For this model we guessed that there would be 3 disulfides (later
confirmed by information found by Alexey Murzin and announced on
FORCASP). 

We tried all 15 possible pairings and built models with constraints
for each.  The better models were further optimized.

We have decided to submit the best model we obtained for each of the
top 4 pairings, sorted by how well they scored on an undertaker cost
function that did not know what pairing was used, but which rewarded
any ssbonds found.

This is model try32-opt2.