A public-health lab finished cleaning lab-results data and must promote exploration v1 to a production training input without renaming folders by hand. What practice should they use?

Publishing a new version of the Azure ML data asset (v2) and pointing production training at that version is the correct promotion practice because cleaned lab-results data should advance through immutable versions rather than destructive renames. Exploration v1 remains available for lineage and comparison while production training explicitly consumes v2. Public-health labs gain a clear promotion gate without hand-editing folder names or losing history.

Overwriting the v1 Azure ML data asset with the cleaned results and keeping production training pointed at v1 destroys immutability: it erases the exploration snapshot, so you lose the v1-to-v2 lineage, audit trail, and rollback safety this scenario requires. Publishing a new version of the Azure ML data asset (v2) and keeping production training pointed at the v1 version creates the right version but fails the promotion: production training would still consume the uncleaned exploration data instead of the cleaned v2 input. Creating a separate Azure ML data asset named cleaned-results v1 and pointing production training at that asset fragments lineage into a disconnected asset name instead of advancing the same asset from v1 to v2, so you lose the explicit promotion gate and version history without avoiding manual renaming.

Exam caveat: promotion means a new data-asset version plus updating the consumer pointer—not publishing v2 while leaving production training pointed at v1.

Operational check: publish v2 from the cleaned lab-results set, update the production training job or pipeline parameter to data asset v2, re-run a validation job, and verify v1 still exists for audit and comparison.