🏈 2025 Season Recap

Can Machine Learning
Beat Vegas?

A QB-adjusted EPA model tracked across 18 weeks of regular season and all four rounds of playoffs.

65%
Regular Season Winners
52%
Regular Season Spread
83%
Playoff Winners
58%
Playoff Spread

Regular Season

Weeks 1–18
Games Predicted
272
Winners
177–95 (65.1%)
Against the Spread
139–128–5 (52.1%)
✅
Strong Through Week 12
When all starters were playing and the season had enough data behind it, the model was consistently above 50% against the spread. That's the sweet spot.
⚠️
Fell Apart Late Season
Weeks 13–18 were rough. Backup QBs, resting starters, teams tanking for draft picks. The model had no way to account for any of it and spread performance dropped to 48.6%.
🔍
What the Model Can't See
The equal 8-game window can't detect a team's winning streak, a running back quietly declining, or whether a team has already mentally checked out of the season.

Wild Card Round

January 11–13
Winners
4–2 (66.7%)
Against the Spread
2–4 (33.3%)
Biggest Edge Miss
SF upset PHI
GB @ CHI
Vegas: CHI -1.5 | Model: CHI -0.7 | Result: CHI won, didn't cover
W Winner Didn't Cover
SF @ PHI
Vegas: PHI -4.5 | Model: PHI -10.2 | Result: SF upset win
L Winner Didn't Cover
LA @ CAR
Vegas: LA -10.5 | Model: LA -13.3 | Result: LA won by 3
W Winner Didn't Cover
LAC @ NE
Vegas: NE -3.5 | Model: NE -15.3 | Result: NE dominated, covered easily
W Winner Covered
BUF @ JAX
Vegas: BUF -1.5 | Model: JAX -1.5 | Result: BUF won on road
L Winner Didn't Cover
HOU @ PIT
Vegas: HOU -3.5 | Model: HOU -3.8 | Result: HOU won and covered
W Winner Covered
📉
Too Confident on the Margins
Predicted PHI by 10 and LA by 13. Both teams won — neither covered. Playoff games have a way of tightening up that a regular season EPA model doesn't anticipate.
🔥
Hot Teams Were Real
NE crushed LAC and HOU covered at PIT continuing their 8-game win streak. Tracking team momentum turned out to be one of the better human adjustments to the model all season.
🏈
Never Bet Against Josh Allen
Model liked JAX at home. Allen went into Jacksonville and won anyway. There are a handful of QBs where the game situation just doesn't matter — he's one of them.

Divisional Round

January 17–18
Winners
4–0 (100%)
Against the Spread
4–0 (100%)
Round Grade
PERFECT 🔥
SF @ SEA
Vegas: SEA -7.5 | Model: SEA -18.0 | Result: SEA 41, SF 6 (won by 35!)
W Winner Covered
LA @ CHI
Vegas: LA -3.5 | Model: LA -0.9 | Result: LA 20, CHI 17 (CHI +3.5 covered)
W Winner Covered
BUF @ DEN
Vegas: DEN -1.5 | Model: DEN -4.3 | Result: DEN 33, BUF 30
W Winner Covered
HOU @ NE
Vegas: NE -3.0 | Model: NE -10.6 | Result: NE 28, HOU 16
W Winner Covered
🏠
Bye Week Was a Real Edge
SEA destroyed SF 41-6 after their bye — won by 35 when the model predicted 18. DEN handled BUF at home too. Two extra weeks of rest and preparation genuinely matters.
🎯
The Best Week of the Season
4-0 on winners, 4-0 against the spread. NE was predicted to win by 10.6 and won by 12. Sometimes the model just locks in — this was that week.
🏡
Home Field Hit Different in Playoffs
All four higher seeds won at home. Worth more than the 2.5 points Vegas typically assigns — closer to 3.5 or 4 once the stakes get this high.

Conference Championship

January 25
Winners
2–0 (100%)
Against the Spread
1–1 (50%)
Both SB Teams
Correctly Picked
LA @ SEA
Vegas: SEA -2.5 | Model: SEA -18.1 | Result: SEA 31, LA 27 (covered)
W Winner Covered
NE @ DEN
Vegas: NE -5.5 | Model: NE -11.0 | Result: NE 10, DEN 7 (snow game, didn't cover)
W Winner Didn't Cover
❄️
Snow Killed the Cover
NE vs DEN ended 10-7 in a blizzard with no scoring in the fourth quarter. NE won on a missed Denver field goal. No model was ever going to predict that.
🎯
Called Both Finalists
The model had NE and SEA going to the Super Bowl, and they did. For all the spread misses, identifying the two best teams in the league is what the model was built to do.
📊
Still Overshooting Margins
Predicted SEA by 18 against LA. They won by 4. The model keeps seeing blowouts that don't materialize — playoff football just doesn't cooperate with large margins.

Super Bowl LX

February 8, 2026
New England Patriots vs Seattle Seahawks
Neutral Site
SEA 29, NE 13
Seattle Seahawks are 2025 NFL Champions 🏆
Vegas Line
SEA -4.5
Model Pick
NE +4.5
Result
SEA by 16
L Winner (NE picked) Didn't Cover (NE +4.5)
🛡️
Defense Won the Trophy
Seattle's defense smothered NE all game. The 29-13 final wasn't a shootout — it was a shutdown. The model focused too much on SEA's offensive explosiveness and missed how dominant their defense had become.
📈
EPA Told Half the Story
NE had better season-long EPA (0.282 vs 0.234) but that didn't matter on the day. Darnold was sharp, the defense was elite, and SEA peaked at exactly the right moment.
🏆
10 of 13 Playoff Winners
The model correctly predicted 10 playoff winners including both conference champions. Picking outright winners was genuinely its strongest suit all season long.

Full Season Summary

2025 Final Stats
Regular Season (272 games)
Winners
177–95 (65.1%)
Against the Spread
139–128–5 (52.1%)
Best Stretch
Wks 1–12
Playoffs (12 games)
Wild Card
4–2 / 2–4
Winners / Spread
Divisional
4–0 / 4–0
Perfect 🔥
Championship
2–0 / 1–1
Winners / Spread
Super Bowl
0–1 / 0–1
Winners / Spread
Playoffs Total
Winners
10–3 (76.9%)
Against the Spread
7–6 (53.8%)
Best Round
Divisional 4–0
💡
What Worked
Picking winners — especially once the playoffs started. Bye week teams, home field, hot team streaks, and the Bo Nix injury call all landed. The model is genuinely good at identifying the better team.
🔧
What Needs Work
Margins. The model kept predicting blowouts that turned into nail-biters. And the equal 8-game window is blind to hot streaks, declining players, and what's actually at stake for a team in Week 16.
🚀
What's Next
Recency weighting so recent games count more, a QB playoff experience factor, and better situational awareness — weather, divisional rivalries, and whether a team is fighting for their season.

How It Works

1
EPA Ratings

Calculates Expected Points Added (EPA) for each team's offense and defense based on the last 8 games of performance.

2
QB Adjustments

Adjusts spread predictions based on who's actually starting — accounting for injuries, backups, or starter rests week-to-week.

3
Edge Calculation

Compares the model spread to Vegas lines. Tier 1 edges (8+ points) indicate highest confidence picks worth tracking.

⚠️ Disclaimer

This model is for entertainment and educational purposes only. Past performance does not guarantee future results. Gambling involves risk — please bet responsibly. If you or someone you know has a gambling problem, call 1-800-GAMBLER.