Enforcing a Manual Model Approval Gate in a SageMaker Pipeline Before Production Deployment

Why the Answer Is Correct

The requirement is a manual approval-based workflow that ensures only approved models reach production endpoints, and the approval state itself lives in the SageMaker Model Registry, where each registered model version carries an approval status. A SageMaker Pipeline can register a model version and then, through the AWS SDK, change that approval status to Approved, which is the documented way to update approval programmatically. Because the pipeline consults that status, an unapproved version cannot pass the gate, so the approval workflow is enforced structurally rather than by convention. The vote was unanimous at 100 for D. GiorgioGss quoted the SageMaker documentation stating that you can update the approval status of a model version by using the AWS SDK, and tigrex73 and Laxma99 both described the registry providing the functionality to manually or programmatically approve models for production deployment.

Why the Other Options Are Wrong

Using SageMaker Experiments to facilitate approval during model registration (A) is a category error, because Experiments organizes and tracks experiment runs and their metrics, and it has no approval state to gate a deployment on, as ninomfr64 noted. Using SageMaker ML Lineage Tracking and creating tracking entities for the approval process (B) has the same defect, since lineage tracking records relationships among models, datasets, and artifacts and cannot authorize a deployment. Using SageMaker Model Monitor to evaluate model performance and manage the approval (C) confuses monitoring with authorization, because Model Monitor watches deployed endpoints for quality, bias, and drift, and its findings can inform a decision but cannot record or enforce an approval status.

Community Comment Notes

The community was unanimous at 100 for D, and ninomfr64 explicitly flagged that the option is not clearly worded while still identifying it as correct, then walked through all three rejected options on exactly the grounds that approval state belongs only to the Model Registry. S_201996 described the intended design, noting that you can define a pipeline step where a manual approval is required before the model status changes to Approved for deployment. The consensus rests on the same point the documentation makes: approval status is a Model Registry field, and the AWS SDK is the supported way to change it.