model-evaluation

A step-by-step playbook for testing pre-release AI models without exposing secrets. Use sandboxes, synthetic data, telemetry, and an incident checklist.