The numbers, explained

How many points a play added to the drive's expected points, from down, distance, field position and clock.

Every snap starts with an expected-points value: what an average offense scores from that down, distance, yard line and game clock. EPA is the value after the play minus the value before it. A 9-yard gain on 3rd-and-8 is worth far more than the same 9 yards on 3rd-and-15, and EPA captures that where raw yards cannot.

Share of plays with positive EPA: the play left the offense better off than it started.

Success rate asks a yes-or-no question of every play: did it raise the offense's expected points? It rewards staying on schedule and is less swayed by one 70-yard bomb than EPA per play.

How often a team throws compared with an average team in the same situations.

Everyone passes on 3rd-and-12 and runs while protecting a fourth-quarter lead. PROE removes the situation: the nflverse model estimates the pass probability of each snap from down, distance, field position, clock and score, and PROE is how far the team's real choices sit above or below that.

A pass that gains 20+ yards or a run that gains 10+.

Big gains flip field position and drives in a single snap. The thresholds are the common analytics convention, and they count the same for every team.

Average EPA on every play where the quarterback dropped back: throws, sacks and scrambles.

The cleanest single number for a quarterback's season. A dropback is any designed pass play, whether it ends in a throw, a sack or a scramble, so the quarterback is charged for the sacks he takes and credited for the yards he makes with his legs.

Completion rate above or below what the throws' difficulty predicts, in percentage points.

The nflverse completion model estimates how often an average passer completes each throw from its depth, location, down and distance. CPOE is the actual result minus that estimate. +3.0 means he completes three more passes per hundred than the degree of difficulty suggests.

Average air yards per pass attempt: how far downfield the ball travels before the catch point.

aDOT measures intent, not result. A receiver with a 14-yard aDOT is used as a vertical threat; a 5-yard aDOT is a short-area, yards-after-catch role. For a quarterback it says how aggressive the passing game is.

Share of his team's offensive or defensive snaps a player was on the field for.

The truest measure of role. A receiver who plays 90% of snaps is an every-down starter whatever the depth chart says; one at 35% is a rotational piece.

Share of the team's targeted passes thrown to this player.

Targets are opportunity. A 25% target share means one in four of the team's aimed throws went his way, the clearest sign of a featured receiver.

Share of the team's air yards on targets that went to this player.

Target share counts throws; air yards share weights them by depth. A deep threat can own 35% of the air yards on 18% of the targets.

Yards gained after the ball is caught, averaged per reception.

YAC splits a receiver's production between where the ball arrived and what he did with it. High YAC with a low aDOT is the classic screen-and-slant role.

The live chance each team wins, updated after every play.

Win probability turns score, time, field position, down and distance into the chance a team goes on to win. The swing on a single play shows how much that play mattered.

Share of 10,000 simulated seasons in which the team reaches the playoffs.

We play out the rest of the schedule 10,000 times, seed both conferences with the current format (four division winners and three wild cards, only the No. 1 seed gets a bye) and run the bracket to the Super Bowl.

Average strength of the opponents left on the schedule, in points above an average team.

Records at midseason flatter some teams and punish others. Strength of schedule rates every opponent by how it has actually played this season and averages the games still to come, and the games already played.

How a team played in games with a player on the field against games without him.

Football has no lineup stints, so the unit is the whole game. The split shows what a team's offense or defense looked like when a starter played and when he sat.

How many plays or games stand behind a number. Small samples swing wildly.

Efficiency stats settle down only as plays pile up. Three good games can put a backup at the top of any per-play table. We show the plays and games behind our numbers and set minimums before ranking anyone.

The average of a player's percentiles on his position's card, from 1 to 99.

The number on the trading card. It summarises five or six position-specific measures into one figure against qualified peers. Tiers: Elite 90+, Pro Bowl 75+, Starter 50+, Depth below 50.

Where a player ranks among qualified players at his position: 90 = better than 90% of them.

A percentile puts any number on the same 1-99 scale, so yards per carry and sack rate can sit on one radar.