We had a lot of difficulty getting the strands to orient the way we
wanted (with the conserved polar residues inside the barrel).  Adding
constraints to the undertaker cost function was never strong enough to
get the strands to flip over.

The problem, apparently, was that our set of initial alignments did
not have one with the strands oriented correctly, or models built from
such an alignment had other serious problems that kept it from scoring
well enough to be kept.

What we did to get correctly oriented strands was to make a structural
alignment (using VAST) for a good barrel (try7-opt2) that had a couple
of strands misaligned.  We then took the alignment (to 1pz1A) and
shifted three of the strands (by 1, 3, and 1) in the alignment to flip
them over.  The result was optimized with a cost function that had
constraints to hold the strands the way we wanted them.

After some optimization, we have try19-opt2, which we submit as model 1, 
as our best-scoring model with an unconstrained cost function.

Note: we are still worried about the burial of T111 and E139, but
don't have time to do more realignment and optimization.