It runs under the the GNU Linear Programming Kit (GLPK) like this:
glpsol -m 8642K-vs-7432Q.mod -o outputfile
Thanks for any comments you guys have.
On the sometimes-fold-the-best-hand front, Alex Selby has some weasel words in his README that may be applicable:
BTW exercise: if you run an=0 bl=1 it tells you that the big blind should fold some hands (e.g. 42s) when the small blind has only called, so the big blind could have checked it out for a free showdown, but decides to muck his hand. Why is this not necessarily a bug? (This had me worried before I realised what was happening.) Alex [Answer: because in this situtation small blind should fold or raise, never call. The only way to exploit BB's "mistake" of mucking 42s after a call, is for SB to call with something some of the time. But this presumably works out as a bigger "mistake". This is an example of a degenerate optimal strategy, and the program picking a non-intuitive representative. It could equally well have said "check" with 42s, but it is being confusing because it hasn't yet been instructed not to be. Another (simpler) example of something similar is numbers vs numbers using the parameters d=2 nr=1 an=0 bl=1, where BB may muck 0 after a SB call.]
It's certainly possible that something similar is what's going on in my case as well. I do know that if the pot is large enough those folds go away, and artificially constraining B's strategy to never fold his 76432 and 75432 hands does not decrease his EV in smaller pots.