Week 1 · Wed, 8:20 PM EDT · NBC
New England Patriots at Seattle Seahawks
Lumen Field, Seattle, WA
Market line for comparison: SEA -3.5 · O/U 44.5 (DraftKings). The model never sees this.
Rating inputs
Each starting rating is a weighted blend of five real football inputs, not just wins and losses. Every term below is measured from the 2025 season.
New England Patriots
| Input | Rating | Weight | Contribution |
|---|---|---|---|
Elo from results 14-3 in 2025, opponent-adjusted | 1607 | 80% | 1285.9 |
Scoring strength 490–320 points (+170.0 differential) → 73.3% expected win rate | 1675 | 5% | 83.8 |
Late-season form same replay with the closing 6 weeks counted double | 1641 | 15% | 246.1 |
Offense / defense split 28.8 PPG scored, 18.8 allowed → offense 1645, defense 1605 | 1750 | 0% | 0.0 |
| Blended 2025 rating | 1615.8 | ||
Offseason regression 25% back toward the 1500 league mean | 1586.8 | ||
Roster / QB adjustment No offseason adjustment applied. | 0.0 | ||
| NE 2026 starting rating | 1587 | ||
Seattle Seahawks
| Input | Rating | Weight | Contribution |
|---|---|---|---|
Elo from results 14-3 in 2025, opponent-adjusted | 1643 | 80% | 1314.2 |
Scoring strength 483–292 points (+191.0 differential) → 76.7% expected win rate | 1707 | 5% | 85.4 |
Late-season form same replay with the closing 6 weeks counted double | 1724 | 15% | 258.6 |
Offense / defense split 28.4 PPG scored, 17.2 allowed → offense 1635, defense 1646 | 1781 | 0% | 0.0 |
| Blended 2025 rating | 1658.2 | ||
Offseason regression 25% back toward the 1500 league mean | 1618.6 | ||
Roster / QB adjustment No offseason adjustment applied. | 0.0 | ||
| SEA 2026 starting rating | 1619 | ||
The calculation
Every term the model applies, in order, from both teams' starting ratings to the number on the card.
Win probability
1 / (1 + 10−(56.8 / 400))
58.1% SEA
Model spread
56.8 ÷ 25, rounded to the nearest half point
SEA -2.5
Confidence band
Move the factors yourself
Every term feeds one number: the adjusted Elo difference. Drag any of them and watch the spread and confidence follow.
Drag to see how much a stronger or weaker host changes the call.
The two ratings only matter through their difference.
The tuned value is a flat 25 points for every host.
5 Elo per extra day of rest, capped at 25 either way.
A manual thumb on the scale, in the home team's direction.
Win probability
58.1% SEA
Model value
Model spread
SEA -2.5
Elo diff +56.8 · Lean
Probability curve
The logistic curve from −600 to +600 Elo, home perspective. The dot is your current setting; the faint vertical line is where the model itself sits. Notice the curve is flattest at the extremes — near a coin flip, every point of Elo moves the number most.
Where the ratings came from
Each rating is the end state of a full 2025 season of Elo updates, regressed toward the league mean for the new season.
- 2025 final rating
- 1607.4
- Regress 25% toward 1500
- −20.5
- 2026 starting rating
- 1586.8
2025 Elo trail
Biggest swings
- W 42–13 vs CAR+49.9
- L 14–21 vs PIT−34.4
- L 13–20 vs LV−31.6
- 2025 final rating
- 1642.8
- Regress 25% toward 1500
- −24.1
- 2026 starting rating
- 1618.6
2025 Elo trail
Biggest swings
- W 31–17 at PIT+45.9
- W 20–12 at JAX+36.3
- W 44–13 vs NO+34.4
Matchup factors
Everything beyond the two base ratings that moved this specific number.
Home field
+25.0 pts
worth 3.5 points of win probability
SEA hosts at Lumen Field. Every host gets the same flat 25 points — the model does not rate crowds individually.
2025 head to head
These teams did not meet in 2025. Past meetings carry no weight in the model anyway — they are already baked into each team's rating.
Strength of schedule
Who each team actually played in 2025, measured by their opponents' current ratings. This is context for reading the ratings — it is not an input to the prediction above, because beating good teams already raises a team's Elo more.
1477
average opponent rating · 30th toughest of 32
- vs above-average teams
- 4-2
- vs below-average teams
- 10-1
1503
average opponent rating · 13th toughest of 32
- vs above-average teams
- 8-2
- vs below-average teams
- 6-1
What the model does not know
- · Injuries, suspensions and depth-chart changes.
- · Coaching and roster turnover between seasons.
- · Weather, travel distance and short-week fatigue beyond raw rest days.
- · Anything a betting market has priced in since the season ended.