William Byron has won three of the last nine Cup races at Martinsville, more than anyone. The model's board for Sunday has him sixth, at 7.4%.

Ryan Blaney leads it at 18.7%, then Denny Hamlin at 16.5% and Kyle Larson at 14.7%.

Before anyone reaches for a pitchfork, this board is worth more than the last one. Last week this site published a Talladega board and said in the headline to ignore it. Martinsville asks a different question, and the gap comes down to what kind of racing happens there.

The model carries two settings per track type. One is how closely the finishing order tracks who actually has the fastest car, and at Martinsville it sits at 0.536 against 0.101 at Talladega. In plain terms, the quick car here usually ends up near the front, while at Talladega it ends up wherever the draft leaves it. The other setting is how much of the field gets collected in something: 5.7% at Martinsville, 36% at Talladega. The last nine Martinsville races lost between one and five cars from fields of 36 to 38. Talladega lost 14 of 40 last Sunday.

So the board steepens. At Talladega the favorite got 12.0% and the eighth choice 5.6%, barely a gap worth naming. Sunday the favorite gets 18.7% and the eighth choice is still 5.6%. Same floor, much higher ceiling. The model has an opinion this week.

Whether that opinion is any good has never actually been checked in public. Across 2023 to 2025, short tracks were its strongest kind of track, 21% better than pulling a name from the entry list, against 8.8% at the pack races. But the running scorecard here, where every board is written down before the green flag and graded afterward, has judged five races: Kansas, Las Vegas, Charlotte, Phoenix, and Talladega. Three intermediates, a flat one, and a superspeedway. Not one short track. The surface it is supposed to handle best has never been marked. Sunday it will be.

Which brings us back to Byron. The model counts how often a driver runs near the front, and wins are only one way to do that. Since 2022, Blaney's average finish at Martinsville is 4.4, the best of anyone who has started all nine, with six top fives and two wins. Byron's is 8.8, with four top fives and three wins. Byron owns the trophies. Blaney has been there every single time.

Both halves of that are worth holding. A short track rewards the driver who survives contact and claws position back, and Byron has twice turned an ordinary afternoon into a win. The model cannot see that happen. It sees finishes that swing more widely and prices them accordingly.

Hamlin is the same problem pointed the other way. He has led 1,027 laps at Martinsville since 2022, more than Byron and Larson put together, and won once. He is second on the board anyway, because laps led is not what gets counted either.

The stakes are what they are. Larson arrives with 2,441 points, 73 clear of Hamlin, with Christopher Bell 74 back, Chase Briscoe 79, Joey Logano 100, and Blaney 107, all of it here. Two races remain. Blaney is the favorite on Sunday and sixth in the standings, which is the cruel arithmetic of a late surge: the fastest car this weekend may be the one it no longer rescues.

One figure on that board comes with a warning the site attaches itself. Blaney, Hamlin, and Larson are all above 55% for a top five, and the model's own documentation calls anything over 55% there the least trustworthy number it prints, never once tested on a season the model hadn't already seen. Read the win column and leave that one alone.

If Blaney wins, nobody learns much, because favorites winning is what favorites do. The instructive result is the other one. Should Byron take a fourth, or Hamlin finally convert all those laps into something, the model will have been wrong at the kind of track where it has the fewest excuses available, and that is worth considerably more than being right at Talladega ever was.