At one point in optimizing the models for this target, we had a model
that was the best-looking in domain 1 and another that was the
best-looking in domain 2.  We made a chimera by pasting the two
domains together from the two models (try6-try4-chimera), which then
scored better than either parent model, and needed only minor
polishing to produce our final model.

This is try7-opt2, our best scoring model with an unconstrained
undertaker cost function.