/api/healthLiveness, database readiness, and which build is running
Doubles as a readiness probe: it pings the database and returns 503 when the DB is unreachable, rather than a cheerful 200 that hides a dead database. The only endpoint here that needs no session.
It also reports the running build (version.commit / version.branch / version.startedAt), so "is my change live?" is one unauthenticated curl instead of an SSH session — which matters for anyone who deploys but has no shell on the box.
startedAt is the PROCESS start, not the deploy. A current commit with an old startedAt means the code was updated on disk and nothing restarted — the exact failure that once made a deployed route look like a 404.
The block rides on the 503 too: a failing server is when you most want to know which build is failing. It is absent entirely when the commit cannot be resolved, never sent as null — a null would let a monitor match null against null and call it a successful deploy.
A 503 is returned: The database did not answer.