Design an A/B test for pinned-unread feature

Design a rigorous experiment to evaluate the pinned-unread chat feature. Cover: (a) the unit of randomization and assignment (user_id vs person_id given some persons have multiple user_ids) and how you’ll prevent cross-contamination; (b) eligibility criteria and pre-exposure; (c) primary and secondary metrics (e.g., reply rate within 24h, messages read per day, time-to-first-response, net unread count change), plus guardrails (crash rate, client CPU/memory, API latency); (d) power/MDE and sample size approach, accounting for heavy-tailed engagement; (e) test duration and seasonality controls; (f) novelty/learning effects and ramp plan; (g) handling network effects (group chats) and SUTVA risks; (h) bucketing for users who frequently switch accounts; (i) instrumentation/data quality checks (duplication, clock skew, delayed reads); and (j) a pre-registered analysis plan, including variance reduction (e.g., CUPED) and how you’ll interpret conflicting metric movements.