Every number, explained

Expected Points Added: how much a play moved the team’s expected score, given down, distance and field position before and after.

How it’s computed here

nflfastR’s expected-points model per play, aggregated from play-by-play over scrimmage plays only — kneels, spikes and two-point tries excluded. League average sits near +0.015 per offensive play.

How to read it

The best offenses live above +0.10; the worst below −0.10. Defensive EPA is points *allowed* per play, so lower is better there.

Caveat

Not adjusted for opponent on stat sheets. The model’s Net Rating is the opponent-adjusted version.

Appears in: Stats · Team pages · Explore · Matchup edges

Share of plays that were successful — kept the offense on schedule.

How it’s computed here

nflfastR’s indicator: a play is a success when its EPA is positive. League average ≈ 44%.

How to read it

A consistency measure. High success with low explosive rate = methodical; the reverse = boom-or-bust offense.

Appears in: Stats · Team pages · Explore

Share of plays gaining chunk yardage.

How it’s computed here

Passes of 20+ yards and rushes of 10+ yards, over all scrimmage plays.

How to read it

Explosives correlate with scoring more than steady gains; defenses that prevent them force long drives.

Appears in: Stats · Team pages

Completion Percentage Over Expected — accuracy adjusted for throw difficulty.

How it’s computed here

Actual completion minus nflfastR’s expected completion probability per throw, averaged over attempts.

How to read it

Positive means completing throws harder than the average QB completes. One of the more stable QB skill signals.

Appears in: Stats · Passing · QB form in the model

Pass Rate Over Expected — play-calling intent versus what the situation predicts.

How it’s computed here

Actual pass/rush choice minus nflfastR’s expected pass probability per snap, averaged.

How to read it

Positive = passes more than situations suggest. Identity, not quality — pair it with EPA to see if the identity works.

Appears in: Stats · Passing · Explore

How quickly the offense snaps the ball.

How it’s computed here

Clock elapsed between consecutive snaps of the same possession, keeping only 0–60 second gaps (filters kneel-outs, quarter breaks, two-minute chaos). 2025 spans 26.6 (fastest) to 31.9 (slowest).

How to read it

The totals bettor’s tempo stat: fast pace = more plays = more scoring chances for BOTH teams. Lower number = faster.

Appears in: Stats · Offense · Explore

How often the QB was hit or sacked per dropback.

How it’s computed here

(sacks + QB hits) ÷ dropbacks. Labelled a proxy because true pressure charting (hurries) is not in the public feed.

How to read it

On defense, higher is better. On offense (allowed), it is a pass-protection read.

Caveat

Undercounts pressures that produce neither hit nor sack.

Appears in: Stats · Defense · Matchup edges

Scoring output per possession — efficiency with the game’s natural unit.

How it’s computed here

Drive outcomes valued (TD ≈ 6.95 with the average conversion, FG 3, safety −2), divided by drives.

How to read it

League-best offenses exceed 2.5; best defenses hold under 1.6.

Appears in: Stats · Matchup edges

The market’s handicap: expected margin between the teams.

How it’s computed here

Historical lines are closing spreads from the nflverse schedule, home-team convention (DEN −9.5 = Denver favoured by 9.5).

How to read it

Covering means beating your own number, not winning the game.

Appears in: Everywhere a game is shown · Edge

Against The Spread: wins-losses-pushes when results are settled against the closing line.

How it’s computed here

Team margin minus its own spread per game; positive covers, zero pushes. League-wide ATS wins always equal losses — this product asserts that invariant in its test suite.

How to read it

52.38% is break-even at standard −110 juice. Descriptive, not predictive — last year’s cover rate does not persist.

Appears in: Stats · Against the spread · Team betting profile · Game rivalry block

Margins where NFL games disproportionately land, because scoring comes in 3s and 7s.

How it’s computed here

Fitted from 6,967 games since 1999: a margin of exactly 3 occurs 15.0% of the time — 9× more than margin 9. The model re-weights a smooth curve by these factors (w(3)=2.79, w(7)=1.85, w(9)=0.50).

