It'd clear that this one isn't being scored in a traditional way - is there any chance someone from FoldCentral can give some more details?
This puzzle definitely is scored a bit differently, I think I mentioned it briefly in the description of the puzzle, but I should have made it bold or something. Sorry.
Basically, there are three scaffolds in this puzzle. When you reset the puzzle, it randomly chooses one of them. The score that shows up at the top reflects only the score of the ligand from the best scoring scaffold.
There is also a condition which requires that the entire design (including the score of the side chains) needs to be within a certain score (or better) of the original starting structure. This is meant to prevent crazy clashes within the protein which may (for some reason) make the ligand really stable at the cost of the rest of the protein's stability.
Also, the automatic features like mutate all and shake sidechains works on the score of the entire protein. (Some may notice that these features sometimes makes your score worse, that happens when a mutation is found that makes the score of the entire protein better, but makes the ligand score worse).
The goal of all these restrictions is to get the best design that binds that ligand as tight as possible without making the rest of the scaffold too energetically unfavorable. Additionally, the backbone is fixed in place and will not move. This is because the designs (when we actually go and make them) are more reliable when we don't use the backbone movement algorithms.
I hope that clears up the scoring. Let me know if you have any other questions!