The thing I particularly like about this is that the problem spec is very rigid and clear: no generic solutions (ints only), don't worry about balancing, don't

The thing I particularly like about this is that the problem spec is very rigid and clear: no generic solutions (ints only), don't worry about balancing, don't worry about visibility, make it work, and then they give example usage of the API. It's really easy to screw up a one-page toy problem spec: I've done that myself, and got many replies that failed to either follow the spec or failed to identify ambiguities in it. Having a simple problem as a baseline is really helpful. That said, what I don't like about this is that it doesn't really test the candidate's ability to reason about systems or architectural things--then again, as a better "fizzbuzz", this seems to fit perfectly.