Every term on this site, in plain English · predictions are published to show the model works. nothing here is betting advice.
Words we use
If a word on the site isn't here, tell us and we'll add it. See also the FAQ and How this works.
Win probability
The model's estimate of how likely a team is to win a game, as a percentage. 65% means that, over many games where we say 65%, that team should win about two out of three. It is not a guarantee about any single game.
Pick
The team with the higher win probability. Every pick is stored before kickoff with the probability and the reasons, and graded after the final. A pick is never edited after the fact.
Straight-up
Picking the winner of the game, with no point spread involved. Our headline record is straight-up.
Calibration
Whether stated probabilities match reality: when the model says 70%, do those teams win about 70% of the time? A well-calibrated model can be trusted to say how sure it is; an accurate but badly calibrated one cannot.
Brier score
A single number for how good a set of probabilities was: the average squared gap between each probability and what happened (1 for a win, 0 for a loss). Lower is better. Always saying 50% scores 0.25; a good NFL model is near 0.21.
Elo
A team rating that goes up when a team wins and down when it loses, moving more for surprising results and for larger margins. Ratings are pulled part of the way back toward average each offseason. Elo is one of the model's inputs, not the model itself.
EPA (expected points added)
A measure of how much a single play changed a team's expected points, based on down, distance and field position. Averaged per play and adjusted for opponent quality, it captures how efficiently a team actually moves the ball and stops the other side, which predicts future results better than points or yards.
Opponent-adjusted
Numbers corrected for who a team played. Gaining 400 yards against a weak defense is worth less than 400 against a strong one; the adjustment estimates each team's strength and each opponent's at the same time.
Strength
A team's expected margin against an average team on a neutral field, in points. The number behind the power rankings.
Named starter (QB)
The quarterback the schedule data lists as starting the next game. The model rates that specific player, so a backup starting changes the number; it also flags when the starter has changed.
Starters out
A position-weighted count of listed starters ruled Out or Doubtful on the official injury report. Quarterbacks are handled separately through the named starter.
Rest / short week
Days since a team's last game. Short weeks (Thursday games) and byes are small inputs to the model.
Home field
The average advantage of playing at home, a little over two points in the modern NFL. Set to zero for neutral-site games.
Projected score
The most likely final score, built from the model's projected margin and projected total. Real scores vary far more than projections do: the projection is the middle of a wide range, not a prediction of exact numbers.
Spread (point spread)
The market's expected margin. 'PIT −6' means Pittsburgh is expected to win by six; a bet on Pittsburgh at that line wins only if they win by seven or more. We show the closing spread for comparison and never use it as a model input.
Against the spread (ATS)
Whether a team beat the spread rather than just won. If PIT −6 wins by three, PIT won straight-up but lost against the spread. Our ATS record compares our projected margin with the closing spread.
Total (over/under)
The market's expected combined score. Our projected total is compared with it and graded like a pick: over if ours is higher, under if lower.
Push
A result that lands exactly on the line (a team favored by 3 wins by 3). Pushes are excluded from records, not counted as misses.
Moneyline
Odds on which team wins, no spread. −200 means bet $200 to win $100 (a favorite); +170 means bet $100 to win $170 (an underdog). Moneylines can be converted to implied probabilities, which is how we compare the market's view with ours.
Opening line / closing line
The first price a market posts for a game and the last price before kickoff. The closing line is the best public forecast there is, because it has absorbed a week of information and money. We snapshot both and show how our number moved between them.
Closing-line favorite
The team the market favored at kickoff. Our benchmark: any model has to be judged against it, and over sixteen seasons it wins about 66.5% of games.
Vig (juice)
The bookmaker's margin, built into the odds so the implied probabilities add up to more than 100%. We remove it before comparing market probabilities with ours.
Analyst layer
A capped, logged adjustment on top of the statistics model that reads news reports (a quarterback change, a key player ruled out, resting starters) and moves the probability by a few points at most, only for things the model has not already counted. Every claim shows its source, and each is graded afterwards.
Claim
One piece of news the analyst layer turned into a structured statement: type, team, player, direction, confidence and the points it is worth. Declined claims are kept and graded at zero weight, so we learn whether the filters were right.
Walk-forward backtest
Replaying past seasons week by week, training only on games that had already been played at that point, so the model never sees the future. The only honest way to estimate how a prediction system would have done.
Leak audit
An automated check that no feature uses information from after kickoff (final scores, later injury reports, closing lines). Run before every backtest.
Polymarket
A prediction market where people trade on the outcome of events, including NFL games. Its prices are another public probability we compare against.
Season projection
Thousands of simulated seasons played out from the current ratings and remaining schedule, giving each team's expected wins, playoff odds and championship odds.