It's half term and I have been away. However, I have been tinkering a bit. I made a program that models 2D collisions between balls (link to come) and I have made some significant improvements to Harmomiza, ahead of its proper launch.
Previously, Harmoniza used a big chunk of luck to find the best arrangement of groups. Leaving it running for a long time gave you better results, but only because the longer you left it the more likely it was to stumble upon a good solution. The new version uses an evolution algorithm. It still uses the same semi-random method to find a base solution, but then switches to a system which makes small changes and tests them to see if they make an improvement. If the new form lists are better, they are saved and the process continues from there. If the "mutated" version is no better, it is discarded. This means that leaving the program running allows the score to "ratchet" up. Already it is giving good results in less time than before, and is much better than the pen-and-paper method.
I think we're ready to go live!