Chase Briscoe won at Charlotte on Sunday from the outside of the front row, leading 62 laps and holding off Christopher Bell at the end. The model gave him 4.06%.
A week before that he won at Las Vegas, where he led 175 of the 267 laps and swept both stages. The model gave him 2.23%.
Twice in three races, then, the winner has been a driver this site more or less dismissed. When it happened the first time we said we would come back to it after Charlotte, so here we are.
The first thing to say is that the model was paying attention. His number went 1.99% at Kansas, 2.23% at Las Vegas, 4.06% at Charlotte. It roughly doubled in three races and his rank climbed from fourteenth to eighth, which is what should happen when a driver posts the kind of weekends he has been posting. Las Vegas scored a 147.0 driver rating. Charlotte scored a 131.8. Both are among the five best races of his career.
Here is the odd part. By one of the model's own measures, Briscoe is currently the best driver in NASCAR.
Strength is built from three pieces, and one of them is simply recent form, the last fifteen races anywhere. Briscoe's form number going into Phoenix is 92.84. That is the highest of all thirty-nine cars entered, four hundredths clear of Ryan Blaney, which is to say level with him, and comfortably ahead of Bell at 91.92. Across those fifteen races he has averaged a 103.8 rating. Nobody has been better since June.
He is the model's eighth choice on Sunday at 3.60%, with a 20.9% shot at a top five.
That is not a glitch. Form is 35% of a driver's strength, and the other 65% is what he has done at the specific track. Phoenix gives Briscoe a 78.9, lower than any driver ranked above him. Blaney is at 109.7. The hottest car of the late summer lines up behind seven men his own form outranks, and the model is entirely comfortable with that.
Spend a minute with his Phoenix record and you can see why. He won there in the spring of 2022, leading 101 laps, and ran fourth in the fall. Since the start of 2023 his best finish at the track is seventh, and accidents have ended two of the last three. It has not been a kind place to him for four years.
But the record is the record of five different seasons, and Briscoe did not spend them in the same equipment. Three years at Stewart-Haas produced mean ratings of 71.5, 62.0 and 70.0. Two years at Joe Gibbs have produced 90.7 and 95.2. On intermediates and large ovals the gap is almost comic: a 66.6 average across forty-four starts from 2022 through 2025, with a single win, against 103.7 in twelve starts this year, with three.
The model does not know he changed cars. Nobody told it, and there is nowhere to tell it. It averages nine Phoenix races with old and new counted alike, and because nine is a thin sample it leans that average toward his record at similar flat tracks, 21 of whose 33 races came at Stewart-Haas. It returns a confident number about a driver who has not existed since 2024. Every track rating here works that way. Briscoe is just the case where it shows.
The obvious fix is to lean harder on recent seasons, or to throw the Stewart-Haas years out. Both are easy and neither is honest, not this week. Choosing which seasons count after you have seen which answer they give is how a model stops forecasting and starts flattering itself. Whatever replaces the current setup has to earn it on races the model has not seen, and that work belongs to the refit after Homestead.
So the number stands, and it is a real claim rather than a shrug: the form is genuine, and it does not travel to this particular mile. Briscoe is fourth in the standings, sixty-five points back with four to run, which means a third win in four weeks would be about considerably more than one wrong probability. Sunday tells us whether the model was respecting history or hiding behind it.