Week 4 was not up to par. The model went 9–7 straight-up—a 56.3% accuracy rate that sits 8.3 points below the season-to-date average of 64.6%. The market fared identically, also hitting 56.3%, which means the week itself was genuinely difficult to predict across the board, but that offers no comfort when the baseline is supposed to be calibrated better than consensus.
The model agreed with the market on every single disagreement (zero divergences), so there was no margin opportunity in straying from consensus—the two moved in lockstep and both got hit hard.
The highest-confidence miss came at BUF, where the model favored the Bills at 70% confidence. NE won 26–29, a 10-point margin error. NE entered that game without Christian Barmore (DT, shoulder), Brenden Schooler (S, back), and Christian Gonzalez (CB, shoulder), while BUF was without T.J. Sanders (DE, illness, doubtful).
The largest projection-versus-reality gap arrived at ATL @ NO. The model projected a 1-point margin for NO; ATL won 24–45. That's a 22-point swing—the model's worst miss of the week. NO was missing three linebackers: Kaden Elliss (calf), Carl Granderson (ankle), and Anfernee Jennings (knee), all out.
The week also featured seven upsets across 16 games—a notably elevated upset rate. Six of the seven went against the model's pick: NE beat BUF, CAR beat DET, DAL beat HOU, CLE beat PIT, NYG beat ARI, and JAX beat CIN. Only one chalk pick, ATL, landed as an upset win. The model's overconfidence showed most clearly in the NE @ BUF call at 70%, but the structural problem runs deeper: too many 55–62% confidence picks all broke the same direction.
This was a week where the entire market struggled equally. That's not vindication—it's confirmation the model needs recalibration against low-confidence scenarios.