This was a dark day for the model. A 57.1% straight-up accuracy—down from a season 62.8% baseline—represents performance that was not up to par. Eight wins in fourteen games tells the full story: the model whiffed on six of the week's tightest contests and never found solid ground.
The worst damage came on ARI @ LAR. The model rated LAR a 16-point favorite with 84% confidence. ARI won 16-18. That 15-point margin miss—favored team losing by 2—was the week's most painful calibration failure. The market also backed LAR at 13-point favorites and also lost, but the model's confidence was materially higher, making the miscalibration uniquely sharp. ARI entered the game missing defensive backs Andrew Wingard and Wydett Williams Jr., plus RB Jeremiyah Love, but the model's overestimate of LAR's edge was severe enough to dwarf any injury context.
Four more upsets followed in tight spots: TEN beat IND (26-20, model favored IND 6.5 points with 69% confidence); CLE beat BAL (13-10, model favored BAL 59% confidence); HOU beat JAX (34-31, model favored JAX 58% confidence); DEN beat SEA (17-14, model favored SEA 57% confidence). CLE's win came with five key BAL absences—including QB Lamar Jackson—while two first-season coaches (Todd Monken at CLE, Jesse Minter at BAL) arrived without extensive film to train on. GB also beat DAL 23-13 despite the model backing DAL at 53% confidence.
The model agreed with the market on every pick this week and matched its record: 8-6. That identical performance matters. It wasn't a week where the model uniquely failed to spot market mispricings; the entire market read Week 6 poorly. But the model's confidence levels were often misplaced—especially at LAR—and that's the distinction. The margin of error averaged 6.7 points, with several predictions landing 11-15 points away from reality.