The model went 11-5 straight-up this week, matching the published market's 68.8% accuracy—a 6.2-point dip from the model's season-to-date 75% baseline. More telling: neither the model nor the market got this one right. Both agreed on all 16 games and both got the same 11 correct, suggesting the week's difficulty was structural, not a model-specific failure.
Five upsets defined the week. The model's highest-confidence miss came in NO @ BAL, where it backed BAL at 75% confidence in a 9.1-point projected win. BAL lost 17-24, a 16.1-point margin error. The market shared that conviction, posting a 9-point spread; both were wrong by the same magnitude. BAL's defense entered that game without DT Nnamdi Madubuike (Neck, Out), CB T.J. Tampa (Knee, Out), and LB Teddye Buchanan (Knee, Out), while Jesse Minter is in his first season as head coach.
The biggest projection-vs-reality gap came in a game the model actually won: CAR @ ATL. The model projected CAR by 2.6 points; CAR won by 31 (34-3). That 28.4-point swing is instructive—it was a blowout the model got directionally right but severely underestimated. ATL entered with QB Michael Penix Jr. ruled Out (Knee) and QB Tua Tagovailoa Doubtful (Oblique), leaving the offense in crisis. The model and market both got the winner right; the market happened to be closer on the spread.
Three other upsets—CLE @ TB, LV @ LAC, and CIN @ HOU—ranged from 10.5 to 18.6-point misses, all favored games where the model and market aligned and lost together. MIN @ CHI saw the model back CHI at 63% confidence; CHI lost 3-9, a 10.5-point gap.
This was a week where consensus failed across the board. The model didn't outthink the market, and the market didn't outthink the model. That's neither a validation nor an indictment—it's evidence the week was genuinely hard to predict.