A supervisor-rating scale is validated against annual performance reviews. Supervisors see employees' scale results before writing the reviews. What modification would best protect the validity study?