An experimental technique that tests whether a feature responds differently to target concepts versus carefully chosen control examples.