Health dashboard for EPS services. Polls ~/.epc/services.toml on an interval, stores results in ~/.epc/observatory.db, and serves a live status page.

Stale service entries — Observatory persists service state in its own SQLite database (~/.epc/observatory.db). If a service is removed (deleted project, epm services stop + manual removal from services.toml), Observatory has no way to know — it will continue showing that service with its last known status indefinitely.

The right fix would be a reconciliation pass that drops DB entries for services no longer in services.toml, but that's risky: it would silently delete historical health data and logs for services that are just temporarily stopped or being debugged. For now, stale entries have to be cleaned up manually:

sqlite3 ~/.epc/observatory.db "DELETE FROM service_state WHERE service = 'name'; DELETE FROM health_checks WHERE service = 'name';"