How to read it

Half a point around 3 or 7 is worth several times more than around 5 or 9.

Appears in: Game pages · the lattice chart · Edge · spread board

Chance the game lands exactly on the spread, refunding spread bets.

How it’s computed here

The model’s probability mass on the whole-number line. At a spread of exactly 3, observed pushes run ≈8.7%; this model prices 7.3%; a smooth normal says 2.7%.

How to read it

Only whole-number lines can push. High push mass is the argument for buying the half point.

Appears in: Game pages · Edge

How much outcome probability flips if the line moves half a point against you.

How it’s computed here

On a whole number: the push mass that becomes a loss. On a half number: the win mass that becomes a push. NOT the change in cover probability — that is zero on whole numbers, which is precisely where the half point matters most (a bug this product shipped and documented).

How to read it

Highest on key numbers. 7%+ at 3; near 2% on numbers like 5.5.

Appears in: Edge · spread board · Game pages

The bookmaker’s margin baked into prices.

How it’s computed here

Sum of implied probabilities minus 100%; ≈ 4.5% on a standard −110/−110 pair. Removed multiplicatively before any model-vs-market comparison.

How to read it

The tax you must beat. It is why 50% ATS loses money.

Appears in: Edge · Methodology

The market’s expected combined score.

How it’s computed here

Closing totals from the schedule feed; team O/U records settle actual combined points against them.

How to read it

Pace and points-per-drive are the underlying drivers; weather suppresses.

Appears in: Game cards · Stats · betting splits

This product’s expected final margin, home perspective.

How it’s computed here

Opponent-adjusted net EPA difference × 15.1 points, + era-fitted home field (1.71 pts for the 2020s, down from 2.56 in the 2000s), + QB adjustment, + rest. Every constant fitted walk-forward — no game predicted with data from its own future.

How to read it

Always shown with an (est.) marker. The component breakdown on game pages shows where the number comes from.

Caveat

The closing line beats this model (10.26 vs 11.06 MAE). Where they disagree, the market is usually right — stated on the Edge page, measured over 5,067 games at 49.93% ATS.

Appears in: Game cards · Game pages · Edge

A team’s opponent-adjusted strength in EPA per play.

How it’s computed here

Offense and defense solved jointly by iterative adjustment (who you played matters), recency-weighted with a 6-week half-life, regressed toward a 45% carry-over of last season by sample size.

How to read it

+0.10 is elite; the rating × 15.1 is roughly the point-spread value of the team.

Caveat

Preseason ratings are pure carry-over priors and are labelled as such.

Appears in: Team pages · Stats · Overview · Model margin

Line movement for who is actually starting at quarterback.

How it’s computed here

Starter-vs-established EPA/dropback gap × 32 points, after each QB is regressed to league average by dropback volume (small samples span −0.99 to +0.74 EPA/db — four times the spread of real signal). Clamped at ±10.

How to read it

The single largest routine shock to a line; a star-to-backup swap is worth about a touchdown.

Caveat

Not covered by the backtest — testing it honestly would require week-by-week QB form the archive does not carry. Stated on the methodology page.

Appears in: Game pages · component breakdown · QB watch

Chance of winning at any moment of a game.

How it’s computed here

nflfastR’s in-game model per play (score, time, field position, timeouts), drawn as the game-flow chart.

How to read it

The area between the line and 50% is dominance; sharp cliffs are the season-turning plays.

Appears in: Game pages

Probability of making the field, winning the division, or earning each seed.

How it’s computed here

10,000 simulations of every remaining game — sampling full scores, not just winners, because the tiebreakers need points — with the complete cascade applied in every simulated season.

How to read it

A team “at 62%” misses in 4 of 10 simulated futures. Preseason odds are priors and move fast.

Appears in: Standings · Team pages · Division pages

The three scoring formats: 1, 0.5 or 0 points per reception on top of yardage and touchdowns.

How it’s computed here

Points come from nflverse directly, never re-derived — so they cannot drift from the reference. Half-PPR is exact: PPR minus standard equals receptions.

