Week 1 by the Numbers: A Model's Look at the Opening NFL Slate
2026.SEP.04 · nfl · forecasting · betting · side-project
Every number below comes from a Bayesian model that predicts the full distribution of final scores for each game — not just a point spread, but the probability of every margin and every total. Team strength is estimated from play-by-play efficiency with the quarterback rated separately, so a line here is the model’s honest expectation, and every bet evaluation is done against the actual posted price, never a generic −110.
One note on discipline before the picks: edges between 2% and 10% are playable. Edges above 10% are listed but not endorsed — when a model claims that much against a competitive market, history says the model is missing information (a September market knows things about new starters and offseason change that no statistical system fully captures), so those games are treated as “the model disagrees loudly” rather than free money.
The Board
| matchup | market | model | mkt total | model total | home win % |
|---|---|---|---|---|---|
| NE @ SEA | SEA −3 | SEA −6.6 | 44.5 | 42.9 | 68.6% |
| SF @ LA | LA −3.5 | LA −4.2 | 48.5 | 54.6 | 63.9% |
| ATL @ PIT | PIT −3 | PIT −3.2 | 42.5 | 44.7 | 60.1% |
| BAL @ IND | BAL −3.5 | IND −3.1 | 47.5 | 50.4 | 59.0% |
| BUF @ HOU | BUF −1.5 | HOU −1.1 | 44.5 | 45.1 | 55.4% |
| CHI @ CAR | CHI −3 | CAR −2.2 | 47.5 | 45.3 | 43.4%* |
| CLE @ JAX | JAX −8.5 | JAX −7.8 | 40.5 | 37.7 | 73.0% |
| NO @ DET | DET −7 | DET −8.0 | 49.5 | 44.3 | 74.2% |
| NYJ @ TEN | TEN −1.5 | TEN −1.7 | 39.5 | 43.9 | 56.2% |
| TB @ CIN | CIN −3.5 | pick 'em | 50.5 | 49.8 | 49.2% |
| ARI @ LAC | LAC −10 | LAC −6.9 | 47.5 | 45.0 | 71.8% |
| GB @ MIN | MIN −1.5 | MIN −1.3 | 46.5 | 45.4 | 53.6% |
| MIA @ LV | LV −3.5 | MIA −1.4 | 41.5 | 41.5 | 44.5%* |
| WAS @ PHI | PHI −5.5 | PHI −7.6 | 44.5 | 44.6 | 73.4% |
| DAL @ NYG | DAL −3 | NYG −0.8 | 48.5 | 52.8 | 51.3% |
| DEN @ KC | KC −3 | KC −2.8 | 43.5 | 42.8 | 59.5% |
Bold = the model’s favorite differs from the market’s. Asterisk = home win % below 50 (road team favored by the model).
The Plays (2–10% edge at the posted price)
- Seahawks ML (−170) vs New England — model win probability 68.6% vs the 63% the price implies. +9.0%
- Under 44.5 (−112), NE @ SEA — model total 42.9; under hits 55.4%. +4.9%
- Texans ML (−105) vs Buffalo — the model makes Houston a small favorite at home against a near pick-’em price. +8.2%
- Under 47.5 (−108), CHI @ CAR — model total 45.3. +8.6%
- Lions −7 (−105) vs New Orleans — the model lays the week’s biggest number happily: Detroit by 8, covering 52.5% with the push priced in. +5.7%
- Under 47.5 (−112), ARI @ LAC — model total 45.0. +9.1%
- Eagles −5.5 (−108) vs Washington — the model has the champs by 7.6 at home; covers 57%. +9.8%
- Eagles ML (−218) — win probability 73.4% against an implied 68.6%. +7.0%
Game Notes
New England at Seattle (tonight). The model’s strongest side of the night: Seattle 68.6% to win, expected score roughly 25–18. Sam Darnold’s debut in Seattle against Drake Maye’s second-year Patriots prices as SEA −6.6 against a market of −3 — the raw spread edge is actually above the trust threshold, which is why the play is the moneyline and the modest under instead.
San Francisco at Los Angeles. Rams by 4.2, right on the market’s 3.5. The eye-catcher is the total: the model says 54.6 against a posted 48.5, an over probability of 68% — far past the trust cap. When a week-1 total disagrees by six points, respect the possibility the market knows something.
Atlanta at Pittsburgh. Rodgers’s Steelers price almost exactly on the market number (model −3.2 vs line −3). Mild over lean (56.5%) that just misses the playable band.
Baltimore at Indianapolis. The loudest disagreement on the slate: the market has Baltimore −3.5; the model flips it entirely to Indianapolis −3.1, driven by Indianapolis’s rating with Daniel Jones against how it weighs Baltimore’s profile. A claimed 44% moneyline edge is exactly the kind of number that fails the smell test — flagged, not played.
Buffalo at Houston. Market says Buffalo by 1.5; model says Houston by 1.1. Small flip, honest price, and the moneyline at −105 is the clean way to take it.
Chicago at Carolina. The model makes Chicago a road favorite but by less than the market (−2.2 vs −3). The playable number is the under 47.5: model total 45.3.
Cleveland at Jacksonville. Jacksonville by 7.8 against a market of 8.5 — close agreement on the side. The model leans under (59%) but past the point where the edge is trusted.
New Orleans at Detroit. Detroit by 8 — the model lays the touchdown against second-year starter Tyler Shough. Also the slate’s most emphatic under lean (64.6% under 49.5), too emphatic to trust at full claimed value; the spread is the play.
New York Jets at Tennessee. Sides agree (TEN −1.7 vs −1.5). The model total of 43.9 against a posted 39.5 is a big over lean — above the cap, no play.
Tampa Bay at Cincinnati. The market lays −3.5 with Cincinnati; the model calls it a coin flip (49.2% Bengals). Baker Mayfield’s Bucs at +3.5 grades far above the trust cap, so it’s a flagged disagreement rather than a play.
Arizona at the Chargers. Double digits is too many: model says LAC −6.9 against a −10 line. The dog-side spread edge exceeds the cap; the under 47.5 at a 57.6% hit rate is the playable piece.
Green Bay at Minnesota. Closest game on the board by both accounts (model MIN −1.3, market −1.5). Nothing playable.
Miami at Las Vegas. Another flip: market LV −3.5, model Miami −1.4 with Kirk Cousins’s Raiders debut priced below what the market thinks. Miami’s claimed edges blow past the cap — flagged, not played.
Washington at Philadelphia. The model’s best-graded playable game: Eagles by 7.6 against a 5.5 line, 73.4% to win. Both the spread and the moneyline sit inside the playable band — the rare game where the number and the price line up.
Dallas at New York Giants. Market: Dallas −3. Model: Giants −0.8, with a 52.8 total against a posted 48.5. Every market in this game grades above the trust cap — the model simply sees a different game than the price does, which in week 1 usually means the price knows the news.
Denver at Kansas City. Mahomes −3 at home vs Bo Nix; the model says −2.8. As efficient a line as the week offers. Pass.
Probabilities are model outputs, not promises. A September model runs on last season’s evidence; the market runs on everything. That’s precisely why the disciplined band exists: the middle-sized edges are where a model’s honest disagreement is most likely to be signal rather than blind spot.