It’s mostly margin of victory/defeat based, with various multipliers based on the strength of opponent (so blowing out an FCS team is worth less than losing by a few scores to a typical P5 team). That includes a progressively larger coefficient the higher the AP ranking of the team that you beat. There’s also a decay multiplier that slightly lowers the weighting of older games as the season goes on.

The weirdness this week is really because it’s cumulative. That is, I don’t average the results by amount of games played, so Notre Dame is higher than the two teams they lost to because they’ve played more. That’s also why teams drop like crazy on their bye week (and this is the time of the season where those tend to occur, but not all at the same time). It works out pretty well by about week 8. (And anyone could take my data and normalize it themselves if they wanted something less “interesting”)

I definitely would do things differently by eye test, but the algorithm does produce interesting insights that I wouldn’t otherwise notice (I watch a lot of cfb, but can’t watch all the sicko games). I think of it as a sort of power rating rather than a proper ranking.

I’ve been tinkering with this formula for about 5 years, but am thinking of overhauling it next year with all the realignment that’s occurring…