How to read it

Format changes player order materially; the board re-ranks per format.

Appears in: Fantasy board · Player pages · Compare

Value Over Replacement — points above the player you could have started instead.

How it’s computed here

Per-game points minus the last starter’s at the position in a 12-team league (QB12 / RB24 / WR36 / TE12 / K12), times games played.

How to read it

Why a top TE outranks a better-scoring WR: scarcity at the position is the value.

Appears in: Fantasy board

Groups of players separated by real production gaps, not adjacent ranks.

How it’s computed here

A new tier opens where the per-game drop to the next player exceeds 2.2× the average drop (and at least 0.6 points).

How to read it

Draft by tier boundary: missing #7 costs little if #8 shares the tier, a lot if he does not.

Appears in: Fantasy board

Share of team targets; WOPR blends target share and air-yards share into one opportunity number.

How it’s computed here

From nflverse weekly data. WOPR = 1.5 × target share + 0.7 × air-yards share.

How to read it

Usage is far stickier week to week than production — the most reliable forward signal in fantasy. Rising target share over the last 3 games flags the trend arrows.

Appears in: Fantasy board · Player pages · Compare

Share of games above a startable line, at ceiling, or at dud level.

How it’s computed here

Position-specific PPR lines, stated so you can disagree: startable QB 18 / RB 12 / WR 12 / TE 9; boom QB 25 / RB‑WR 20 / TE 16; bust QB 12 / RB‑WR 6 / TE 4. Only games with real involvement count — a healthy scratch is not a bust.

How to read it

Two players with equal averages can be a weekly anchor and a coin flip. This is the difference.

Appears in: Fantasy board · Player pages

Fantasy points a defense allows to each position, per game.

How it’s computed here

Every player-week grouped by the defense that faced it; rank 1 = stingiest of 32.

How to read it

The matchup grid behind props and start/sit calls.

Caveat

Schedule-dependent — a defense that faced three elite TEs looks leaky to TEs.

Appears in: Stats · props section

Playoff order: four division winners then three wild cards per conference (since 2020).

How it’s computed here

The league’s full cascade as written: head-to-head, division record, common games (min. 4), conference record, strength of victory, strength of schedule — with wild cards selected iteratively and the sweep rule for 3+ teams. Validated against ground truth: reproduces the actual playoff field in 27 of 27 seasons since 1999.

How to read it

The small grey note under a team names the exact rule that decided its position.

Appears in: Standings · Division pages

Combined record of every team you beat.

How it’s computed here

Win percentage across beaten opponents’ full seasons; a late-cascade tiebreaker.

How to read it

Why beating good teams matters twice.

Appears in: Tiebreaker notes

This number is modelled, not observed.

How it’s computed here

Attached to every derived figure: model margins, ratings, playoff odds, win probability.

How to read it

An unmarked number happened; an (est.) number is an estimate with error bars in spirit.

Appears in: Everywhere

Where a value ranks — and the league’s shape around it.

How it’s computed here

Ranked against the teams or players that actually played that season (31 teams before 2002, never a hardcoded 32). The histogram strips show whether a 78th percentile sits in a pack or on an island.

How to read it

No games yet = no percentile shown, never a fake “0th”.

Appears in: Team pages · Player pages

Before kickoff, player pages and the board show last season’s production under a visible banner.

How it’s computed here

Identity (team, headshot) from this season’s roster; production from last season; ratings carry 45% of last season as a labelled prior.

How to read it

Carry forward what is known, label it, never present absence as zero.

Appears in: Fantasy board · Player pages · Preseason standings

The gap between this model’s margin and the closing line — shown as analysis, never as a tip.

How it’s computed here

Backtested directly: sides the model preferred covered 49.93% over 5,067 games; 52.38% is break-even. An earlier version flagged “value bets” from this gap; it fired on 91.7% of selections and lost money, so it was removed.

How to read it

A model disagreeing with a sharper market is usually evidence the model is wrong. This product says so on its own betting page.

Appears in: Edge · Game cards · Methodology