The PolyEdge College QB Rating

We built a quarterback rating for the NFL by letting ten years of play-by-play tell us what actually repeats. Then we pointed the same idea at college football — and college fought back. There's no public EPA past 2021, the schedules are wildly unequal, and the sport's biggest award is decided as much by story as by snaps. Here's the rating we got, the problems we had to solve to trust it, and an honest answer to the question everyone asks: can it call the Heisman?

The engine is opponent-adjusted efficiency. Where the NFL rating leans on EPA, the college version is built on ANY/A (adjusted net yards per attempt) because that's the deepest signal available across all ten seasons. Two college-specific fixes do the heavy lifting: an iterative national strength-of-schedule that stops a G5 quarterback feasting on bad defenses from outranking an SEC starter, and a garbage-time discountthat down-weights snaps once a game is decided so a 49–point blowout doesn't inflate the line.

Opens on the 2025 season's top 30 — switch to any year (or All seasons), filter by conference or team, raise the minimum games to drop cameos, sort any column, or expand the limit to see deeper.

The “Heisman factors” toggle — who's best vs who wins

Flip the board above to Heisman factors and it stops answering “who's the best quarterback?” and starts answering a different question: who will the voters actually crown? The two are not the same, and the gap is the whole point. The skill rating is pure production efficiency; the Heisman model adds the things voters reward — team record, conference, a College-Football-Playoff proxy, and raw counting stats — and it's fit to the actual Heisman vote shares of every finalist, 2016–2025, leave-one-season-out validated.

It helps a lot, and we'll be honest about how much. On pure skill, the eventual Heisman winner ranks about 8th on average and lands in the top three only ~37% of the time — the rating keeps preferring efficient quarterbacks the voters overlook. Add the Heisman factors and the winner's average rank more than halves to ~4, with the winner in the model's top three 62% of the time and top five 75% (all out-of-sample). That's a real, validated tightening — but notice it's nowhere near our NFL MVP model, which nails the winner at an average rank of 1.6 and top-three 90%. The Heisman is simply less predictable than the MVP — more of it lives in narrative, “Heisman moments,” and preseason hype that no box score captures.

The toggle makes the divergence concrete. In 2025 the skill rating's best quarterback is Vanderbilt's Diego Pavia — and the model correctly fades him, because Vanderbilt isn't a playoff brand. Meanwhile Fernando Mendoza, 24th on pure efficiency, rockets to the model's #1 once you credit his 16–0 Indiana team — and he won. Watch the arrows: they show exactly how far each quarterback moves between “best” and “most likely to win,” and that distance is the part of the Heisman the numbers can't — and shouldn't — pretend is about quarterback play.