What is shadow deployment in classification model rollout?

Shadow deployment runs a new model alongside the production model without affecting live predictions.

In shadow deployment, incoming requests are sent to both production and candidate models. Only the production model’s output is returned, while the shadow model’s output is logged for comparison. This helps evaluate performance, latency, and stability before full rollout.