UFC Sacramento predictions: our model is 0 for 10 vs the market

UFC Fight Night: Hernandez vs Rodrigues runs Saturday from the Golden 1 Center in Sacramento, and it starts early: prelims at 5 p.m. ET (2 p.m. local), main card at 8 p.m. ET. Paramount+ has it in the States, with CBS taking a simulcast, and the full country grid is on our how to watch page. PFL Tampa goes off at 5:30 p.m. ET on ESPN+, so the two cards will overlap for about three hours. Plan your remote accordingly.

Now the uncomfortable part.

We run a machine-learning model on every upcoming UFC bout. It rescores itself every half hour, it publishes a win probability on each fight page, and I have been quoting it in this space all month. So before I quote it again, here is its report card, pulled from our own database this morning.

Across the last five weeks of UFC cards, 60 fights that had both a model number and a closing betting line, the model picked the winner 38 times. That is 63.3 percent. The closing market favorite went 48-12 over the same 60 fights, which is 80 percent, and that is a hot streak for the market rather than a normal month, but it is still 10 fights better than us.

And then there is the number that actually matters. In those 60 fights, the model and the closing market picked opposite corners exactly 10 times.

The model went 0-10.

The ten

Every one of them, in order: Magomed Zaynukov over Damian Rzepecki, Ramazan Temirov over Steve Erceg, Jovan Leka over Alexander Poppeck, Robert Valentin over Dusko Todorovic, Borislav Nikolic over Mark Vologdin, Carol Foro over GiGi Canuto, Mackenzie Dern over Gillian Robertson, Tresean Gore over Vicente Luque, Lucas Fernando over Rafael Tobias, and Jalin Turner over Kaue Fernandes. Ten times the market said one thing, our model said the other, and ten times the market collected.

Sixty fights is five weeks. Ten disagreements is a sample you could fit in a taxi. I am not going to pretend this proves the model is broken, and a 0-10 run inside a 63 percent hit rate is roughly what you would expect if the model's contrarian calls were coin flips that happened to land badly. But I am also not going to bury it under a paragraph about how it is early days, because the honest read is simpler: when this thing tells you the market is wrong, it has so far been wrong itself, every single time.

Which brings us to Saturday, where it does it twice.

The two crossings

Kennedy Nzechukwu (14-6-1) vs. Shamil Gaziev (14-3), heavyweight. DraftKings has Gaziev at -135 with Nzechukwu coming back at +114, so the vig-free consensus across 25 books sits at 55.5 percent Gaziev. Our model gives Nzechukwu 58.0 percent. That is a 13.5 point gap and a flipped side, in a fight the market has priced within a hair of even. This is the model's best kind of disagreement, if it has one: it is not screaming that a big favorite is wrong, it is nudging a coin flip. The line has drifted toward Gaziev since it opened, from 53.5 percent to 55.5, which means whatever money has moved has moved against us.

Prediction: Nzechukwu by decision, and I want it noted that I am borrowing someone else's conviction here.

Chris Padilla (17-6-1) vs. Nasrat Haqparast (18-6), lightweight. A genuine pick'em. DraftKings has it -105 and -115, the consensus is 51.2 percent Haqparast, and the line has not moved a tick since it opened. Our model has Padilla at 57.3 percent. When the market cannot separate two men and a model can, one of them is seeing something and the other is seeing noise, and five weeks of evidence says do not assume it is us.

Prediction: Haqparast by decision. I am overruling the machine on this one, which is either good judgment or exactly the reflex that makes models useful in the first place.

The four fights where they agree on the winner and nothing else

The more interesting splits this week are not about who wins. They are about how sure anybody should be.

Photo: Show Me The Money / CC BY 3.0

Terrance Chatman (5-1) vs. Anthony Wint (7-0), heavyweight. The market has Wint at 87.5 percent. DraftKings is asking -1000. Our model has him at 65.4 percent, a 22 point gap and the widest on the card. The reason the books love him is obvious enough: Wint won his Contender Series contract on August 11 with a first-round TKO of Matt Adams and is undefeated as a pro. The reason a model might hesitate is equally obvious, and it is the same fact from the other side. He won that contract eleven days ago. Whatever the UFC is paying him, it is not paying him enough to take two heavyweight fights inside a fortnight, and Chatman is a 5-1 debutant nobody has film on either.

Prediction: Wint by knockout, at a price I would not go near.

Carli Judice (6-2) vs. Jeisla Chaves (8-0), women's flyweight. The market has Judice at 81.9 percent and drifting further that way (she opened at 77.4). Our model has her at 61.1 percent. The market is essentially pricing Chaves, who is undefeated and won her UFC debut over Yuneisy Duben in June, as a body. Judice has won three straight, two by stoppage, and she is the better-known quantity. I understand the direction. I do not understand the size.

Prediction: Judice by decision.

Jamall Emmers (22-8) vs. Lerryan Douglas (14-5), featherweight. Market: 76.4 percent Douglas. Model: 58.0 percent Douglas. Emmers is 37 and has been in this promotion long enough to be somebody's answer to a trivia question, and the line has moved away from him since it opened. The model thinks the market is burying a veteran. The market has heard that one before.

Prediction: Douglas by decision.

Gauge Young (11-3) vs. Stan Dorsainvil (5-0), lightweight. The only fight on the card where our model is the more confident party by a wide margin: 76.9 percent Young against a market consensus of 59.0. Dorsainvil is 26 and unbeaten, which is worth something to a betting market and apparently very little to a model that has watched what unbeaten 5-0 records do in this promotion.

Prediction: Young by decision.

The top of the card, where everybody agrees

Anthony Hernandez (15-3, 1 NC) is ranked sixth at middleweight and gets Gregory Rodrigues (19-6), ranked eleventh, in a five-rounder. Model: 64.8 percent Hernandez. Market: 62.8 percent. Two points apart, which after the last few hundred words feels almost companionable. The line has actually drifted toward Hernandez since opening (59.1 to 62.8) even though he is the one coming off a knockout loss, to Sean Strickland in February, while Rodrigues has won three in a row with two first-round finishes.

The co-main is Roman Dolidze (15-5) against Reinier de Ridder (21-4) at light heavyweight, and it is the rare fight where our model, the market, and I all shrug and point at the same man. Model has de Ridder at 72.8 percent, the market at 76.4, DraftKings at -375. There is a real question buried in that fight about what a grappler loses moving up a division, and it deserves its own piece rather than a paragraph.

Elsewhere: Serghei Spivac (18-6), the sixth-ranked heavyweight, meets Vitor Petrino (14-2) with the model at 54.8 percent Petrino and the market at 57.1, one of the tightest agreements on the card. MarQuel Mederos (11-1-1) faces Mason Jones (18-2, 1 NC), who is a -340 favorite on his second UFC run. And the prelims close with Shanelle Dyer as an 85 percent favorite over Elise Reed, one of five fights this week where model and market land within four points of each other.

The whole card, both numbers

Market numbers are the vig-free consensus across 25 books as of Friday morning. Model numbers are from our current version, rescored Thursday night.

Our card page updates live as the results land, so you can watch both of us get it wrong in real time. The model, at least, will not make excuses afterward.