Question 24
UnclassifiedWhat does a shadow deployment of a model mean?
Correct answer: B
Explanation
A shadow deployment means the model is run alongside the live system so it can "score production traffic in parallel" while users still receive the existing model’s outputs. It is used to evaluate behavior on real traffic without "serving its predictions to users," which lets teams compare performance safely before release.
Why each option is right or wrong
A. The model runs only at night to reduce infrastructure costs.
B. The new model scores production traffic in parallel without serving its predictions to users.
A shadow deployment is the standard pre-release validation pattern in MLOps where the candidate model is attached to live inference traffic but its outputs are not exposed to end users; the production system continues to serve the incumbent model. In practice, this means the new model receives the same requests in real time and can be evaluated on actual production inputs, latency, and drift without affecting user-facing responses, which is why it is distinct from an A/B test or canary rollout.
C. Only the data science team can access the model for testing.
D. The model is encrypted at rest to protect stored parameters.