From reading the their description it has the advantage that it's simple and totally agnostic, it only looks at what a team has actually done as a team based on run differentials and SOS and in direct answer to my question about how far back in time you go to get the best estimate - they regressed on that question to come up with 100 games as the best fit to their dataset. I think the most obvious weakness on a casual view is that since it's purely team based, within the 100 game moving window it assumes a team is a constant. So it has no awareness of roster changes, particularly deadline dealing, which can be pretty huge (think about the impact JV had when he went to Houston!). I guess you can look at it as the concept that is trying to get as much as it can from the least required data.