This is try6-opt2, our best-scoring model created by undertaker.

Because undertaker does not understand multimers and has no way to
freeze residues, we had to add scaffolding constraints to try to keep
the 3-strand sheet in place.  We added constraints for P38,C49,and E77
to Q2,M9, and D20.  These constraints are not quite right, as we could
see by the clashes and missed hydrogen bonds of the dimer formed by
superimposing try6-opt2 on 1ml8A and 1ml8B.

If we had had more time, we would have adjusted our scaffolding
constraints for more precise positioning of the sheet, but we were
feeling the press of other targets and left this one as "good enough".
The CORRECT solution---making undertaker understand multimers---would
taker far too much time during CASP season.