Model line
–
NFL spread and over/under projections from team scoring form, starting QB rating, efficiency, pace, return units, weather and starters ruled out. Trained on 2015 to 2025 games.
Enter the line the way a sportsbook shows it for the home team: -3 means home favored by 3.
Enter the kickoff forecast. Wind only starts to matter above 15 mph; cold matters below 45°F; rain or snow takes about 2.4 points off the total.
QB rating is EPA per dropback, weighted toward his last 12 starts. A typical NFL starter is about +0.07; a typical backup about −0.06.
Ranks are out of 32, from recent form weighted toward the last 8 games. Lower is better for defense stats and for seconds per snap. Punting and kicking rows are shown for reference; the model found they do not move scores. "2026 against Vegas" rows describe the season so far only. Game to game volatility combines swings in points, EPA per play and yards; ±x/±y is the typical swing in points scored and allowed. Neither volatility nor early results against the line have carried over in past seasons; since 2015, early-season results against the line have not carried over to the rest of the season.
For when one team falls behind by 17 or more in the second half. Uses the teams, quarterbacks and Vegas line above.
"Gets within one score" means cutting the deficit to 8 or fewer at any point, which is when live spreads and second half bets on the trailing team cash. "Lets teams back" is how often an average team down 17 gets within one score against this team. From 1,239 second half deficits of 17+ points, 2015 to 2026.
Enter where the game stands right now. Uses the two teams and the pregame Vegas line and total from the matchup above; type in any live lines you see to check them.
Both teams are guaranteed a possession (regular season since 2025, playoffs since 2022), then it is sudden death. Regular season overtime is 10 minutes and can end in a tie; playoff overtime is 15 minute periods until someone wins.
Points for a one standard deviation gap between teams (about the gap between an average team and a top 6 to 8 team), holding everything else equal. Factors are correlated, so QB rating and passing efficiency share credit.
Turnover margin, red zone TD rate, rushing efficiency, sack rates, bye weeks and short weeks, punter pinning inside the 10, net punting and kicker accuracy. Each made the projections slightly worse or no better on unseen seasons, usually because they swing week to week more than they repeat.
Each season was predicted by a model trained only on earlier seasons. "Before" is the PPG, weather, injury and returns model; "after" adds QB rating, success rate, passing efficiency, explosive plays, pace and penalties. Records count bets where the model and Vegas differed by at least the edge shown; pushes excluded. The added factors were picked by testing on these same seasons, so treat the "after" records as a best case.