[objectstack-fleet]

Part of #23181
Clause-②: no

The @objectstack/runtime row of #23181 (slice 1): three test files the hourly full battery read over 10 s of wall. This PR edits those three test files only: no runtime source, no vitest config, no timeout value, no test removed or moved.

Rows

file	card's CI reading	this box, isolated, median of 5 (before to after)	this box, whole-package run (before to after)
src/declared-capability-absence-warn-once.test.ts	19.4 s (38325245250)	23145 ms to 101 ms	10990 ms to 91 ms
src/stored-metadata-reader-contexts.pin.test.ts	20.7 s (38314183560)	5756 ms to 5085 ms	7241 ms to 8207 ms (one run each, see notes)
src/stored-metadata-body-boundary.pin.test.ts	13.7 s (38210849670)	4976 ms to 4149 ms	6012 ms to 7085 ms (one run each, see notes)

• warn-once: this PR completes that file's row. Its clocked work goes from seven module-graph evaluations to none (margin to 10 s: about 9.9 s on this box, isolated and in the whole-package run).
• The two pin files: this PR removes the one test-side cost found and does not claim the rows. Both were already under 10 s on this box before the change (isolated and in the whole-package run); their CI readings are the slowest of 20 contended hourly runs. What remains in their clocked windows is CPU-bound boot and sandbox work (measured below), not a sleep, retry or poll. The next hourly full battery's timing refresh (shard-timings-refresh.yml, measure-test-shard-timings --diff) reads all three; if it names a pin file again, the remaining cost is outside the test (Acceptance notes).

"Isolated" is the file run alone, five times, with the agent-detection variables unset so vitest picks the default reporter CI reads; the number is the file line's result.duration (hooks plus tests, collection excluded). The pin files' before and after runs were interleaved (base, after, base, after), with the base bytes run from an untracked copy byte-identical to the merge base's blob (9423d04d44 for body-boundary). Every after run sat below every before run. This is a shared 4-core box, so read the ratios as well as the absolutes.

Where the time went, and the change

declared-capability-absence-warn-once.test.ts

• Before: each of the 7 tests called vi.resetModules() and re-imported the dispatcher and sendError, re-evaluating the same 733 modules (spec/src 422, core/src 93, platform-objects/src 71, runtime/src 64, rest/src 24, types/src 22, plus the inlined objectql and metadata-protocol dist bundles): about 0.9 s per test isolated, 1.1 to 1.4 s per test in the whole-package run. In isolation the first test also paid the graph's cold transform inside its clock (17.0 s; vitest transform 15.5 to 17.0 s).
• Change: one module graph, imported at module top so the transform is paid during collection (the clocked-window rule, scripts/check-test-source-alias.mjs). The per-test isolation the reset bought is kept by the route key instead. The dedupe registry is keyed on METHOD plus the mounted route pattern (parked by observability/instrument.ts), so each boot() mounts the dispatcher under its own prefix (/t1/api/v1, /t2/api/v1, and so on). Every key a test touches is one the registry has never reported, the same precondition a fresh process gives those keys. Handler lookups go through route('GET /notifications'); asserted paths through path('/notifications').
• Whole-file wall with collection included also fell: vitest Duration 22.2 to 23.5 s before, 17.4 to 19.7 s after (isolated).

stored-metadata-reader-contexts.pin.test.ts and stored-metadata-body-boundary.pin.test.ts

• Where it goes: one boot per file in beforeAll (already shared), then the tests. On the base bytes of the reader file (3 runs), beforeAll took 2.7 to 3.1 s wall against 3.1 to 3.4 s CPU, and the tests 2.6 to 2.9 s wall against 5.5 to 6.2 s CPU. CPU at or above wall in both windows means no idle wait to remove. bootstrap() is about 2.0 s, of which the objectql plugin's start() is 0.54 to 0.58 s and init/ready the rest. The two waitFor polls (100 ms interval, for a /meta-authored action or hook to bind through the event-driven resync) settle in 153 to 274 ms per test, and in 826 ms for the body-boundary file's hook bind, whose probe runs three hook bodies per attempt.
• The test-side cost found: bootstrap() first-evaluates two workspace modules inside beforeAll, driver-sqlite-wasm/dist and driver-sql/dist, reached through service-datasource's lazy import(). These are the transform-queue-exposed kind of load the clocked-window rule is about. Every other first load in the boot is a native external (better-auth, knex, sql.js, nanoid), and better-auth is not a runtime dependency, so a module-top import of it would be a phantom dependency.
• Change: module-top import '@objectstack/driver-sqlite-wasm' and import '@objectstack/driver-sql', the shape datasource-visibility.test.ts already carries. Both pairs were already registered for runtime in KNOWN_UNALIASED_TEST_IMPORTS. Measured: bootstrap() 2.0 to 2.1 s became 1.5 to 1.8 s in a probe; the files went 12% and 17% faster isolated.

