Stories explain a request lost while a pod stops
When probe errors fall between a stopping pod's SIGTERM and its removal from the EndpointSlice, the story's takeaway says so and why (nodes update their rules after the pod is marked terminating; play-app exits 0.4 s after SIGTERM without serving the connection; a preStop delay avoids it).
Checked against the live site's data before the change: the one error of a rolling restart (1/72) was attributed to the stopping pod, and no other story matched.