The system design answer that finally clicked was admitting I didn't need to shard anything at the scale they described

Every mock system design round I'd done before this reached for sharding, queues, and a cache layer within the first five minutes, mostly because that's what every YouTube walkthrough does. This time the interviewer gave me a number — a few thousand daily active users — and I actually paused and asked if that number was firm before designing anything. It was. Ended up with a single Postgres instance, a read replica, and nothing else, and spent the rest of the round going deep on the schema and indexing instead of drawing boxes. Got told afterward that most candidates never ask about scale and just default to "for millions of users" architecture regardless of what's stated. Small thing, changed how I approach every round since.