The same way I verify my own proofs of textbook exercises: very cautiously. Subtle errors are a feature of the problem domain, not a new novelty.