The model's board for the YellaWood 500 has Ryan Blaney at 12.0%, Joey Logano at 10.1%, and William Byron at 9.3%. Kyle Larson is fourth at 7.8%.

Ignore all of it.

This site publishes odds for all 36 races, and at drafting tracks it publishes them with a caveat on the page, because the measurement behind them is poor and has been from the start. Win skill over a uniform guess by track type, on the 2023 to 2025 seasons the model was fitted on: short tracks +21.0%, intermediates +19.6%, flat tracks +17.6%, road courses +12.1%, drafting +8.8%. Drafting was the weakest type on the board even where the model had every advantage.

Then came the honest test. On 2026, held out of the fitting entirely, drafting scored -3.7% across five races. Below zero means worse than picking a name at random from the entry list.

What makes this more than an apology is that the model's own settings agree. Two numbers get fitted per track type. Coupling controls how tightly the finishing order follows a car's pace, and at Phoenix last week it was 0.477, while at Talladega it is 0.101. Trouble is the chance a car finishes but loses track position on the way, and it runs 13.2% at Phoenix against 36% here. The model is saying that pace barely survives to the flag and that more than a third of the field gets shuffled. It has encoded the problem accurately and still cannot solve it.

You can see the flatness in the board itself. At Phoenix the favorite sat at 15.4% and the eighth choice at 3.6%, a gap of more than four times. Here the favorite is 12.0% and the eighth choice, Chase Briscoe, is 5.6%, barely double. No driver gets above 37.1% for a top five, where Blaney alone was given 51.6% a week ago. Thirty-seven cars, and the model cannot separate them.

None of which requires a model to believe. Talladega has run nine Cup races since 2022 and produced nine different winners: Ross Chastain, Chase Elliott, Kyle Busch, Blaney, Tyler Reddick, Ricky Stenhouse Jr., Austin Cindric, Briscoe, and Carson Hocevar. They started 19th, 16th, 17th, 10th, 18th, 32nd, 7th, 2nd, and 12th. Stenhouse won from 32nd. Across those nine races 20.1% of starters retired, against 11.5% at the tracks that don't draft in the same period.

That is the race the championship walks into. Larson leads with 2,370 points and Christopher Bell is five behind, with Briscoe 24 back, Denny Hamlin 31, Blaney 40, and Logano 44, all of it here. Three races remain, and the closest title fight in years gets handed to the one Sunday where running well is worth less than being in the right line at the right moment.

It is also where the model's other admitted blind spot bites hardest. It treats every car as independent and has no teammate term at all, which matters least at Phoenix and most at a track whose entire outcome turns on who agrees to push whom.

The forward test will score Sunday like any other race, and the result will probably be ugly. That is the point of scoring it. A model that only reports the weeks it looks good is a marketing document, and the drafting number is the clearest evidence on this site that these are measurements rather than claims. Blaney at 12.0% is not a prediction worth acting on. It is the honest output of a method that knows, and says, where it stops working.