It's brilliant. Love it. I want to know more about the generation, it looks very well thought out. Worth an article in itself.

Thank you! It's been a few years so I can't remember exactly without reading through the code but it's something like this:

It uses notes from the selected scale and octave (from the dropdowns). If the pattern is of an even length, say 16, it will split it into 4 chunks of 4, then randomly decide if it should generate new data for the chunk or copy the previous chunk. It uses the repeat slider for the probability on this.

It randomly applies the 303 modifiers (up, down, accent, slide) using probability set with the sliders on the pattern tab.

There's also an 'empty' slider which sets the probablity of an empty note appearing in a chunk.