Ryan Blaney won the season finale at Homestead from ninth on the grid, having led seven laps all afternoon. The model had him fourth on its board at 10.2%, behind a favorite who finished twelfth.
That makes seven races scored on this site's public log since it opened, and seven times the model's top pick has failed to win. Written that way it sounds like an indictment. It is closer to a rounding error.
Add up what the board actually claimed and the picture changes. Across those seven races the favorites were priced at 16.1%, 16.2%, 15.7%, 15.4%, 12.0%, 18.7%, and 15.8%. Sum them and the model expected its favorite to win 1.1 races. Not five, not three. One. And the chance of the favorite winning none of the seven, given exactly those numbers, is 30.2%.
So the headline record is a thing that happens roughly one time in three. A 15% favorite is a driver the model expects to lose about five times out of six. Stringing seven of those together is not a slump, it is an ordinary Tuesday with a small helping of bad luck.
Meanwhile the thing the board is actually measured on finished the year at plus 24.4%, which is how much better it was than pulling a name out of the entry list. Six of the seven races scored positive. The one that did not was Las Vegas in October, where Chase Briscoe won at 2.2% and the model took a 6.1% beating for it.
Sunday was close to its best day. Blaney at 10.2% scored plus 38.2%, second only to Kansas in September. The model did not pick the winner, and it got closer to him than a coin would have, which has been the shape of the whole season.
As for the championship, it was over in the way these things usually end, which is quietly. The preview here on Thursday worked out that Larson only had to finish 28th to be champion no matter what anyone else did. He started fourth, finished twelfth, and led not one lap of the race that made him a three-time Cup champion. Christopher Bell came fourth and Denny Hamlin sixth, and neither was ever close enough for the arithmetic to matter.
Hamlin's afternoon deserves its own line, because it is the season in miniature. He started second, won the second stage, and led 87 laps, more than anybody. He finished sixth. Across the ten Martinsville races since 2022 he has led more laps than any other driver, with a single win to show for it. Leading laps is a wonderful way to spend a Sunday and a poor way to collect trophies.
Then there is Tyler Reddick, who won six races this year, more than any driver in the series, and ended up fifth in the final standings, 104 points behind a champion with three. Fifteen different drivers won a race in 2026. The one who won most is not the one holding the trophy, and the format did exactly what it was built to do.
The last number worth recording is the one this site spent the season telling people to ignore. Three times all year the model put a driver above 55% for a top five, all three at Martinsville, and all three missed. Zero for three against an average prediction of 61%. That was the figure labeled the least trustworthy thing here, and the first time it was ever graded in public it earned the label.
One season of forward testing is a small sample and nobody should pretend otherwise. But the record now reads seven races, a favorite that never won, a board that beat random by about a quarter, and a published warning that turned out to be correct. For a model that publishes before the race rather than after it, that is a reasonable place to stop for the winter.