A model's ability to change its predictions based on whether evidence supports or contradicts a claim.