How grading works

Fantasy points reward volume — a running back who gets 25 carries for 80 yards on a bad offense scores higher than one who was far more efficient on fewer touches. They also don't care whether those yards came against a good defense, in the fourth quarter of a blowout, or in a tied game with the season on the line.

PFF-style grades try to fix that with manual, subjective grading: a human watches every play and assigns a score. That's more context-aware, but it's slow and expensive to produce, and opaque — nobody outside PFF can see why a player got the grade they got.

This site's grades are built entirely from public, play-level data using formulas anyone can inspect. Same inputs always produce the same grade — no film room, no house grader, no black box.

We tested the grading formula on the 2021–2025 seasons, which it was never tuned on. Each player was graded on his games up to week 8 only, exactly as this site would have shown him then, and compared with what he did over the rest of that season: 692 player-seasons in all.

Rank correlation: 1.00 means the order matched exactly, 0 means no relationship. The best in each group is in white. “How well he played” is his grade over the rest of the season, so that half is judged on our own yardstick; “raw EPA” is expected points added per play before any of our adjustments.

A grade is not a fantasy projection. Points per game so far predicted the rest of a season's fantasy points far better at every position, and adding the grade on top of them improved that prediction by less than 1%. Fantasy points follow volume — targets and carries — which holds steady from week to week. A grade measures what a player did with his chances.

Where it earns its keep is quality, most of all at running back. A back's points so far said almost nothing about how well he would run the rest of the season (0.04), while his grade did (0.30). It isn't the best tool everywhere: for quarterbacks, plain EPA per play did slightly better than our grade, and for receivers, points so far did.

Running back and receiver grades carry a lot of noise. Grade a player on the odd-numbered weeks and again on the even ones, and the two agree only 0.14 for backs and 0.13 for receivers, against 0.66 and 0.50 for their points per game. Quarterback grades are steadier (0.41). We tried pulling small samples toward the average to calm them down; it made no difference on seasons it wasn't tuned on, so we didn't ship it.

So use a grade to judge how well someone is playing, and use our fantasy projections to decide who to start. They are built for that, and they are tested the same way.

Raw EPA doesn't know if a defense is elite or terrible. Before grading, we calculate each defense's season-to-date EPA allowed per play (pass and rush, separately) versus the league average. A player's EPA on a given play is then adjusted up if it came against a stingy defense, or down if it came against a defense that lets everyone move the ball. Beating a top-5 defense counts for more than beating a bottom-5 one.

A defense is judged only on its games against other offenses, so a team's own performance never counts toward the strength of the defense it faced. Early in the season, when a defense has played only a game or two, its record is pulled most of the way back toward average, so the adjustment starts small and grows as the evidence builds up.

Opponent-adjusted EPA/play, success rate, CPOE (for QBs), and WPA are combined into a single composite score, weighted differently by position — for example, a quarterback's grade leans harder on opponent-adjusted EPA and accuracy (CPOE), while a receiver's leans on efficiency plus target share.

That composite score is then converted to a percentile rank against the players at the same position with a real workload that week or season. A grade of 85 means a player graded better than 85% of those peers over that sample — not an absolute score against some fixed ideal. The best qualified player at a position sits just under 100, and no two players with different performances share a grade. A real workload means, per game his team has played, at least 20 dropbacks for a quarterback, 6 carries and targets for a running back, or 4 targets for a receiver or tight end; a defender needs to have played in half his team's games. Players below that still get a grade, but they aren't ranked, and they aren't part of the crowd anyone else is measured against, so a few plays can't skew it.

While a game is in progress, the game page and player pages show a live rating from 1 to 99 next to the season grade. It's a different, simpler calculation, because the EPA and win-probability data behind the season grades isn't available until after the game.

The live rating is built from the running box score. Each player earns points above or below an average performance for their workload, and the total is squeezed onto a 1-99 scale where 50 means an average showing so far. Here's what counts:

Early in a game most players sit near 50 because they haven't done much yet. The rating moves as production adds up, and a turnover or a big play can swing it quickly.

The live rating is a snapshot of the game as it unfolds, not a replacement for the season grade. Once the play-by-play data is published, usually within a day, the EPA-based game grade described above appears alongside it and the season grade updates.