AI model validation checklist
Validation should test whether the model is fit for its intended use, not whether it performs well on a convenient benchmark.
Performance
Measure task quality, calibration, uncertainty, data drift, and segment-level failure.
Responsible AI
Evaluate bias, fairness, privacy, explainability, robustness, and user recourse.
Deployment
Check monitoring, access controls, fallback, versioning, rollback, and approval evidence.