How it works
Model v3.0 · built on 2019–2024, tested on 2025
01 · The line
Yards only, one game, in the stat each position is bet on.
275
passing yds
happens 30% of starts
100
rushing yds
happens 18% of starts
100
receiving yds
happens 11% of starts
70
receiving yds
happens 11% of starts
02 · The data
Refreshed before every board.
Play-by-play
Every snap: down, distance, air yards, pressure, pass rate
Box scores
Every player's yards, targets, carries, game by game
Snap counts
Who was actually on the field
Injury reports
Out, doubtful, questionable, both teams
Depth charts
This week's starters and who fills in
Vegas lines
Spread and over/under, every game
Weather
Wind at kickoff, forecast for this week
03 · The method
50 ideas, 20,000+ games.
1
Build every idea
~50 factors for every starter, every game. Only what was known before kickoff.
2
Test on unseen games
Learn six seasons, predict the seventh. Repeat for each season.
3
Keep what helps
A factor stays only if it helps in most seasons.
4
Each position alone
A 275-yard QB isn't a 100-yard RB. Separate searches.
5
Project the yards
A projection and a likely range, so the chance works for any line, not just the boom.
6
Keep up with the league
Recent seasons count more, and projections adjust week to week to how the league is running.
Checked against a black box allowed all 50: ours matched or beat it at every position.
04 · What predicts a boom
How often the top fifth on each factor boomed, vs the bottom fifth. Bigger gap, more points.
QB275+ passing ydsnormal 30%
Yards lately17 pts
top 44%bottom 18%Game total6 pts
top 44%bottom 19%Team points24 pts
top 45%bottom 16%Opp pass defense13 pts
top 36%bottom 24%Pass-happy team15 pts
top 40%bottom 21%Wind13 pts
top 31%bottom 24%Boom-or-bust0 pts
top 32%bottom 28%Pocket passer12 pts
top 33%bottom 26%
RB100+ rushing ydsnormal 18%
Yards lately30 pts
top 32%bottom 7%Opp run defense24 pts
top 24%bottom 12%Carry share23 pts
top 27%bottom 8%Run-heavy team16 pts
top 24%bottom 13%Offense quality7 pts
top 19%bottom 14%
WR100+ receiving ydsnormal 11%
Yards lately43 pts
top 24%bottom 4%Target quality34 pts
top 23%bottom 4%Game total15 pts
top 15%bottom 8%Injured teammates8 pts
top 12%bottom 11%
TE70+ receiving ydsnormal 11%
Yards lately29 pts
top 29%bottom 2%Target quality35 pts
top 27%bottom 2%Snap share20 pts
top 22%bottom 3%Team points16 pts
top 18%bottom 7%
05 · What didn't make it
Some of these really move yardage. The betting line already prices them, so they add nothing new.
QBs in a dome
24% vs 17%
…but domes already get higher totals
QBs in the cold
12% vs 20%
…but cold games already get lower totals
WRs with a backup QB
6% vs 8%
…but books drop the team total
Home vs roadNo consistent edge once Vegas is inNight gamesNo consistent edgeTurf vs grassDomes and turf overlap; Vegas already prices itDomeReal, but the game total already reflects itCold weatherReal for QBs, but the game total already reflects itBackup QB startingReal, but Vegas lowers the total for itExtra restNo consistent edgeDivision gamesNo consistent edgeA player's history vs this opponentNo signal in 5,133 games; his form vs everyone predicts 9× betterRevenge games (vs a former team)No edge; slightly worse, if anything (369 games)Opposing DBs or front seven outNo consistent edgeOwn offensive line outNo consistent edgePass-rush pressure, pace, explosive plays allowedAlready covered by the factors above
Injured teammates: real, but small
A top teammate ruled out, vs the player's normal:
−6
context
+5
context
+6
scored
+4
context
The one weather factor that stayed: wind, for QBs.