A fraud detection model runs in production on a SageMaker endpoint. A new model version must be evaluated using live traffic without impacting production users. Which strategy achieves this?

Shadow testing, also known as dark launching, is the correct strategy because it allows the new model to process live production traffic without its predictions impacting the actual user experience. The production model continues to serve responses, while the shadow model's predictions are logged for evaluation. This enables thorough testing of the new model's performance, latency, and accuracy under real-world conditions without risk to production users. SageMaker Debugger is used for monitoring and debugging training jobs, not for evaluating new models with live inference traffic. Blue/green deployment, especially shifting all traffic at once, would directly impact production users, which the question explicitly states to avoid. While canary-style traffic shifting is a form of blue/green, it still involves a small percentage of live users receiving responses from the new model, which is not ideal when the goal is zero impact on production users during evaluation.