More random evening noodling: dug back into the focusing front-end to ileansep that I wrote back in like february or something, and tinkered with the test harness a little. I discovered a correctness bug in the translation, whereby it believes it's refuted a => ~~a. Hey now, little theorem prover, I know you're supposed to be intuitionistic, but that's going a bit too far, eh?
Hmm. I seem to have fixed the correctness bugs (I basically just forgot to implement a couple of bits of the translation) but the performance is not too exciting. It only speeds up a small handful of the ILTP problems relative to ileansep, which, if I look at them, are totally the cheating-estly best-possible-case examples for forward chaining focusing.