Salahdine Parnasse knocked out Dan Hooker in the first round in Paris on Saturday, the outcome our model gave an 84.2% chance and the finish it called at 44%.
It was the best result of the model's night, and close to the only clean one. Across the 14 bouts at Accor Arena, it picked nine winners, or 64.3%, and called eight methods correctly, or 57.1%.
The explanation is not complicated. The model picked the same fighter as the betting market in all 14 bouts, and five favorites lost.
Five of the seven bouts the model flagged came in, including both of its two highest-conviction picks. Mario Pinto stopped Ryan Spann; the model had Pinto at 83.1% against a market price of 72.3%, the widest disagreement it carried into the card. Kurtis Campbell finished Trevor Peek at 81.8%.
The two flagged picks that missed are the ones worth naming. Modestas Bukauskas knocked out Oumar Sy in the second round, against a model read of 74.8% and a market price of -225. Pavel Andrusca beat Nathaniel Wood by decision after three rounds, against a model read of 80.1% and a price of -291. Both were bouts where the money was on the same side we were.
Below the flagged tier the card thinned out. The two lowest numbers the model put on a Paris bout both lost: Michael Aljarouj at 52.2% and Klaudia Sygula at 56.5%, both priced within a few points of a coin flip. Fares Ziam, a 65.2% pick at -150, was knocked out by Axel Sola.
The method column reads differently from the winner column. Eight of 14 is the number there, and four of those eight came in fights where the model had the winner wrong. It called a KO/TKO in Bukauskas against Sy at 45% and got one, from the other man.
Where the method read was clear, it mostly held. A decision for Felipe Lima over Morgan Charriere at 55%, and a decision. A KO/TKO for Pinto at 44%, and a knockout. Where the model split three ways it missed: Losene Keita was a 44% decision call and stopped Muhammad Naimov, and Campbell was a 46% decision call who won by submission.
The honest summary of Paris is that the model added very little to the market's own opinion. It agreed on every side, differed only on margin, and lost the same five fights the favorites lost.
The card leaves the 2026 season record at 254-110, or 69.78%, across 30 graded cards, as of Sept. 5. That figure counts every pick, including the toss-ups.