One thing to watch out with these sorts of approaches is you are actually improving on an exhaustive search of all possibilities (perhaps with some smart pruning).
What do you mean watch out? Isn't "improving on an exhaustive search of all possibilities (perhaps with some smart pruning)" a way of saying global optimization? Unrelated, great job with JuliaOpt, if that is you.
I guess what I'm getting at is a GA with 100% mutation is basically randomly generating solutions and evaluating them. Now, no one does that, but if your GA eventually explores a large amount of the search space in order to get a good solution then it suggests that its not the right approach. So yes, some sort of "global optimization" where you acknowledge you are basically going to look at the whole domain but avoid the silly cases/provably worse cases would be the alternative.
