Production model monitoring. Input-feature distribution drift, prediction drift, performance metrics (with label-arrival lag handled), data-quality regressions, train-serve skew detection, alerting tied to business SLOs, retraining triggers. ML observability is a superset of service observability — pair with `observability` skill; service-level metrics PLUS model-specific drift signals. ML/AI Engineer owns the monitoring design; Platform/SRE wires the collectors; incident-runbook handles model incidents like any other incident. Use whenever a model has been deployed and needs production monitoring, when a drift alert fires, when retraining cadence is being designed, or wh…
