5. ML Solution Monitoring, Maintenance, and Security (24%)

This domain catches candidates off guard. At 24%, it's the second-heaviest weight on the exam, yet most study time goes to model development. The exam reflects a production reality: models degrade silently, infrastructure costs spiral without oversight, and a single misconfigured IAM role can expose sensitive training data. Phase 5 covers three interlocking concerns — monitoring model health (5.1), optimizing and observing infrastructure (5.2), and securing the entire ML pipeline (5.3).

The three sections address different failure modes. Section 5.1 asks "is the model still accurate?" through drift detection, Model Monitor, and A/B testing. Section 5.2 asks "is the infrastructure healthy and cost-efficient?" through CloudWatch, X-Ray, and cost optimization. Section 5.3 asks "is the system secure and compliant?" through IAM, VPC configuration, encryption, and audit tooling. Exam questions frequently combine these concerns — a scenario about rising latency (5.2) may have a root cause in data drift (5.1) or a VPC misconfiguration (5.3). Approach this phase as an integrated operations discipline, not three independent topics.