The way Feynman put it was that your theory shouldn't just make the observation that produced it "come out right" but should make something else "come out right" as well