Expand description
GET /health. Credential-free upstream, and the endpoint every bring-up script polls.
It reports liveness only, and that is a known gap rather than a pending one: storage has landed, so this must learn to distinguish “process up” from “database reachable”. A health endpoint that returns ok before its dependencies are connected turns every startup race into a mystery test failure, because the harness stops polling exactly when the server is least ready.
Until it does, the one property it has to keep is that it answers without resolving a request context. A health check that loaded schemas would fail exactly when a caller most wants to tell “process up, database down” from “process down”.