Reference
Glossary
Every metric on this site, defined in one sentence, with what a good value looks like and where the number came from. If a figure appears on a chart and not on this page, that is a bug.
Opportunity and usage
Volume, and whose volume it is. Most fantasy scoring is decided here before efficiency gets a say.
- End-zone targets
Targets thrown into the end zone itself, regardless of where the play started.
What good looks likeThe most direct evidence of touchdown intent. Unstable year to year, so it explains a past season better than it predicts the next.
Sourcenflverse play-by-playAppears onPlayer pages
- High-value touches
Targets plus carries inside the opponent 10, the touches worth materially more than a carry in open field.
What good looks likeAmong backs with 150 or more carries, the median season is 76 and the top quarter start at 93; only eleven of sixty-eight reach 100. A target is worth roughly twice a carry in PPR and a goal-line carry several times one, so this survives week-to-week noise raw carry counts do not.
Sourcenflverse play-by-playAppears onPlayer pages, running back charts
- Inside-5 carries
Carries taken from inside the opponent 5-yard line.
What good looks likeThe narrowest measure of goal-line trust. Double figures in a season is a back his coaches want scoring.
Sourcenflverse play-by-playAppears onPlayer pages
- Red-zone targets
Targets inside the opponent 20-yard line.
What good looks likeTwenty or more in a season is a genuine scoring role rather than a compiler.
Sourcenflverse play-by-playAppears onPlayer pages
- Route participation
The share of his team's pass plays on which a player ran a route, rather than blocking or sitting.
What good looks likeAbove 80% is a full-time receiving role; the median of the twenty-five tight ends charted is 78%. A licensed 4for4 column, published here by the operator's decision. See the tight end page.
Source4for4, via the workbookAppears onTight end and running back charts
- Targets per route run
Targets divided by routes run: how often a player is thrown to when he is actually out in a pattern.
What good looks likeAbove 0.23 is top-quartile among the twenty-five tight ends charted, whose median is 0.21. It separates a player who is ignored from one who is simply not on the field.
Source4for4, via the workbookAppears onTight end and running back charts
- Touches
Carries plus receptions: everything a player actually did with the ball in his hands.
What good looks likeVolume, not value. High-value touches is the version that accounts for where those touches happened.
Sourcenflverse play-by-playAppears onPlayer pages
- WOPR
Weighted opportunity rating, combining target share and air yards share into one measure of receiving opportunity.
What good looks likeShown here as its two components as well as the composite, because the same WOPR can describe a possession role and a vertical one.
Sourcenflverse play-by-playAppears onPlayer pages
- Yards per route run
Receiving yards divided by routes run, the efficiency counterpart to targets per route run.
What good looks likeAbove 2.0 is elite for a tight end. The charted median is 1.74 and only a handful clear two. Rewards a player who does more with fewer looks.
Source4for4, via the workbookAppears onTight end charts
Route running and receiving
What a receiver was asked to do, and what he did with it.
- Average cushion
How far off the defensive back lines up before the snap.
What good looks likeA small cushion is respect. Read with separation: beaten cushion and no separation is a player winning early and losing late.
SourceNext Gen StatsAppears onPlayer pages
- Average depth of target
The average distance beyond the line of scrimmage at which a player is targeted.
What good looks likeNot good or bad on its own; it describes a role. Under 7 is a possession receiver; over 13 is a field-stretcher whose weeks will be volatile.
Sourcenflverse play-by-playAppears onPlayer pages
- Average separation
The distance in yards from the nearest defender when the ball arrives.
What good looks likeAround 3 yards is average. Under 2.5 with a high target share means a player is being thrown to while covered, which is a compliment and a risk.
SourceNext Gen StatsAppears onPlayer pages
- Catch rate
Receptions divided by targets.
What good looks likeRead against depth of target. 75% on shallow crossers and 55% on a vertical role can describe equally good players.
Sourcenflverse play-by-playAppears onPlayer pages
- RACR
Receiver air conversion ratio: receiving yards divided by intended air yards.
What good looks likeAround 1.0 is typical. Above it means a player is adding yards after the catch or converting deep looks at an unusual rate.
Sourcenflverse play-by-playAppears onPlayer pages
- Receiving EPA
Total expected points added on plays where a player was targeted.
