Adopt
ML experiment tracking and model registry. Open source, Azure-compatible, and data scientist-friendly — low adoption barrier for a research-background team. Directly addresses the FDA submission requirement to demonstrate controlled model releases with a full change history.
Can be self-hosted on Azure or use Azure ML's built-in MLflow compatibility, preserving optionality. Chosen over Azure ML (higher operational cost and complexity) and Weights & Biases (SaaS-only, data residency concerns for clinical data). See DEC-006.