The model went 10-for-14 straight-up this week (71%), a sharp nine-point climb from its season-to-date 62% baseline. More tellingly: the published market also went 10-for-14. When the entire week is this hard to predict, a model matching the consensus line isn't a failure—it's a reflection of genuine uncertainty baked into the odds themselves.
The model never diverged from the market this week, so there's no separation signal to measure. What matters instead is which games surprised both.
HOU @ NYG was the costliest miss. The model backed HOU at 78% confidence with a projected 10.5-point margin, but NYG won 27-21 anyway—a 16.5-point gap between what was expected and what happened. HOU entered the game missing RB David Montgomery (groin) among four out-and-out absences; NYG was in John Harbaugh's first season as coach. The market's 10.5-point line on HOU matched the model's conviction exactly, and both were wrong together.
TB @ CAR produced the model's largest projection-vs-reality crater: expected a 2.9-point CAR win, got a 43-13 TB blowout instead—a 32.9-point swing. Again, the market (CAR by 3) was equally blindsided. Four other upsets—MIA @ NYJ, SF @ ATL, and two others—rounded out a week where consensus and model alignment simply meant they shared the same blind spot.
The average margin error was 10.2 points. In raw accuracy, 71% is a good week that outpaces the season floor. In context, it's a week where the model did exactly as well as the market did, which means the variance lived in the games themselves, not in the model's ability to separate signal from noise. Solid performance on a genuinely tough slate.