What good looks likeCounts volume and quality together, so it rewards a player who is both used and effective.
Sourcenflverse play-by-playAppears onPlayer pages
- YAC above expectation
Yards after catch against what a tracking model expected given the defenders' positions at the catch point.
What good looks likePositive means he beats the coverage he actually faced. This is the version that is about the player rather than the scheme.
SourceNext Gen StatsAppears onPlayer pages
- Yards after catch
Yards gained after the ball arrives.
What good looks likeThe part of a receiver's production least dependent on his quarterback.
Sourcenflverse play-by-playAppears onPlayer pages
Rushing
Separating the runner from the blocking in front of him.
- Eight-plus defenders faced
The share of a back's carries run against eight or more defenders in the box.
What good looks likeA high rate is a defense daring the offense to throw. It suppresses yards per carry and is not the back's fault.
SourceNext Gen StatsAppears onPlayer pages
- Rush yards over expected per carry
Yards gained per carry against what a tracking model expected, given the defenders' positions at the handoff.
What good looks likePositive is a back beating his blocking. This is the cleanest separation of runner from scheme the public data offers.
SourceNext Gen StatsAppears onPlayer pages
- Rushing efficiency
Distance actually traveled divided by distance gained downfield, or how much lateral running a back does to get where he gets.
What good looks likeLower is more direct. High efficiency numbers describe a back who bounces runs, which produces both the long touchdown and the two-yard loss.
SourceNext Gen StatsAppears onPlayer pages
- Time to line of scrimmage
How long, in seconds, a back takes to cross the line after the handoff.
What good looks likeFaster is more decisive. Read with efficiency: slow and inefficient is a back waiting for blocks that are not coming.
SourceNext Gen StatsAppears onPlayer pages
- Yards per carry
Rushing yards divided by carries.
What good looks likeThe noisiest number on a running back's page. One long run moves it a full yard, which is why the site leads with opportunity instead.
Sourcenflverse play-by-playAppears onPlayer pages
- YBC / YAC
Rushing yards gained before contact versus after it, separating the blocking environment from the runner.
What good looks likeA high YBC with a low YAC points at the offensive line; the reverse points at the back.
SourceThe workbookAppears onRunning back charts
Quarterback
The passing numbers that describe a good quarterback, and the rushing ones that describe a good fantasy quarterback.
- Aggressiveness
The share of throws made into tight coverage, with a defender within a yard of the receiver.
What good looks likeA style measure. High aggressiveness with high CPOE is a quarterback getting away with it; high without is one who is not.
SourceNext Gen StatsAppears onPlayer pages
- Air yards to the sticks
Average air yards relative to the first-down marker.
What good looks likeNegative means throwing short of the sticks by habit: fine on early downs, a problem on third.
SourceNext Gen StatsAppears onPlayer pages
- CPOE
Completion percentage over expected: actual completion rate minus the rate expected given throw depth, pressure and receiver separation.
What good looks likeIsolates accuracy from scheme. Positive is good, but it correlates with fantasy scoring at only 0.23, which is the quarterback page's point.
Sourcenflverse play-by-playAppears onQuarterback charts, player pages
- Designed carries
Rushes called for the quarterback, as opposed to scrambles.
What good looks likeThe most valuable carries a quarterback gets, because they are repeatable and often come near the goal line.
Sourcenflverse play-by-playAppears onQuarterback charts, player pages
- Dropbacks
Pass attempts plus sacks plus scrambles: every play that began as a pass.
What good looks likeThe right denominator for quarterback rate stats, because attempts alone quietly exclude the plays that went wrong.
Sourcenflverse play-by-playAppears onPlayer pages
- EPA per dropback
Expected points added per dropback, the single strongest per-play measure of quarterback play.
What good looks likeAbove 0.15 is high-end. It correlates with quarterback fantasy scoring at 0.67, the strongest football input on the quarterback page, behind only fantasy points per dropback, which is partly scoring measured against itself.
Sourcenflverse play-by-playAppears onQuarterback charts, player pages
- Fantasy points per dropback
Fantasy scoring divided by dropbacks, or efficiency in the units the game is actually scored in.
What good looks likeThe highest correlation on the quarterback page at 0.77, though it is fantasy scoring divided by volume correlated against fantasy scoring per game, so it describes efficiency rather than discovering it.
