Yeah that could be it. But in that case what's the big deal? The data is a messy bump on a log, the fit is a smooth bump on a log. That kind of fit is useless for prediction, but they say so explicitly.