Player rankings

Where the numbers come from. Free play-by-play data (nflverse) credits three people on every play — the passer, the rusher and the receiver — with the expected points the play added. We total that per player per game, then weight each game toward the present: this season counts fully (with a slow decay across the year), last season counts 30%, two seasons back 10%. New players and players who changed teams are handled the same way as everyone else.

Adjusted for opponent. A carry against a defense that allows a lot is worth less than the same carry against a stingy one. We subtract how far each defense faced sits above or below the league average, weighted the same way, so the "adjusted" column is what a player did against average defenses.

Shrunk, then valued. Adjusted EPA per touch is pulled toward the position average with a prior of 80 touches for backs and 60 for receivers and tight ends, so a dozen great plays do not outrank three hundred good ones. Value multiplies that by weighted touches, above a replacement level set at the 25th percentile of qualifiers, which is why a heavily used, above-average back can rank above a part-timer with a better per-touch number.

What we do not rank, on purpose. Offensive linemen, defenders, kickers and returners get no per-play credit in public data, so we do not rank them; inventing a number would be dishonest. Blocking by tight ends and backs is invisible here too. Positions come from the latest depth charts; a player missing from them is placed by usage (more carries than targets: back; otherwise receiver).

These are not used in the game predictions. The game model rates teams and the named starting quarterback. Whether skill-position rankings would improve the predictions is a study on our list; if they are ever added, the change will be logged and graded like everything else. Nothing here is a tip.