User Engagement Patterns

A user analytics pipeline classifies streaming session engagement and rolls up user-level behavioral segments — a foundational step in retention and recommendation modeling. Given a list of session records, perform a three-part engagement analysis. **Input:** A list `sessions` of dicts with `user_id`, `title_id`, `watch_time`, and `title_runtime`. **Output:** A 3-tuple: (1) list of per-session engagement labels, (2) dict mapping each user to their most frequent label, (3) dict counting users per label across all three categories. **Key Rules:** - Watch percentage = `watch_time / title_runtime * 100`. - Label thresholds: ≤25% → `'Abandoned'`; >25% and ≤75% → `'Sampled'`; >75% → `'Completed'`. - On ties for a user's most frequent label, choose alphabetically first (`Abandoned < Completed < Sampled`). - Part 3 counts users (not sessions); always include all three keys even if count is 0.