The first hurdle a foldit design has to clear to get tested in the lab is getting picked by the scientists to run Rosetta on it. The second hurdle is having Rosetta successfully predict the designed shape from the amino acids. Being predictable by Rosetta is used by the scientists as a proxy for being likely to fold successfully. So making designs that are more predictable hopefully means making designs that are more likely to fold well.
Secondary structure predictors like psipred and jpred don't like all-blue (hydrophilic) sheet strands or all-blue helices - they always predict them to be just loops rather than structures. I assume Rosetta makes the same judgement. So any design we submit with an all-blue strand on the edge of a sheet is unlikely to clear the 2nd hurdle.
In the recent blog posts, all of the edge strands where I can see the tips of the sidechains have one or more hydrophobic sidechains in them. (In the blog pictures, which are rainbow colored, hydrophilic sidechains have red or blue tips, while hydrophobic sidechains are all one color). This supports the idea that successful Rosetta prediction requires some hydrophobics in the edge strands. If Rosetta is predicting well, having the orange sidechains in the edge strands may make the protein truly more likely to fold well. If it is not predicting well, having them there may make Rosetta happy but will not make the protein more likely to succeed in real life.
Foldit is very happy putting all blue sidechains on an edge strand, even if some of the inward-facing sidechains are semi-buried (bright green in core filter view). This is an attribute where the foldit score function and the Rosetta prediction algorithm seem to disagree about what is good. Which one is right? Should we manually replace some of these inward-facing sidechains with orange ones to make our designs more predictable by Rosetta, even if it makes the foldit score lower?
More importantly, wouldn't foldit be more successful at producing predictable designs if its mutate function were more lenient about leaving orange sidechains in positions that are semi-buried? This doesn't just mean putting orange sidechains where they don't belong. Keeping the sidechains orange when they are in a borderline state between hidden and not hidden starts a virtuous cycle: If an edge sidechain is orange, it tends over time to twist the strand inward, making that place even better for an orange sidechain. If the sidechain in that spot is blue, it tends to twist the strand outward, making it better for blue and worse for orange. All the other changes (including sequence changes) made to the protein will adjust to this better-for-orange or better-for-blue position. Leaving more orange sidechains in place will over time result in a sequence for a protein with a better core, which should be more likely to fold well in real life, not just in Rosetta.