Iowa won an in-state rivalry game and ESPN's FPI responded by marking the Hawkeyes down, which sounds like an insult and is really just a machine reading a different box score than the rest of us. The ESPN FPI Iowa conversation flared up this morning because a win is a win in the standings, and it is emphatically not a win in an efficiency model. Both of those things are true at the same time. That is the whole story.

Why did ESPN FPI drop Iowa after a win?

FPI is a predictive rating, not a report card. It is trying to answer one question: if these two teams played again on a neutral field next Saturday, what would happen? To do that it leans on how a team performed relative to opponent quality, down to the efficiency of the drives rather than the final line in the agate. Win by a field goal when the model expected you to win comfortably and you can come out the other side rated lower than you went in, trophy in the case and all.

So Iowa did the thing every Hawkeye fan in the state wanted, and the algorithm shrugged. I don't have the full FPI table in front of me to tell you the exact drop, and I'm not going to pretend I do. The mechanism is what matters here, and the mechanism is margin against expectation.

Rivalry games break efficiency models on purpose

Here is where I'll stop defending the machine. The in-state rivalry on Iowa's schedule is the Cy-Hawk, and that game has never once cared about looking pretty. Rivalry football is a genre unto itself: extra film study on both sides, kids who grew up choosing between these two schools, coaching staffs who know each other's tendencies down to the personnel package. The spread of outcomes is enormous and the quality of play is often worse than either roster's talent suggests. A model built on the average Saturday is going to keep getting this one Saturday wrong.

checked the almanac: iowa and iowa state spent decades not playing each other at all before the series came back for good in the 1970s, which is its own quiet reminder that this game was always about the state, never about the math.

That is not a flaw you can code out easily. FPI is not built to know that a punt-heavy, field-position knife fight in September is exactly what Iowa football has been, under this staff, for a very long time. The Hawkeyes have won a lot of games in the 17-13 neighborhood. A model that punishes that style is going to be permanently annoyed with Iowa, and Iowa is going to be permanently fine with that.

Polls versus models after Week 2

You can see the same split playing out at the top of the sport this week. Texas moved to No. 1 in the AP poll after handling Ohio State, because humans reward the win and the résumé. Oregon lost 39-31 at Oklahoma State and is now getting picked apart by national analysts wondering what is going on inside the program, because humans also reward and punish vibes. The models do neither. They log efficiency, adjust for opponent, and move on without a single feeling about the trophy case.

Neither system is lying. They are answering different questions. The poll asks what you have done. The model asks what you are. Those answers diverge most violently in exactly the games fans care about most.

What comes next for Iowa

The thing Hawkeye fans should sit with: nothing FPI thinks changes the standings, and the standings are the only document that decides who plays in December. If Iowa keeps grinding out low-scoring wins, the rating will lag and the record will not, and eventually the rating catches up because that is what ratings do with enough sample.

What I'm watching is whether the offense gives the model anything to chew on. Efficiency metrics forgive an ugly rivalry night quickly if you follow it with a clean one against a defense that can actually be moved. If Iowa's next few results are more of the same field-goal margins, the gap between how the Hawkeyes are seen and how they are rated is going to be the running story of their season. A win is a win. It just isn't always a data point.