How should liveness, readiness, and health groups work in Spring Boot?

I design probes from orchestrator behavior backward: failed liveness causes restarts, failed readiness changes routing, and repeated failures can amplify an incident. Health groups therefore have explicit budgets, stable status mappings, and cheap checks that do not create load on dependencies. I test startup, dependency loss, graceful shutdown, and recovery with the actual deployment platform. A green probe is not an SLO; it is only one control input, so latency and error signals still drive alerting.