Does a look at total accuracy provide us insight into bias for or against a conference? 75% of games that the model is evaluating are within a conference and the SEC historically hasn't played many relevant cross-conference games against other P5 schools and certainly not outside of the south.

I could see a case that FPI is accurate comparing teams within a conference, but could still have a built in conference bias depending on what data it's using. It would be interesting to see how each model does in non-conference games.