SourceDerived from nflverseAppears onQuarterback charts
- Pass rate over expected
How often a team passes against what its down, distance and score would predict, measured in neutral game situations.
What good looks likePositive is a pass-first offense. It moves quarterback volume less than intuition suggests, which the quarterback page shows directly.
Sourcenflverse play-by-playAppears onQuarterback charts, player pages
- Scrambles
Designed passes where the quarterback ran instead.
What good looks likeThe count correlates with fantasy scoring at 0.56; the rate at 0.24. Volume of scrambling is production, the tendency to scramble is only a description.
Sourcenflverse play-by-playAppears onQuarterback charts, player pages
- Success rate
The share of plays that gained positive expected points.
What good looks likeAround 45% is average. A measure of consistency rather than ceiling.
Sourcenflverse play-by-playAppears onPlayer pages, defense page
- Time to throw
Average seconds from snap to release.
What good looks likeNeither end is better. It describes an offense, and it correlates with fantasy scoring at 0.14, the weakest input on the quarterback page.
SourceNext Gen StatsAppears onPlayer pages
- Yards per attempt
Passing yards divided by attempts.
What good looks likeThe oldest efficiency number on the page and still a reasonable one, correlating with fantasy scoring at 0.50.
Sourcenflverse play-by-playAppears onPlayer pages
Efficiency and expected points
Per-play quality, measured against what an average team would have done in the same spot.
- EPA per play
Expected points added: how many points a team adds on average versus what a league-average team would be expected to score from the same down, distance and field position.
What good looks likeAbove 0.10 per play is a strong offense. The definition here is the prediction pipeline's own, so the site cannot drift from what the model computed.
Sourcenflverse play-by-playAppears onGame Prediction Model, position pages
- Expected points
The points a team in a given situation would be expected to score before the play, from which EPA is the difference afterwards.
What good looks likeThe baseline everything per-play on this site is measured against.
Sourcenflverse play-by-playAppears onRunning back charts, Game Prediction Model
- Explosive pass rate
How often the passing game produces a gain of 20 yards or more.
What good looks likeHigher is better, but it is a tail measure and swings hard on small samples.
Sourcenflverse play-by-playAppears onGame Prediction Model
- First downs per play
The share of plays producing a first down, used as a proxy for play success rate.
What good looks likeA blunter, more stable cousin of success rate.
Sourcenflverse play-by-playAppears onGame Prediction Model
- Rush EPA per carry
Expected points added on designed runs.
What good looks likeNoisier and less predictive than passing EPA, which is why the model weights it less.
Sourcenflverse play-by-playAppears onGame Prediction Model
Defense and special teams
What repeats year to year, and what does not.
- DVOA
Defense-adjusted value over average: per-play performance against a league baseline, adjusted for opponent and situation.
What good looks likeNegative is good for a defense, which is the opposite of most numbers on this site, so the defense page states the direction wherever it appears.
SourceThe workbookAppears onDefense page
- Pressure rate
How often a defense disrupts the quarterback: sacks, hits and hurries as a share of dropbacks.
What good looks likeThe stable half of defensive scoring. Pressure repeats year to year where turnovers largely do not, so it is what is worth streaming on.
SourceThe workbookAppears onDefense page
- Sack rate
How often an offense's dropbacks end in a sack.
What good looks likeLower is better for an offense. Carried by the prediction model as an offensive input rather than a defensive one.
Sourcenflverse play-by-playAppears onGame Prediction Model
- Simulated pressure
A pressure look created by threatening more rushers than actually come, dropping one or more into coverage.
What good looks likeThe only 2025 defensive block in the workbook carrying real values rather than a ranking, and it is noted as such on the defense page.
SourceThe workbookAppears onDefense page
- Turnover margin
Takeaways minus giveaways in a single game, from the perspective of the team named.
What good looks likeThe single largest driver of the model being wrong: its picks won the battle by +0.82 per game when right and lost it by 0.97 when wrong.
Sourcenflverse team-week statsAppears onCase study
- Turnover rate
Interceptions plus lost fumbles per offensive play.
What good looks likeLower is better and barely repeats. The model carries it as a rolling rate, which is why turnovers remain the largest gap between what it knows and what decides games.
