System design interviews feel open-ended in a way algorithm interviews don't — there's no single test case that tells you whether your answer is "right." That's not an accident of a badly written question; it's the entire point. Every real system design problem is a bundle of trade-offs between competing constraints — consistency against availability, latency against cost, simplicity against scale nobody has asked for yet — and the interview is measuring whether you can navigate those trade-offs out loud, not whether you can recite one specific architecture from memory. This is a complete reference: work through it in order to build the mental model from scratch, or jump to the section — or one of the three fully worked examples near the end — you actually need.