Another slight caveat -- as well as predicting things which should work, it should also predict what won't work. If your theory is that your medicine cures an illness, it's not enough for people who take it to get better - people in the control group who don't take it must not get better.