FizzBuzz is an incredible filter for devs. Any kind of async JavaScript that needs to do something with the data after it fetches it has also been a winner. Lately the biggest weeding tool has been asking candidates to fetch some JSON from a bucket, then visualize that data. I'd say that maybe half the people manage to fetch the data in 30 minutes, and of that, maybe 10% get the data fetched within 10. The other 20 minutes is trying to help them recognize that their data is initially being rendered with "undefined" and they have to update the viz.
But are you filtering for realistic performance, or for performance under time & observer pressure?
I did a .. bit more complex than fizzbuzz but similar toy problem sort of thing recently, and it's nothing I can't do, it passed the visible test cases (HackerRank if you're familiar, I haven't used it before but I assume it's generally similar) etc. but it was clumsy and far from representative of what I'd actually write on the job I think. Even if I still only spent an hour on it, just not having that pressure. Plus I suppose I would likely have seen the problem before in planning, and realistically it would be something within the context I was familiar with from working on it every day, not the toy problem. (This wasn't it, but imagine something like calculating the possible states of a chess board after each player makes another move. Bit simpler though.)
