After looking at the two native folds that were just released for CASP refinement puzzles, it appears that the portions needing attention did not need a great deal of movement away from the best predicted fold. Delivering refinement puzzles to us with sections stretched out straight forces us to make radical changes from the prediction in those areas, which in turn tends to push other portions of the protein away from the prediction. Perhaps the portions needing change could be delivered to us in the shape predicted, only slightly displaced (translated and/or rotated a bit), with short cutpoints at either end. This would ensure that we do not just duplicate the given prediction, but would be a better start than making those sections into a de novo challenge. If slight movement would bias us too much toward a single solution, you could rotate the cut portions 180 degrees, forcing us to choose an angle to reattach them, while still giving us the information of the shape they are predicted to have.
You are absolutely correct, Susume.
We will bring this up at this week's Foldit meeting.
Until we get this implemented, the only other Refinement target where they have specified the regions (where we gave you an extended start) is TR679 (Puzzle 584). We will post another puzzle for that one where we just give you the starting model to refine (with no stretched out regions).