Raw `bun test` mis-loads Playwright e2e specs and reports phantom failures#

Problem#

Running bun test (the bare command) in this repo collects and tries to execute the Playwright specs under e2e/*.spec.ts with Bun's own test runner. Playwright's test() refuses to run outside the Playwright runner, so those files error out — producing a scary "15 fail / 15 errors" summary that looks like a real regression but is purely a test-runner mismatch.

Symptoms#

$ bun test
error: Playwright Test did not expect test() to be called here.
...
      at .../e2e/theme.spec.ts:21:1

 276 pass
 15 fail
 15 errors
Ran 291 tests across 46 files.

The 15 failures/errors all originate from files under e2e/; the 276 passes are the real src/ unit + integration suite.

What Didn't Work#

Reading the failures as a code regression. After a value-preserving CSS-token rename, seeing 15 failures immediately after the change suggests the change broke something. It did not — none of the failing files are exercised by the change, and the failure text is a runner error, not an assertion failure.
Assuming bun test is the project's test command. It is the Bun default, but this repo does not use it directly; the real suites are wrapped in scoped scripts.

Solution#

Use the project's scoped scripts, never bare bun test:

// package.json
"scripts": {
  "test": "bun test src", // unit + integration only (Postgres-gated on DATABASE_URL)
  "test:e2e": "playwright test" // Playwright runner; Docker/Testcontainers required
}

Unit / integration: bun run test (scoped to src/, so e2e/ is never collected).
End-to-end: bun run test:e2e (the Playwright runner).
Full gate: just ci-check (lint, format-check, typecheck, pre-commit-run, test, test-e2e) — the mandatory pre-commit gate.

Run through the scoped scripts and the phantom failures disappear (276 pass / 0 fail); the e2e suite runs green under Playwright (24 passed).

Why This Works#

bun test <path> globs every *.test.ts / *.spec.ts it can find under the given path. Bare bun test defaults to the whole repo, which sweeps in e2e/*.spec.ts — Playwright specs that must run under the playwright binary, not Bun. Scoping to src (bun test src) excludes e2e/ by path, and playwright test owns the e2e specs with the correct runner and its Testcontainers/Docker setup. The two runners never see each other's files.

Prevention#

Always invoke the scoped scripts (bun run test, bun run test:e2e) or the full just ci-check — treat bare bun test as "wrong tool" in this repo.
When a post-change test run shows failures, check the failure type before the count. A Playwright Test did not expect test() to be called here (or any "runner X did not expect Y") message is a collection/runner mismatch, not a code regression — look at which runner loaded which files before diagnosing the code.
This mirrors the repo's testing rule: integration/e2e run via Testcontainers under Playwright; see AGENTS.md and [[e2e-dotenv-mise-clobbers-launcher-env]] for the e2e harness details.