You must be using an LLM that cannot navigate formal logic puzzles or hasn't undergone chain-of-thought optimization.