What observability baseline would you expect for a production Spring Boot API?

Every alert has an owner, severity, runbook, and tested route to a mitigation decision. Cardinality and sampling are budgeted so observability does not become a production dependency failure of its own. I validate telemetry during load, partial dependency loss, restart, and deployment rather than only on a healthy laptop. The baseline evolves from incidents: missing evidence becomes a concrete instrumentation improvement, but extra signals are rejected when nobody can name the question they answer.