Identifying misalignment by finding contradictions in a model's reasoning across equivalent scenarios with different framings.