We had no good hits for this target, so were reduced to using new-fold
methods.  The contact predictions were weak.

This is try22-opt2, the best-scoring model with an unconstrained undertaker
cost function.