Question
Design the 'already-seen' / dedup and pagination-consistency layer for an ML-ranked feed serving 200M DAU. Because the feed is re-ranked on every request (not a static list), classic offset pagination breaks: as the user scrolls, items shift rank and they see the same post twice or skip posts. You must guarantee a user never sees the same item twice across a session (and ideally for days), across multiple devices, while keeping the seen-set lookup off the critical path of a 100ms feed request. Design the seen-state store and the pagination model.
Clarify scale and constraints first. Propose a clean component breakdown, then go deep on the hard parts — data model, bottlenecks, consistency, failure modes — and name the trade-offs you are making.