One issue is that tests are usually based on limited samples of behavior; we cannot ask every possible question or observe every instance of behavior.