Super Bowl LX: Seahawks vs Patriots EPA Model Breakdown
Two points.
That's the gap between what the model said and what the market said about Super Bowl LX. Seattle opened as 4.5-point favourites against New England. The V3 EPA model, built from 48,578 plays of nflverse play-by-play data and adjusted for strength of schedule, made it Seahawks by 2.5. The market was giving away value on the Patriots.
The numbers
Seattle's defensive EPA ranked second in the NFL (-0.124). New England's ranked fifth (-0.055). On offence, the Patriots actually graded higher on raw EPA (+0.097 vs +0.048), but Seattle faced a significantly harder schedule (opponents +0.060 net EPA better). After SOS adjustment, Seattle graded slightly better than raw numbers suggested, flipping the advantage from New England to roughly even.
That's what made the 4.5-point spread generous. Raw EPA favoured New England by 2.1 points. SOS-adjusted numbers favoured Seattle by about 1 point. Add a 1.5-point neutral site adjustment for the Seahawks and the model fair line landed at Seattle -2.5.
Against common opponents (seven teams), Seattle graded better both offensively (+0.109 vs +0.082) and defensively (-0.158 vs -0.099).
The total
Classic under spot. The number-two defence against the number-five defence, and New England's offence had been struggling throughout the postseason: -0.149 offensive EPA in playoff games, averaging 18.0 points per game. Seattle held San Francisco to 6 points and Los Angeles to 27. Both defences were elite. Lean under 45.5.
Matchup exploits
Seattle was vulnerable on the left side (+0.208 EPA on pass plays) and deep (+0.139 EPA). New England had the weapons to attack both.
Kayshon Boutte: 57% of targets to the left, 54% deep, with a deep ball EPA of +1.022. Perfect alignment with Seattle's weakness. Stefon Diggs: 38% left with a +1.244 EPA on deep balls. Mack Hollins: 33% left, 33% deep, a solid secondary option.
Going the other way, New England was weakest on the right side (+0.070 EPA on pass plays) and deep (+0.165 EPA). Jaxon Smith-Njigba ran 34% of his routes to the right and 27% deep. AJ Barner sat at 41% right with safe underneath targets.
Player props
The model's strongest edges on rushing yards: Rhamondre Stevenson over 49.5 (projected 64.7, clear RB1 at 17 attempts per game in the playoffs) and Kenneth Walker under 72.5 (projected 58.2 against New England's elite run defence).
On receiving yards: Hollins over 24.5 (projected 34.8, left side alignment) and Boutte over 29.5 (projected 35.7, left-plus-deep combination). Smith-Njigba at 93.5 was elite but priced fairly. No edge.
Boutte anytime touchdown at $4.50 carried the best value in the TD market: the model gave him a 30.1% chance against the book's implied 22.2%. Hunter Henry at $3.70 showed a smaller edge (29.3% vs 27.0%).
The model flagged one trap: Jaylen Henderson's anytime TD looked like a +15.4% edge on paper, but his playoff usage had collapsed to three carries in the AFC Championship Game. The number was right; the context said stay away.
The bet card
Three-star plays: Patriots +4.5, Stevenson over 49.5 rushing, Walker under 72.5 rushing, Hollins over 24.5 receiving, Boutte over 29.5 receiving. Two-star leans: under 45.5 total, Boutte anytime TD at $4.50.
Bets to avoid: Walker anytime TD at $1.58 (overpriced, model had him at 35%), Smith-Njigba receiving yards (priced fairly), Henderson rushing (role collapsed).
How it went
Seattle 29, New England 13. Not close.
The Seahawks held New England scoreless through three quarters, kicked four field goals to build a 12-0 lead, then broke it open in the fourth with an AJ Barner touchdown, another field goal, and a Uchenna Nwosu pick-six. New England scored twice in garbage time. The model's central thesis, that the 4.5-point spread was too wide and the Patriots were live, was wrong by a country mile.
Scoring the bet card honestly: two wins from seven plays.
The under 45.5 landed (total: 42). Seattle's defence was every bit the unit the model said it was; the model just picked the wrong side of the spread to back. Hollins over 24.5 receiving yards was the card's best call. He finished with 78 yards and the Patriots' first touchdown on a 35-yard catch, exactly the left-side deep threat the matchup section predicted.
Everything else missed. Stevenson over 49.5 rushing yards looked solid on paper (projected 64.7, 17 attempts per game in the playoffs), but he got just 7 carries for 23 yards as New England fell behind early and abandoned the run. Walker under 72.5 was the card's worst pick: he ran for 135 yards on 27 carries, nearly double the projection, as Seattle controlled the game on the ground. Boutte, the model's favourite matchup exploit, managed 1 catch for 21 yards on 5 targets. No touchdown.
The matchup analysis got the mechanism right on both sides. Hollins did damage on the left. Barner did score from the right side (4 catches, 54 yards, the opening touchdown of the fourth quarter). But the volume predictions were off because the game script ran away from New England early, and a model built on regular-season usage patterns can't predict a 29-0 fourth-quarter deficit.
The model's biggest lesson: SOS adjustment convinced it that Seattle's 4.5-point line was generous when in reality it was close to fair. Seattle's defence wasn't just statistically elite. It was historically dominant on the day, and the model had no mechanism to distinguish "very good defence" from "defence that holds an opponent scoreless for 45 minutes". That's a ceiling, not a bug, and it's worth stating plainly.
Two from seven. The total and one prop. That's the scorecard, published alongside the predictions, because that's how this works.
V3 EPA Model, nflverse play-by-play data. For informational purposes only. Gamble responsibly.