# Predict Next-Week Streaming Engagement at Event Scale Design an ML system that predicts whether a streaming user will be active next week. Discuss label choice versus duration or an engagement score, features from region, tenure, platform, sparse artists, lifetime statistics and event history, leakage prevention, and modeling at trillions-of-events scale. ### Constraints & Assumptions - Features must be available before the prediction cutoff. - The prediction must support a defined product decision. ### Clarifying Questions to Ask - What action consumes the score and what error is costlier? - How is activity labeled for new users? ### What a Strong Answer Covers - Target validity, temporal splits, scalable aggregation, baselines, sequence models, and monitoring. ### Follow-up Questions - How would threshold drift be handled? - When would a Transformer be justified?