The answer to that is above my pay grade. But it isn't above theirs. All I know is when your models concerning one team in particular are pretty drastically wrong year-after-year, and you are a website that prides yourself on accuracy with your predictions, something is pretty demonstrably broken.
They'd probably hate the suggestion, but it likely starts with adding some sort of automatic positive regression to prospective Brewers rotation members. The teams' pitching development mechanism, at this point, has proven that it can regularly out kick its coverage.