Which statement correctly distinguishes static analysis from dynamic analysis in ML systems?

• A. Dynamic analysis examines code and configurations without running the model.
• B. Static analysis observes the model at runtime.
• C. Dynamic analysis evaluates code during deployment.
• D. Static analysis examines code and configurations without running the model; dynamic analysis observes behavior during execution.

Answer: (D)

Understanding the difference between static and dynamic analysis in ML systems is the key idea. Static analysis examines code, configurations, data schemas, and model artifacts without running the model. This lets you catch issues like syntax errors, misconfigurations, insecure dependencies, and mismatched input/output schemas early, in a fast and deterministic way.

Dynamic analysis, on the other hand, looks at how the system behaves while it’s actually executing. It involves monitoring runtime behavior such as latency, throughput, memory and compute usage, and the model’s outputs under real or simulated workloads. This is essential to validate performance, reliability, and security in practice, including spotting issues like data drift or unexpected predictions during operation.

The statement that captures this distinction—static analysis examines code and configurations without running the model, and dynamic analysis observes behavior during execution—is the best fit because it aligns with how these analyses are fundamentally used in ML development and deployment. The other choices mix up when analysis happens or what is being examined, for example by implying static checks happen without execution but in a way that’s not accurate, or by limiting dynamic analysis to deployment rather than broader runtime observation.