A permits MLOps lead selected a winning experiment run and needs it available for controlled production promotion. What should they do in Azure Machine Learning?

Registering the winning MLflow model in the Azure ML model registry with a clear version tag is correct because controlled production promotion for permits depends on a durable, discoverable artifact—not an ephemeral experiment run. The registry stores the model with version metadata so MLOps can stage, approve, deploy, and roll back with an audit trail. Selecting a winning run is incomplete until that run’s model is registered for the permits promotion path.

Registering the winning MLflow model as a versioned data asset in the Azure ML workspace store fails because data assets are for datasets, not deployable models with registry versioning, lineage, and controlled promotion gates. Registering the winning MLflow model in the Azure Container Registry with a clear version tag fails because Container Registry stores container images, not MLflow models with registry versioning, lineage, and controlled promotion gates. Deploying the winning MLflow model to a managed online endpoint in Azure ML for production use fails as the controlled promotion step because it skips registry registration, bypassing versioning, approval gates, and rollback controls needed before serving.

Exam caveat: registration is not automatic production approval—stage labels, tests, and responsible-AI evidence still gate endpoint deploy. Ensure the registered package includes the intended MLflow flavor and any required conda/env cues. Operational check: from the winning run, register the MLflow model with an explicit version tag, verify it appears in the registry with lineage to the run, and only then proceed to staging evaluation for the permits service.