I haven't thought about the speed at all yet, but seeing that the same decision tree is executed in traditional code as well, I don't think this will be much of a problem. Maybe the Prolog approach will be a performance problem, but first I want to get the syntax convenient before I worry about the implementation too much.