The Rosetta program goes through a complete folding process every few minutes (usually). Thus, the values are often not far off from the final result. However, this is repeated thousands of times for each gene using slightly different perturbations so that the researchers can be pretty sure about the correct final fold for that particular protein. About 10,000 such folds are performed for each gene (not all in one work unit) so that a statistical map can be developed which shows the most likely way the protein will fold.

You can think of this kind of like throwing darts. Lets say the dart throwers know where they are aiming, but you, the observing scientist, doesn't. The scientist will assume the center of the target was in the middle of a region where the most dart holes lie. If however, there are several clearly distinct bunches of holes, then more investigation is needed (more steps in their research).