²Or rather, cross-validation can only show that the training set is insufficient: it can invalidate the training set, not (really) validate it.