https://ww2.amstat.org/mam/2010/essays/PasteurPredictive.pdf
No one worth their salt is assuming superiority or weighing conference or team expectations in final results; that would defeat the entire purpose of computer models.
Like I said, here's the main take away:
Start each season with teams having unequal ratings, based on results from the previous season, to improve early-season predictions, but diminish the effects of the initial inequalities as more games are played.
Eventually the initial inequalities are fully removed.