You need a method that can't be distinguished from the native pattern. How much trust do you want to put in your algorithm being invulnerable to future analysis?