We rated every college quarterback. Then we learned the Heisman isn't a quarterback award.

A few weeks ago we rebuilt the NFL quarterback rating from scratch and let ten years of data tell us what makes a passer good. The obvious next question: does the same idea work in college? Who, by the numbers, are the best quarterbacks in the country?

College said: not so fast. Three things make it harder than the NFL. There's no public play-by-play EPA past 2021 — the metric our NFL engine runs on simply doesn't exist for recent seasons. The schedules are wildly unequal — a Group of Five quarterback can feast on defenses an SEC starter never sees. And the sport's marquee individual award, the Heisman, is decided as much by narrative as by production. We had to solve all three before we could trust a number. Here's the honest version.

The full sortable leaderboard — every qualified FBS season, 2016–2025, plus the Heisman model — lives at the PolyEdge College QB Rating page.

A different engine, because EPA isn't there

In the NFL the engine is EPA. In college, for seasons past 2021, it doesn't exist publicly — so we built the rating on ANY/A (adjusted net yards per attempt: yards, plus a touchdown bonus, minus a heavy interception penalty, per dropback). It's the deepest efficiency signal available across all ten seasons. The blend, scaled like passer rating to a mean of 100:

College QB Rating = 0.55·z(opponent-adjusted ANY/A) + 0.15·z(completion %) − 0.15·z(sack rate) + 0.12·z(rushing yards/game)

Same philosophy as the NFL rating — a fixed blend, no fitting, because the forward-fit version overfits and collapses out of sample. But two of those words — opponent-adjusted — are doing an enormous amount of work, and they're the whole reason the college number is trustworthy.

The two fixes that make it real

Fix one: who you played. Raw college stats lie. Run the numbers unadjusted and the leaderboard fills with Group of Five quarterbacks who put up video-game lines against defenses that couldn't tackle. So we built an iterative national strength-of-schedule — an SRS-style model that solves every team's offense and defense rating simultaneously, iterating to convergence, then discounts each quarterback's production by the actual defenses he faced. It's the single most important college-only adjustment. Without it, the rating is fan fiction.

Fix two: garbage time. A 49-point blowout inflates the winning quarterback's line with snaps that didn't matter. The naive fix — delete them — is worse than the disease: hard-filtering by win probability dropped dominant quarterbacks (Tua, Lawrence) below the qualification line, because they were the ones sitting in the fourth quarter with the game won. So we don't delete garbage time, we discount it — leverage-weighting each snap down to about a quarter of a competitive one — and we qualify on total dropbacks so nobody is punished for volume.

Do both, and the leaderboard passes the eye test. The best single seasons since 2016: Jayden Daniels (2023), Baker Mayfield (2017), Kyler Murray (2018) — three Heisman winners, in that order — then undefeated UCF's McKenzie Milton and Jalen Hurts' 2019. When consensus is strong, the rating agrees with it. That's the bar a new metric has to clear before its disagreements are worth anything.

Then we asked it to predict the Heisman. It stumbled.

Here's where college stopped being a translation exercise and started teaching us something.

We ran the rating against the eight quarterback Heisman winners from 2016–2025. Across the full field of ~130 qualified FBS starters a year, the eventual winner landed at an average rank of about 8th, in the rating's top three only 38% of the time. For contrast, our NFL MVP model nails its winner at an average rank of 1.6 and top-three 90%. Either the college rating was broken, or — more interestingly — it was measuring the wrong thing on purpose.

It was the second one. The Heisman is not a production award. It rewards an undefeated record, a national-title chase, a blue-blood stage on Saturday night — exactly the things a quarterback rating is built to ignore. The rating wasn't failing. It was telling us, precisely, what it refuses to see.

So we gave it eyes — and fit it to the actual votes

If the trophy rewards team context, model the team context — and don't guess at the weights, learn them from the ballots. We rebuilt this as a vote-share model: the efficiency rating plus the things voters actually reward — a College-Football-Playoff proxy (each team's end-of-season Elo rank), team record, conference, and raw counting production (yards and touchdowns) — fit to the actual Heisman vote shares of every finalist, 2016–2025, and leave-one-season-out validated.

Every figure is out-of-sample — each season scored by a model that never saw it, so the lift is real, not a curve fit to eight winners.

That's a genuine, validated tightening: the model more than halves the winner's average rank. Fernando Mendoza 2025 rockets from 24th to the model's #1 once you credit his 16–0 Indiana team — and he won. Bryce Young 2021 climbs from outside the top fifteen to #1 on Alabama's playoff run. Murray, Burrow, and Daniels stay near the top.

But here's the honest headline, and it isn't tidy: the Heisman model tops out at 62% — nowhere near the MVP model's 90%. The misses are a catalog of what no box score sees. Caleb Williams 2022 lands 10th — a brilliant individual season carrying a merely-good USC team; he won on highlight reels and brand. Lamar Jackson 2016 stays low — a historic rushing season the passing metric under-credits. The NFL MVP is a remarkably predictable, team-success-driven award. The Heisman is not: more of it lives in narrative, "Heisman moments," and preseason hype than any model can or should pretend to quantify.

What actually moves the needle

When we first built a cruder version of this, the headline read "the Heisman is a power-conference award." The richer vote-share model tells a more precise story. Once you hand it a real team-quality signal — where a team finished in the playoff picture — the raw "power conference" flag stops mattering almost entirely; it was only ever a stand-in for "is this a good team," and the CFP proxy does that job directly. What the model leans on, in order, is how good your team was (the playoff/Elo proxy), how much you produced (yards and touchdowns), and how efficiently you played (the skill rating). The Heisman is a team-success-and-production award layered on quarterback skill — with a large, irreducible slice left over for story.

So who wins it — and who's just the best?

This is the useful split, and the Heisman factors toggle on the board above flips between the two questions for any season:

• Who's the best quarterback? The skill rating. In 2025 it's Diego Pavia (Vanderbilt) — elite opponent-adjusted efficiency with real rushing value, on a team that never gets prime-time billing.
• Who wins the trophy? The Heisman model. In 2025 it crowns Fernando Mendoza (Indiana, 16–0) at #1 — and he won — lifting him from 24th on pure production, while correctly fading Pavia, because Vanderbilt isn't a playoff brand.

That's the whole game. The skill rating finds the quarterback the voters are sleeping on. The Heisman model tells you who they'll actually crown. They're different questions, and the distance between them — visible in the toggle's movement arrows — is the part of the Heisman that was never about quarterback play.

What it's for — and what it isn't

Use the skill rating to find the best quarterback in the country and the one the Heisman conversation is overlooking. Use the Heisman model to anticipate the trophy — but hold it loosely: at top-three 62% it's a strong filter, not an oracle, because the award keeps a big chunk of itself in narrative. Don't read the Heisman model as a quality ranking — it's deliberately contaminated with team context, which is the entire point. And the whole thing rests on one load-bearing wall: the opponent-adjusted strength-of-schedule, now rebuilt to cover every FBS team — the prior version quietly dropped realignment-stranded programs like Oregon State and Washington State, which is fixed.

We set out to find the best quarterbacks in college football, and we did. Then the Heisman taught us that "best quarterback" and "Heisman winner" are two different questions — and that the gap between them is team success, production, and a stubborn, unmodelable layer of story. We didn't want an opinion about the trophy. We let the data have one. It turns out the most honest thing a model can say about the Heisman is that it gets you close — and that the last stretch belongs to the voters.

Explore the full leaderboard, the Heisman backtest, and the 2025 Index at the PolyEdge College QB Rating page. Companion to the NFL PolyEdge QB Rating.