Jev plays chess
Legal moves as a Choice, scored against reasoning models.
Maxim Saplin had Jev pick from the legal moves in a position, which turns chess into a constrained Choice rather than free generation, and compared where it landed against reasoning models. Illegal moves are impossible by construction here, so the test is about judgment alone.
Why it matters
Constraining the answer space removes a whole class of failure before the model is even asked.