Study Databricks ML-ASSOC UC Registry and Aliases: key concepts, common traps, and exam decision cues.
This lesson is about controlled model management after experimentation. Databricks expects you to know what the Unity Catalog registry adds, how aliases work, and when promoting code is better than promoting a model version.
| Need | Better first instinct |
|---|---|
| governed model lineage across versions | Unity Catalog registry |
| stable pointer to the currently preferred model | alias |
| champion or challenger decision | promotion via alias or controlled version change |
| decide whether to promote code or a model | classify whether the change is workflow logic or trained-model artifact |
| Ask this first | Why it matters |
|---|---|
| are you moving experiment evidence, model lifecycle state, or application logic? | run, registry, and code promotion are different lanes |
| do you need a stable label for the active model? | that points to aliases |
| is the meaningful change inside the trained artifact or in the serving workflow around it? | this is the core code-vs-model promotion distinction |
| Trap | Better rule |
|---|---|
| treating registry as a better experiment log | registry is for managed model lifecycle, not raw run history |
| using aliases without understanding what they point to | aliases refer to model versions, not to runs directly |
| promoting a model when the real change is code logic | sometimes the code change is the thing that should move, not the artifact alone |
A higher-scoring model version is not always the thing that should move forward. If the business change is really about:
then code promotion can be the better answer. The registry helps manage the model artifact, but it does not make code-level changes disappear.
| Scenario clue | Stronger answer shape |
|---|---|
| “team wants a stable production label for the preferred model” | alias |
| “team needs governed named versions across environments” | Unity Catalog registry |
| “team changed scoring logic but not the trained artifact” | code promotion may fit better |
| “team wants to move from challenger to champion” | controlled alias or version promotion |
Registry questions usually test whether you can separate experiment history from promotion and deployment naming. Aliases provide stable labels over concrete registered versions. Promotion decisions may be about code or about the trained artifact itself, and the exam wants you to tell those apart. The weak answer usually assumes every promotion is just “move the model.”