The trick is to find out what works and then when you experiment, only change one factor. That way you'll always be able to diagnose the glitch.