Verification — debug-startup-smoke-flake

Step 5 of the 5-step executor chain. Captures the test evidence that the contract in docs/contracts/debug-startup-smoke-flake-contract.md has been fulfilled.

Acceptance gates

Gate	Result	Notes
`npm run build`	green	`tsc` exits 0; no type errors.
`npm run lint`	green	`eslint src` exits 0; the inline `// eslint-disable-next-line no-console` covers the diagnostic-emit line.
5 sequential `npm test` runs	5/5 green	All 1357 tests pass; wall time 20.7–21.3 s per run. See “Run-by-run evidence” below.

Pre-fix (baseline) reproduction

Before applying the fix, the standalone tsx cold-start measurement showed:

Run	Time to `[Startup] Phase 1`
1 (cold)	4205 ms — exceeds the old 3000 ms `spawnSync` timeout by 40 %
2	3055 ms — also over budget
3	1439 ms
4	1799 ms
5	1864 ms

A 10-run pre-fix npm test reproduction (after clearing node_modules/.cache):

RUN 1: 1357 passed
RUN 2: 1 failed, 1356 passed       ← startup — subprocess smoke
RUN 3: 1357 passed
…
RUN 10: 1357 passed

Pre-fix flake rate: ~10 % (1/10).

Post-fix verification

Run-by-run evidence (10 sequential runs, fresh `node_modules/.cache` cleared before run 1)

=== RUN 1 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.012 s
=== RUN 2 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.250 s
=== RUN 3 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.037 s
=== RUN 4 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        20.727 s
=== RUN 5 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.109 s
=== RUN 6 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.180 s
=== RUN 7 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.024 s
=== RUN 8 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        21.036 s
=== RUN 9 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        20.752 s
=== RUN 10 ===
Test Suites: 30 passed, 30 total
Tests:       1357 passed, 1357 total
Time:        20.712 s

Post-fix flake rate: 0 % (0/10) for startup — subprocess smoke.

Build gate

$ npm run build
> colibri@0.0.1 build
> tsc
$ echo $?
0

Lint gate

$ npm run lint
> colibri@0.0.1 lint
> eslint src
$ echo $?
0

Contract compliance

Contract clause	Status
Test passes deterministically in isolation	YES — `npm test -- --testPathPattern=startup` is the original isolated condition; still green.
Test passes deterministically under full-suite parallel	YES — 10/10 green at default Jest `maxWorkers` (~7 on dev box).
Subprocess invocation preserved (`tsx src/server.ts` end-to-end)	YES — same `spawnSync(process.execPath, ['--import', 'tsx', SERVER_MODULE_FS_PATH], …)` invocation. Only the timeout literal changed.
`[colibri] starting` and `[Startup] Phase 1` regex assertions preserved	YES — both regex matches unchanged at lines 822 and 824 of the modified test file.
No `.skip()` / `.todo()`	YES — the test still runs every invocation.
Named timeout constants	YES — `STARTUP_SUBPROCESS_TIMEOUT_MS = 30_000`, `STARTUP_SUBPROCESS_TEST_TIMEOUT_MS = 35_000`.
Diagnostic enrichment on failure	YES — `console.error` emits subprocess `status` / `signal` / `error` / truncated `stdout`+`stderr` only when an assertion is going to fail. Green-path output is unchanged.
No production code changes	YES — `git diff origin/main...HEAD --stat` shows zero src-side .ts files modified (only the test file + 4 docs).

Out-of-band observations

Pre-existing unrelated tools/merkle.test.ts flake. During one transition between cache-clear and warm batches in the verification pass, a single tools/merkle.test.ts failure surfaced (Test Suites: 1 failed, 29 passed; Tests: 1314 passed, 1314 total — note: no 1 failed count, suggesting the suite crashed during load, not during a test). It did not reproduce in any of the 10 contiguous post-fix runs. This is out of scope for this task but worth flagging in memory for whoever picks up suite-wide flake hygiene next.
No production bug surfaced. Per the task escalation clause, the audit explicitly checked src/server.ts, src/startup.ts, and src/db/index.ts for genuine race conditions. None were found. The flake is purely the timing race between the subprocess cold start and the externally-imposed spawnSync deadline. No escalation needed.
Diagnostic enrichment was not exercised in the verification pass — every post-fix run was green, so the on-failure console.error branch never fired. This is the desired outcome (no failures means no diagnostic output). The branch’s correctness is type-checked by tsc (build green) and lint-checked by eslint (lint green); a future regression will be the first real test of its output.

Files changed

.worktrees/claude/debug-startup-smoke-flake/
├── docs/
│   ├── audits/debug-startup-smoke-flake-audit.md           [new]
│   ├── contracts/debug-startup-smoke-flake-contract.md     [new]
│   ├── packets/debug-startup-smoke-flake-packet.md         [new]
│   └── verification/debug-startup-smoke-flake-verification.md [new — this file]
└── src/
    └── __tests__/
        └── startup.test.ts   [edited — +40 / −2]

No src/server.ts, src/startup.ts, src/db/index.ts, or any other production source modified.

Conclusion

The flake is fixed. The contract is met. All three gates (build, lint, test × 5) are green.