The Case for Validity Generalization