Houston Texans vs Jacksonville Jaguars — Win Probability

Houston Stuns Jacksonville as Model Underestimated Upset Magnitude

Houston's 34-31 victory over Jacksonville qualified as a genuine upset by both simulation and market standards, but the model's pregame read significantly understated how dramatically the Texans would outperform their projection.

The model assigned Houston a 42% win probability entering the game, reflecting Jacksonville's status as the market favorite at -2.5. That spread aligned closely with the simulation's expected margin of -2.5 (favoring Jacksonville by that amount), and the model projected a final score of Houston 20, Jacksonville 22.5. The math was internally consistent: Jacksonville appeared to hold a modest but clear edge.

Instead, Houston won 34-31—a 3-point victory that reversed the expected 2.5-point deficit into an actual 3-point win. That represents a 5.5-point swing from projection to result, meaning the Texans outscored their model expectation by 14 points while Jacksonville underperformed its expected total by 8.5 points.

The magnitude of the miss matters here. This was not a nail-biter where the model nailed the favorite and the game came down to a possession or two. Houston didn't just win; it dramatically outpaced what the model believed was possible given pregame conditions. The Texans scored 34 points against a projection of 20. That 14-point gap is substantial and suggests the model either misread Houston's offensive capacity or Jacksonville's defensive vulnerabilities.

Jacksonville's fourth-quarter collapse—falling short of even its already-modest 22.5-point projection—compounds the model's error. The Jaguars not only failed to win as favored; they failed to score as the higher-seeded team facing an inferior projection.

For a model whose brand rests on accuracy, this represents a significant miss in both direction and magnitude. The upset call itself was correct in probability terms, but the scale of Houston's overperformance was far larger than the pregame numbers suggested.