If you attach it, I'd be happy to take a look. If it seems random, try taking any trials randomization OUT of the input file and run the trials in a constant order--does that make the freeze more predictable? If so, it may help narrow the search for some problematic syntax.