Sourcenflverse play-by-playAppears onGame Prediction Model, case study
Kicking
Volume and distance, and the offense that produces both.
- Distance bands
Field goal attempts grouped by range rather than counted together.
What good looks likeBands matter more than raw attempts once the long-range rules are accounted for: a 50-yard attempt is worth more and is made less often.
SourceThe workbookAppears onKicker page
- Field goal attempts
Attempts per game, the volume half of kicker scoring.
What good looks likeKicker scoring needs an offense good enough to reach field goal range and inefficient enough to stop there.
SourceThe workbookAppears onKicker page
Market and game context
The numbers the betting market publishes, and what they imply.
- Game total
The combined points the market expects both teams to score.
What good looks likeTotals and spreads move in halves, so implied totals land on quarters and carry two decimals here rather than being rounded to a tenth.
SourceThe betting marketAppears onGame Tracker
- Implied team total
The points a team is expected to score, derived from the game total and the spread.
What good looks likeAbove 26 is a strong scoring environment; below 18 suppresses every skill position in the game.
SourceThe betting marketAppears onGame Tracker, kicker page
- Moneyline
The price on a team to win outright, convertible to an implied win probability.
What good looks likeThe model's benchmark: the market's moneyline has called 66.7% of games correctly against the model's 64.5%.
SourceThe betting marketAppears onGame Prediction Model
- Spread
The margin the market expects, quoted from the home team's perspective.
What good looks likeThe sharpest single number published about a game. On this site it is also what defines the market's side in the model's case study.
SourceThe betting marketAppears onGame Tracker, Game Prediction Model
The prediction model
How the model is scored, and what each score is worth knowing.
- AUC
The chance the model gives a randomly chosen winner a higher probability than a randomly chosen loser.
What good looks like0.5 is a coin flip and 1.0 is perfect. The model's 0.69 says it orders games well even where it cannot call them.
SourceThe prediction modelAppears onGame Prediction Model
- Brier score
The mean squared error of the probabilities, or the average of how far each forecast sat from what happened.
What good looks likeLower is better. Read next to log loss: Brier treats a confident miss more gently, so the two together describe how the errors are distributed.
SourceThe prediction modelAppears onGame Prediction Model
- Calibration error
The average gap between a stated probability and the rate that outcome actually occurred.
What good looks likeLower is better; 1.7 points is good. It is the measure that decides whether a 70% call can be read as 70%.
SourceThe prediction modelAppears onGame Prediction Model
- Confidence band
The bucket a prediction falls into (coin flip, low, medium or high), with the accuracy that band has actually achieved.
What good looks likeHigh confidence has been right 74.5% of the time; the coin-flip band hits 49.8%, which is chance, and the page says so rather than dressing it up.
SourceThe prediction modelAppears onGame Prediction Model
- Feature importance
How much each input contributes to the fitted model, relative to the rest.
What good looks likePublished in full rather than as a top few, and the page names the heaviest inputs it cannot yet show per team rather than implying the chart is the model.
SourceThe prediction modelAppears onGame Prediction Model
- Log loss
A scoring rule that punishes confident wrong answers far harder than uncertain ones.
What good looks likeLower is better. The model's 0.629 against the moneyline's 0.608 is the clearest statement that it does not beat the market.
SourceThe prediction modelAppears onGame Prediction Model
- Market-aware features
Inputs derived from the betting line, included alongside the team-performance features.
What good looks likeIncluding them makes the model better and makes it less independent. The wider feature sets tested scored worse: 202 features reached 61.2% and 455 reached 62.8%, against 63.2% for the shipped set.
SourceThe prediction modelAppears onGame Prediction Model
- Walk-forward validation
Training on every prior season and testing on the next, one season at a time, so the model never sees the games it is scored on.
What good looks likeThe only honest way to quote accuracy on a time series. Every figure on the prediction pages comes from these folds, not from data the model was fit on.
SourceThe prediction modelAppears onGame Prediction Model, case study
- Win probability
The model's estimate that a given team wins, before kickoff.
What good looks likeCalibrated, so it means what it says: across 4,166 out-of-sample games the average gap between stated confidence and actual hit rate is 1.7 points.
SourceThe prediction modelAppears onGame Prediction Model