A measure of how unusual or unreliable a prediction is, used by conformal methods to decide which predictions to include in the answer set.