Question
Design a multi-armed-bandit experimentation platform that product teams use to run hundreds of concurrent experiments. Each experiment can be a fixed A/B split or a bandit (Thompson sampling / epsilon-greedy) that shifts traffic toward winning variants automatically. Assignment must be deterministic and consistent per user (a user always sees the same variant within an experiment), sub-5ms at the SDK call, and correct under overlapping experiments. The platform must report per-variant metrics with valid statistical inference and let teams safely ramp, pause, or hold out a control.
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.