A model is deployed in shadow mode alongside the production system. Both the production system and the shadow model receive all live traffic. The shadow model's predictions are logged but never shown to users. What is the purpose of shadow mode deployment, and what limitation does it have?
A Shadow mode is used to A/B test models — half of users see the shadow model's results B Shadow mode validates a new model's behavior in production conditions without user exposure risk — it captures real production traffic distributions, real latency behavior under load, and real edge cases not present in offline test sets; the shadow model's predictions are compared to production predictions or to eventual ground truth labels; the limitation: shadow mode measures prediction distribution and latency but cannot measure user impact (conversion, satisfaction) because users never see the shadow model's output; it validates technical correctness, not business value C Shadow mode is used to monitor for data drift on the production model — not related to testing new models D Shadow mode eliminates the need for A/B testing — if shadow model predictions match production, it can be deployed without further testing