The readiness probe example stands out: a pod reporting 0/1 because it probed /does-not-exist shows how a small probe misconfiguration can look like an application failure while the container itself is fine. The note that the LoadBalancer IP stayed stable when scaling from two to four replicas is also a useful check, since it separates the Service from the replica count. The point about keeping AI-assisted triage read-only is the right guardrail — gathering and analysing is helpful, but letting a script act on a cluster without a human in the loop would undo the safety the rest of the walkthrough builds in.