Use when shipping production behavior, adding logs, metrics, traces, alerts, retries, queues, or external calls; when an incident is hard to explain from available telemetry; or when a feature needs operational evidence.
