Skip to content
ML Academy

Introduced on day 19

Shadow deployment

Running a candidate model on live inputs beside the incumbent without letting its output affect any decision, to test plumbing, latency and agreement.

In detail

Shadow mode shows whether the candidate runs, how its outputs compare with the incumbent and whether features match training. It cannot prove business impact because its decisions are never acted on; that needs a canary or an experiment.