Chase Briscoe won at Las Vegas last week. The model behind this site had him thirteenth in its order going in, at 2.23%.

Being beaten by a long shot is routine. Any honest forecast gets beaten by one regularly, and one that never did would be lying about how predictable this sport is. What makes this particular loss awkward is the manner of it. Briscoe led 175 of 267 laps, swept both stages, and took the flag 1.376 seconds clear after Kyle Larson clipped the wall chasing him on the final lap.

He was not lucky. He was the fastest car for two thirds of the afternoon, and the model had twelve drivers rated ahead of him.

Those are two different failures. A 2.2% driver stealing a race on a late restart is the model working as designed, because the probability has to land somewhere and sometimes it lands there. A 2.2% driver controlling the race from the front means the ratings were wrong about his speed before the engines fired, and no amount of correct calibration covers that.

The log shows the same blind spot a week earlier. Two weeks ago at Kansas the model had Briscoe fourteenth at 1.99%. He finished fourth. For Sunday at Charlotte it has him eighth at 4.06%, movement in the right direction and probably still short of how he has been running.

The Kansas forecast got the top of the order close to right, for what it is worth. Larson sat second at 15.17%, just under a point behind Denny Hamlin, and Larson won. Scored on who won, that race came out 48% better than giving every starter an equal chance. Las Vegas came out 6% worse than that same baseline.

Two races is two races. Every prediction in the log carries a timestamp from before the green flag and gets scored afterwards, so none of it is reconstructed, but a two-race record is an anecdote and will be one for a while. The number worth weighing came from twenty-nine of them: fitted on 2023 through 2025, then run across the 2026 races it had never seen, the model beat a uniform guess by 16.9% on who won, with every win-probability bucket landing within 2.3 points of its own prediction.

Sunday's numbers are already recorded. Hamlin leads at 15.68%, which reads like Kansas repeated until you look past him. Christopher Bell is at 13.78%, William Byron at 11.98%, Tyler Reddick at 10.49%, and Larson at 8.18%. Five drivers at 8% or better, where Kansas and Las Vegas each had three, and seven and a half points covering first to fifth where ten covered it at both earlier races. The top line looks familiar and the race underneath it has closed up.

Hamlin is the cautionary note inside his own number. He started from pole at Las Vegas and led 39 of the first 80 laps before spinning twice and finishing sixteenth, which the 2026 standings duly absorbed. The model reads speed, not afternoons.

The point of writing the numbers down first is that they stay written, including the ones that look foolish by Sunday night. Las Vegas sits in the log at 2.23%. Briscoe is eighth going into Charlotte, and whether that is a correction or still a lag is the thing to check on Monday.