Whenever you're ready, walk me through how you'd approach this.
Off-policy learning: you have logged data from a model trained on an old policy -- how would you fit a new model to update the policy?
Asked in the onsite round 3 stage. ML theory discussion. Address how you would use logged actions, rewards, and behavior-policy probabilities, evaluate a candidate policy offline, deploy it safely, and monitor for distribution shift and training-serving skew.