An organization manages multiple versions of a fraud detection machine learning model in Model Registry. The engineering team wants to streamline the model release process so that downstream serving applications can automatically consume newly promoted models (such as staging or production) without needing application code or configuration updates whenever a new version ID is generated. Which approach should the team implement in Model Registry?

Assign a mutable model alias to the target model version and configure downstream applications to reference the model using the alias name.: ### What model version aliases are A **model alias** in Model Registry is a mutable, named pointer assigned to a specific version of a machine learning model. Similar to Git branch references or Docker tags (such as `:latest` or `:stable`), an alias allows users and automated systems to identify, deploy, or query a model version using a human-readable text string rather than an immutable, generated version ID. ### How it meets the requirements - **Environment and lifecycle designation (✓):** Custom aliases like `staging` or `production` can be assigned directly to specific model versions to clearly demarcate their lifecycle stages. - **No code changes required (✓):** Downstream client applications can be hardcoded or configured once to reference the model via its alias (e.g., `fraud_detector@production`). When a new model version is validated, moving the alias to the new version instantly redirects application traffic without modifying serving code. - **Simplified promotion workflow (✓):** Moving an existing alias to a new model version automatically removes it from the prior version, ensuring atomic and straightforward model promotion. ### Key advantages and capabilities - **Mutability:** Aliases can be reassigned across versions dynamically as models progress through testing and production stages. - **Human-readable references:** Eliminates the need to memorize or look up lengthy numerical or hash-based version identifiers. - **Automatic default fallback:** The first version of any registered model automatically receives the `default` alias, ensuring standard calls without explicit versions function smoothly. - **Strict uniqueness per model:** An alias can only point to a single model version at any given time within a model resource, preventing conflicting routing. ### Why this is optimal Using model version aliases decouples model deployment and serving infrastructure from model retraining pipelines. Data science teams can train, test, and promote models independently while client applications consistently query stable alias endpoints.