As I stated in my journal, I just recently discovered an awesome programming game titled Core Wars. I've been playing with the it for a week or two, and I've decided to write my own evolver for the game. Many different types of mutation are obvious, but I'd like to discuss other methods of evolving. It seems to me that RedCode is uniquely difficult to represent in a way useful for quality crossover or other graph-based approaches because of the way each instruction can modify the next (or previous, etc) in a way that makes the program as a whole act like an almost-atomic unit. Do you have any ideas on how operators other than mutation could be implemented in a way that provides useful results any significant portion of the time (vs making a program take a huge step backwards)?