I'm taking time to actually write some code (!) over the next few days, so this will be brief.
Looking at my and/xor production system, it occurred to me that it was an excellent start on representing a Holland classifier system in schema theorem notation. Which prompts the question: it was easy to come up with a rulset for bitwise addition, but can a learning classifier system generalize a ruleset for addition using my machine?
To that end, I am re-writing, modularizing, and generally improving the code. Then, I'll write a genetic algorithm system to try to evolve the rules. I'll post each module here as I get them complete, but I may also start a code project somewhere to make downloads and updates easier.