Week 9 was gold—93% straight up accuracy, matching the market point for point across 15 games, and a 29.1-point jump above season-to-date baseline (93% vs. 64.2% through 120 prior games). This is the model's tier-one performance: almost perfect across the board.
Then came GB @ NE.
Both the model and market favored NE by 3 points. Both were wrong—spectacularly so. GB won 27-45, a 21-point margin miss on a play where the model expressed 59% confidence in the home pick. That's the highest-confidence wrong call of the week, and it's also the single biggest projection-versus-reality gap: the model projected NE by 3; the game landed GB by 18. The error of 21 points stood alone.
What makes this week truly exceptional is that it wasn't a general market collapse. The market posted 93% accuracy too. Every other game—14 out of 15—played to script. The model never disagreed with published consensus, meaning there were no opportunities for either divergence to save it or straying to amplify the damage. This was a week where both the model and the market nailed nearly everything, then collectively whiffed on a single matchup.
NE entered that game missing WR A.J. Brown (ankle) and RB TreVeyon Henderson (illness), among others. GB was without LBs Zaire Franklin (shin) and Micah Parsons (knee), plus C Sean Rhyan (ankle). Both rosters carried real losses.
The 8.6-point average margin error across all 15 games reflects a week where edge-case misses were rare. One brutal game doesn't erase 14 cleanly predicted ones, but it does remind us that even 93% accuracy leaves room for the unexpected—and sometimes the margin of that miss can be enormous.