grim7/14/2026, 5:58:17 AM

A down API silently produces no run — the harness would push, then wait out the
full verdict timeout. Add a bounded /healthz probe (TEST_API_URL, defaulting to
VITE_API_URL) after the DB ping: an unreachable ingest fails fast with "start it
(bun run dev:api) … a down API means no webhook lands, so no run". The worker has
no HTTP surface, so its silence still surfaces as the no-run-in-60s timeout —
documented, not faked.

Parent370a8ac