2026 · Week 2 · Sep 17–Sep 21
16 games, simulated 10,000 times each.
A play-by-play NFL simulator, scored against the market in public. Every projection is published before kickoff and graded against the closing line afterwards — including the weeks it gets wrong.
See week 2 projections Track recordWhere the model disagrees most with the market week 2
| Matchup | Projected | Home win | Close | Edge |
|---|---|---|---|---|
 CAR at  ATL |
18.1–24.7 | 70% | -1.5 | +8.1 |
 MIA at  SF |
19.6–25.9 | 67% | +13.5 | -7.2 |
 NO at  BAL |
19.6–23.5 | 59% | +8.5 | -4.6 |
 CIN at  HOU |
19.6–27.1 | 71% | +3.0 | +4.5 |
 GB at  NYJ |
26.7–18.4 | 27% | -4.5 | -3.7 |
Edge is the model's projected home margin minus the market's. A big edge is the model making a claim, not a tip — on the 2025 holdout the closing line was still the more accurate of the two.
Scored on 2025, out of sample
Games
272
never used for tuning
Winner picked
66.5%
market 65.4%
Margin MAE
10.06
market 9.72
90% interval
89.0%
target 90%
The full breakdown, season by season and game by game, is on the track record page.
Published weeks
Model output, not advice. These are simulated distributions from a statistical model, published to record how it performs. Nothing here is a betting recommendation. Lines shown are the closing numbers carried in the underlying data and may differ from any book. 21+; if gambling stops being fun, call 1-800-522-4700.
Every Possible Sunday — built Sep 15, 2026. Play-by-play and lines from nflfastR via nfl_data_py. Team marks belong to their owners and are used here to identify clubs.