A measure of whether a QA model produces consistent answers across different decompositions of the same question.