it's pretty clear what it does. It uses the evolutionary information expressed in multiple sequence alignments to make reasonable judgements about interatomic distances, which are used as constraints for a force field. We've been doing variations on this for decades. The evolutionary information encoded in multiple sequence alignments is pretty much all you need to fold homologous proteins (apparently). No, this technique doesn't do anything about the harder problems of actually understanding the raw physics of protein folding (nor, does it seem, that we need that to solve downstream problems).