Okay, I think development is going to slow down a bit now, but I spent too much time tonight fixing up the twiddler.cfg parsing, so symbols should work properly (at least, they all worked for me on my old-Tabspice layout). It also reads MCC's, but doesn't link them up to keypresses yet.
And thinking about how to handle training MCC's, it's going to take a fairly decent reworking of the stats, and there are some questions around how to handle them - if you're training on "er", does that also count as "e" and "r"? I suppose that if you're strictly enforcing MCC's for training purposes, then no; you won't actually be using the "e" character, so it won't help your muscle memory on "e".
This also gets back into my musings when I first got the twiddler that MCC's can significantly skew distributions; e.g., if I'm reading my ngrams data right, it looks like >50% of all H's occur as part of a TH digram. So it might actually be more efficient to give "th" a top-level key, and then use the "t+th" chord to produce h. Likewise, most Q's are followed by U, so that should be the easier key there. Though I suppose exploring this would probably be a separate post...