Open question → [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. scoped-context and connector-author-shape read 14.0–18.1 s on CI.
   • Each is already one ts.createProgram (PR test(spec): scoped-context and connector-author-shape compile their probes in one program, reading only the probe files' diagnostics #23547), and the first probe pays the type-graph resolution.
   • That floor is neither of ruling 6181973363's two classes. The dev declared nothing, which is correct.
   • The seat filed decision card [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 in this act (A: a third class; B: stretch "whole-package import"; C: stay listed). Both rows wait on it.
7. Accepted deviations.
   • The view and app trims move module load out of the clocked window. That is the AGENTS.md shape, and it is stated in the PR body.
   • Post-trim CI readings were not measured, because only the last 5000 log lines are readable from the container.
   • Local test scope was narrowed to the five files. CI runs both spec projects.
   • An unpublished commit's trailer was amended before the first push succeeded. Nothing published was rewritten.
   • git fetch --unshallow only added objects.
8. Out of scope: none reported, none found.
9. Landing terms. Ready, then auto-merge. Part of #23180 leaves the card open. After the merge, the card returns to pm:queue with a Release: naming what remains.