I think that minimizes the most important and most inarguable aspect of any game. Who won and who lost. We can argue over should have and could have won. We can argue over why the margin of victory was big or small but who won and lost is in stone and it’s the entire reason these games are played.
So, this question cuts to the center of what we want from our rankings.
OAM and a lot of others say, we want rankings that say, this is who we think wins between those two teams. So in theory, on a neutral field, a team on average wins against the team a spot below it, loses to the team a spot above it. This would be a forward looking ranking. We might use data from the past, but we are not rewarding a team for accomplishments.
And it turns out, relying on wins is not a good way to build a forward looking ranking. They're too blocky of a piece of data. Looking at wins, a 1-point win and 40 point win have more in common than a 1-point loss and a 1-point win. Those last two are a stark difference, even though they're likely functionally the same game, give or take a play. So we can't build a great forward looking ranking by only using what amounts to a standing approach looking backwards.
That being said, I don't really see any need for the biggest rankings to be forward looking. I much prefer them to be backwards looking. I like rankings as a comparison of resumes and accomplishments. But that means I have to do away with the idea the No. 1 team should usually beat 2 or 3. It means I have to accept the No. 4 team might really be better than No. 1.
In the end, I think there's room for any of these. I love some predictive ones. I don't mind the resume ones. To be honest, I'm mostly not bent out of shape by the big ones because they're not really held to any standard. I'm sure some use the old poll movement logic, and nothing says they can't. Some try to build their own resumes each week. Some use a bunch of advanced stats and then slightly adjust. And it all kind of comes out in the wash.