UFC Paris scored: the favorites held, our numbers did not
We scored every UFC Paris prediction against the closing odds: our model went 8 of 14, the market 9 of 14, and three upsets beat both of us in Paris.
Every card we cover gets a pre-fight probability on every bout, and every card we cover gets scored afterwards whether the numbers looked good or not. UFC Paris is one of the ones that did not.
Our model went 8 for 14 in Paris. The closing betting market went 9 for 14. On Brier score, which punishes confident wrongness rather than just counting picks, the market finished at 0.205 and we finished at 0.227. Losing to the market on a fourteen-fight card is not a crisis, it happens often and it will happen again, but it is worth walking through where the two of us diverged, because that is the only part of a scoring exercise that teaches anything.
The main event was the easy one
Our model gave Dan Hooker a 19.6 percent chance of beating Salahdine Parnasse. The market, once you strip the bookmaker margin out of the two-way price, had him at 20.2 percent. Those are the same number. A debuting KSW champion was a heavy favorite over the number 12 lightweight in the world and nobody at all disagreed about it.
He was knocked out in 155 seconds. Both sides of the ledger got the main event right, and it was the least interesting call on the card precisely because there was no argument to have.
Where the numbers and the market actually disagreed
Three bouts, out of fourteen, had our probability on one fighter and the market's on the other. We went 1 and 2.
| Bout | Our probability | Market (vig removed) | Winner |
|---|---|---|---|
| Aljarouj vs Sintes | Sintes 50.8% | Aljarouj 53.4% | Sintes, unanimous decision |
| Duclos vs Dias | Dias 58.9% | Duclos 52.2% | Duclos, KO/TKO round 1 |
| Benouaich vs Montenegro | Montenegro 57.2% | Benouaich 55.4% | Benouaich, submission round 2 |
Look at those margins. All three were coin flips on both sides, and the one we won, Fabia Sintes over Michael Aljarouj, was decided by a two-point edge. Nothing in that table is a signal. If a numbers system is only beating the market by picking the correct side of a 51-49, it is not beating anything, it is being lucky on the right days.
The place we lost real ground was Delphine Benouaich, where our 42.8 percent on her was the single most wrong confident number of the night: she landed 146 significant strikes, 104 of them to the head, and submitted Sofia Montenegro in the second. There was nothing close about it.
The three real upsets beat everybody
Photo: Hrhr prph / CC0
The bouts that decided the night were the ones where we and the market agreed and were both wrong together. Axel Sola was a 41.7 percent underdog to the market and a 48.7 percent one to us, and he knocked Farès Ziam out in 100 seconds. Modestas Bukauskas sat at 31.3 percent to the market and 29.3 percent to us, and he knocked Oumar Sy out in the second after losing most of the grappling. Pavel Andrusca was at 29.7 percent and 41.4 percent respectively, and he out-wrestled the number 8 featherweight in the world for fifteen minutes.
That is the honest shape of a bad scoring night. It was not that our numbers read a fight differently from the market and lost the argument. It is that three fighters did something neither a market of thousands of people nor a system trained on twenty years of results expected, which is the entire reason anybody watches.
The one place both sides were properly rewarded was Michael Page at 64.2 percent to the market and 52.5 percent to us, winning a 29-28 across the board that nobody enjoyed. Our number was closer to what the fight actually looked like than the market's was, which is a small consolation on a night short of them.
The full card, scored
| Bout | Our pick | Market pick | Result |
|---|---|---|---|
| Hooker vs Parnasse | Parnasse 80.4% | Parnasse 79.8% | Parnasse, both correct |
| Ziam vs Sola | Ziam 51.3% | Ziam 58.3% | Sola, both wrong |
| Page vs Ruziboev | Page 52.5% | Page 64.2% | Page, both correct |
| Donchenko vs Soriano | Donchenko 71.1% | Donchenko 67.6% | Donchenko, both correct |
| Campbell vs Peek | Campbell 59.6% | Campbell 76.9% | Campbell, both correct |
| Keita vs Naimov | Keita 56.8% | Keita 76.4% | Keita, both correct |
| Charriere vs Lima | Lima 73.2% | Lima 64.8% | Lima, both correct |
| Pinto vs Spann | Pinto 76.3% | Pinto 76.8% | Pinto, both correct |
| Sy vs Bukauskas | Sy 70.7% | Sy 68.7% | Bukauskas, both wrong |
| Wood vs Andrusca | Wood 58.6% | Wood 70.3% | Andrusca, both wrong |
| Aljarouj vs Sintes | Sintes 50.8% | Aljarouj 53.4% | Sintes, we were right |
| Cornolle vs Sygula | Sygula 57.7% | Sygula 53.5% | Cornolle, both wrong |
| Duclos vs Dias | Dias 58.9% | Duclos 52.2% | Duclos, market right |
| Benouaich vs Montenegro | Montenegro 57.2% | Benouaich 55.4% | Benouaich, market right |
Every probability on this page was published before the fights. The per-bout numbers live on the event page and the running accuracy record is at /power-index/accuracy.
What I take from it
Nothing about the system, honestly. Fourteen fights is noise, one card cannot validate or invalidate anything, and the correct response to a 0.227 Brier is to write it down and keep publishing the numbers before the fights rather than after. The response I am more interested in resisting is the other one: quietly not scoring the cards that go badly.
The next UFC card is in Las Vegas on September 12, and the probabilities for it will be up before the walkouts, same as always. Come back and hold them against the results.