المدربIn a shadow deployment a new model receives the production inputs and records its predictions, but its decisions are not executed; its outcomes are computed as if they had been. Champion–challenger is the discipline of keeping the production model (the champion) in place until a challenger, run beside it on the same data, has shown with a test chosen in advance that it is better.
ml_monitor.shadow_run.