Days 1-2 examined "is the model correct (drift)"; today we ask "is the system healthy". Even accurate models mean nothing if endpoint response slows, instances die from overload, or calls fail with 5xx errors—to users it is broken service. Operational monitoring' heart is Amazon CloudWatch. MLA-C01 asks "how do you track operational issues via metrics, trigger alarms, and auto-respond?"
CloudWatch provides three things: Metrics (time-series numbers), Logs (text records), Alarms (threshold-triggered auto-response). Adding AWS X-Ray for distributed tracing completes the picture.