Every assertion kept

• All three files: it/describe title lines are hash-identical at the merge base and head (reader 21 it sites / 27 tests, body-boundary 5 / 7, warn-once 7 / 7). The two pin files gained two import lines and a comment; no test body changed.
• warn-once: 25 expect( calls before and after, and each test proves the same thing:
  • reports a declared 501 ONCE across N requests, at warn, naming the missing service: five requests to a route the registry never saw, one warn, zero error. The asserted path is this boot's mounted pattern.
  • a SECOND, DIFFERENT route still reports — the dedupe key is the route: two never-seen routes, one line each, same code and prose.
  • an UNDECLARED 500 on the same door is still loud, once per request: three faults, three error lines, no warn.
  • a declared absence and a fault coexist in one process without either changing the other: one warn and two error on one mount.
  • the envelope the dispatcher answers is classified identically by the REST envelope writer: the dispatcher and sendError still share one @objectstack/types registry (one module graph), and both lines name the service and the suppression.
  • answers the same bytes it answered before this change, on every request: request 1 is still the reported occurrence on a never-seen route, so the first-versus-suppressed byte comparison still holds.
  • answers the same bytes for a fault, too: unchanged.

One-shot proofs (run from the committed state, not kept as tests)

Both went through scripts/ablation-replace.mjs (anchor hit counted, blob hash moved, restore proven by the head blob hash and an empty git diff HEAD), against the warn-once file at 8d5fab4bdc:

1. Isolation removed (every boot on the shared /api/v1 prefix): 3 failed, 4 passed. a SECOND, DIFFERENT route… got 1 line where 2 were expected, …coexist… 0 where 1 was, and the envelope… (door 1) 0 where 1 was. That is the order-dependence the reset used to prevent, so the prefix is load-bearing. Blob a643630be43d to e35018939f4d, restored to a643630be43d.
2. Dedupe removed (packages/types/src/server-fault-log.ts, the has(key) early return; aliased to source for this suite, so no build leg): 3 failed. reports … ONCE… got 5 lines where 1 was expected, a SECOND… 4 where 2 were, and …coexist… 2 where 1 was. The file still pins the dedupe itself. Blob 7ed86cc374e8 to 758f68960025, restored.

Verification at 8d5fab4bdc (branch merged with main at d4aaeaf0e2)

• pnpm --filter @objectstack/runtime test: Test Files 364 passed (364), Tests 5178 passed | 19 skipped (5197), lock verdict command-exit 0.
• pnpm --filter @objectstack/runtime typecheck: exit 0. tsc --noEmit is clean, and check:test-typecheck: OK. The three files are in the tsconfig.test.json program (3 of 367 runtime test files) and carry no entry in test-typecheck-debt.json.
• Lint, narrowed to the touched files with its proof:
  • eslint, invoked as pnpm lint invokes it with --no-inline-config --format json, over the 3 files: 3 files, 0 errors, 0 warnings, 0 fatal.
  • --print-config resolves each file to an active config (not ignored).
  • eslint.config.mjs enables no type-aware linting (no parserOptions.project, no typed rules; its own comment says so), so this diff cannot move any untouched file's verdict.
• node scripts/pm/dispatch-gates.mjs --commands: 50 commands, run as printed with exit codes captured before any pipe. 49 exit 0 with their own pass lines. pnpm check:dual-build-cjs-loads first answered PREREQUISITE NOT MET (exit 3: 38 packages had no dist/). After a full build (74/74 tasks), the same command answered exit 0: 108 require entry points across 67 packages load. --ran: 50 derived, 50 run, 0 NOT-MEASURED, 0 UNRUN. (Corrected by the seat from the dev's report; the PR body had been written before that run.)
• No changeset: runtime's files is dist, README.md, CHANGELOG.md. A string unique to this change has 0 hits in them; a control string from dispatcher-plugin.ts has 1 hit in dist/index.js. Label skip-changeset.

Acceptance notes

• The pin files' residual, if a refresh names them again. Their clocked work is the composed boot (about 1.5 s after this change) plus one sandboxed body per assertion. The sandbox instantiates a fresh WASM module per invocation by design (ADR-0102 D2), which shows up as about 2x CPU over wall in the test window. A 4-vCPU shard running turbo --concurrency=4 multiplies exactly that. The card's readings sit 2.3x to 2.9x over this box's whole-package run. Neither lever is a fixed delay or poll the test waits on, so neither is in this slice: a cheaper boot, or reusing compiled WASM across invocations while keeping separate instances. Observation only; carrier: none.
• Whole-package before/after for the pin files is not an A/B. It is one run each, on different trees (merge base versus merged main), on a shared box (the after run queued 187 s behind sibling work), and with a changed file order (vitest schedules by previous duration, and warn-once is no longer the slowest file). The module-top imports cannot add clocked time, since they move evaluation into collection. The interleaved isolated runs are the comparison.
• warn-once's vi.setConfig({ testTimeout: 30_000 }) was raised for the per-test reload this PR removes. It is left as it stands, because this slice's claim excludes timeout values. Its comment now says why it was raised, so whoever next owns the file can drop it.
• @objectstack/runtime CI reads (Test Core) will be the first contended reading of the new shapes. The hourly timing refresh is the pin.

---

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 44e2d505c7e6f5da8efc4419fc9de40bbee9ebc0 → packageMentionDocs.