test(cli): record the measured cause of this suite's cost in the vitest config header

[os-elon]

Fixes#10152

Comment-only change to packages/cli/vitest.config.ts. No test block was added; the executable config is byte-identical.

What was measured

One machine (4 cores), warm build, npx vitest run --maxWorkers=2 per package, vitest 4.1.10, tree at 04096f1. Gate union re-run on the final commit b1ff122.

Both normalisers, because per-file cost alone cannot separate "expensive suite" from "more tests per file":

package	files	tests	tests/file	wall	s/file	s/test
@objectstack/cli	137	1498	10.9	495.81s	3.619	0.3310
@objectstack/spec	415	11045	26.6	325.31s	0.784	0.0295
@objectstack/service-automation	83	991	11.9	57.97s	0.698	0.0585
@objectstack/driver-turso	39	1003	25.7	38.50s	0.987	0.0384
@objectstack/client	23	314	13.7	23.75s	1.033	0.0756
@objectstack/example-showcase	21	342	16.3	31.15s	1.483	0.0911

The null hypothesis — "cli simply carries more tests per file, so per-file cost is the wrong normaliser" — is falsified, and in the direction that strengthens the finding: cli has the lowest tests-per-file of the six, so its 2.4–5.2× per-file cost becomes 3.6–11× per test.

Where the cost is not

cli 495.81s (transform 29.47s, setup 0ms, import 192.91s, tests 774.95s, environment 17ms)
spec 325.31s (transform 13.79s, setup 0ms, import 63.90s, tests 456.36s, environment 49ms)
svc-auto 57.97s (transform 14.54s, setup 0ms, import 96.78s, tests 4.73s, environment 9ms)
turso 38.50s (transform 16.85s, setup 0ms, import 65.98s, tests 3.56s, environment 4ms)
client 23.75s (transform 18.31s, setup 0ms, import 39.72s, tests 2.25s, environment 3ms)

Per file, cli's import cost is 1.41s (mid-band — client 1.73s, driver-turso 1.69s) and its transform cost 0.215s (the lowest of the six). The wide dependency closure in package.json is not what the test files import. setup is 0ms everywhere.

Where it is

Test-body work, concentrated rather than uniform: median file 0.03s, 105 of 137 files under 2s, top 20 files = 87.7% of the wall. The 20 files that spawn the real CLI as a subprocess (bin/run-dev.js through tsx, against a mkdtemp project) are 56.1% of the file wall (300.1s) while carrying 177 of 1498 tests.

Each spawn re-executes the CLI's module graph in a cold process — so the standing per-file module-graph theory is the right mechanism, relocated out of vitest's worker where neither its transform cache nor its module registry can reach it. Floor per spawn, printing nothing but a version:

tsx bin/run-dev.js --version 6.5-6.8s (the source entry the e2e files use)
node bin/run.js --version 2.9-3.2s (the built entry)
node -e 0 0.031s (process floor)

Two levers measured and rejected here

• test: { maxWorkers: 4 } — 2→4 workers on an idle box is 495.81s → 337.13s, but CPU is flat (user+sys 1291.3s → 1256.2s) and per-file wall inflates (sum 535.1s → 748.3s; longest file 74.3s → 104.1s). The box is saturated: that is packing, not work. In CI the box is not this package's — ci.yml runs turbo run test --concurrency=4 — so pinning a worker count here spends cores belonging to whatever else lands on the shard. Worker allocation is a shard property, decided in ci.yml.
• NODE_COMPILE_CACHE — 6.98 / 6.24 / 6.41 / 6.18s cached vs 6.39 / 6.72s uncached: inside noise, for 42MB of cache. The per-spawn cost is module-graph execution, not compilation.

So the work is real and nothing contained in this package removes it without changing what the e2e tests assert. Swapping the spawns to the built entry would halve per-spawn boot and is exactly the source-vs-dist trade scripts/check-test-source-alias.mjs exists to refuse.

Scope

scripts/partition-test-shards.mjs, ci.yml and the shard count are untouched by design — #10149's A/B choice is parked behind this measurement and is not prejudged here. #10149 is not addressed by this PR and remains open; the reading that bears on it is in the report comment on #10152.

Verification

b1ff122 · all captured by redirecting to a file before reading, never through a pipe:

check:slot-lookup exit=0 baseline key set verified against 04096f1: no files added.
check-affected-docs exit=0
check:test-source-alias exit=0 72 packages with tests scanned; 43 published subpath(s) resolved through every alias table.
check:nul-bytes exit=0 scanned 6075 text file(s); no raw ASCII control bytes.
check:cross-package-test-inputs exit=0 All 60 self-test cases passed.
check:docs-image-tag exit=0
check:override-consistency exit=0
check:route-envelope exit=0
check-changeset-fixed exit=0
check-osv-exemptions exit=0
cli typecheck (tsc --noEmit) exit=0

node scripts/pm/dispatch-gates.mjs (no paths — it derives its own change set) named only check:slot-lookup and check-affected-docs for this diff; the rest of the list above is the dispatch's lead list, run anyway. No family had to be added.

Suite itself: Test Files 137 passed (137) · Tests 1498 passed (1498) at both --maxWorkers=2 and --maxWorkers=4.

skip-changeset: the diff is one comment block in a test-config file. Nothing lands in src/, so nothing publishes.

Generated by Claude Code

---

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 1 changed package(s)), so this run has no opinion about the docs.

What this run could not see

• 1 changed file(s) yielded no anchor (packages/cli/vitest.config.ts) — pages documenting those are invisible to this run

Coarse fallback — 23 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 318f96ae5cc25d04a2650a6ef6d4c2b930619b96 → packageMentionDocs.