UFC Paris parlay: the biggest edge is the one we threw away

The biggest disagreement between our model and the betting market on Saturday's card in Paris is the featherweight fight that opens the main card. The books have Losene Keita at 75.9 percent against Muhammad Naimov. Our model has him at 56.8. Nineteen points is an enormous gap, Naimov is sitting there at +300, and we are not touching it.

That is the most useful thing I can tell you about this card, so let me explain it before I get to the two tickets. Everything below is our model's opinion, not betting advice, and none of it is for anyone under 18.

The edge we threw away

Our model reads a fighter's per-minute striking and grappling profile out of the official fight statistics. When we pulled Keita's, this is what came back: 38.87 significant strikes landed per minute, 24.73 absorbed, 490 percent striking accuracy, and a takedown defense of negative 1,700 percent.

None of those are real numbers. A striking accuracy above 100 percent is arithmetically impossible and a negative takedown defense is not a thing. They are what you get when a fighter has one statistically recorded UFC appearance and the per-minute denominators collapse. Keita has fought once in the promotion, a split decision loss to Nathaniel Wood in March.

So the model is not telling us the market is wrong about Keita. It is telling us it does not know who Keita is, in a voice that sounds exactly like confidence. Our trust rules already flag any model-market gap over 15 points as our own ignorance rather than the book's error, and this is the cleanest illustration of that rule I have seen on a card all year. A 19 point edge computed on impossible inputs is not an edge. It is a bug wearing an edge's clothes.

The same thing happens twice more on this card. Kurtis Campbell vs Trevor Peek carries a 17 point gap and a corner with two UFC fights. Luis Felipe Dias comes back with 98.8 percent striking accuracy and negative 26.3 percent striking defense off three appearances, which is why he is not on the safe ticket even though the probability-optimal version of it wanted him there.

Three fights out of fourteen, gone, on data quality alone. That is the pitch. A model that will not shut up is worth nothing.

The safe ticket: 7.76x

Five market favorites, every one of them with clean inputs.

Lima is the model's largest edge on any favorite anywhere on the card, nine points clear of the price. Charriere is 5-3 in the UFC with all five wins by knockout and no other way of winning a fight, he is coming off a first round knockout loss to Melquizael Costa, and he absorbs more than he lands. Lima takes people down four times per fifteen minutes.

Sy is the one bout on the card where our validated style rule fires: he takes opponents down at the 94th percentile of his division and Bukauskas defends ground striking at the 5th, a combination that has produced a 61 percent win rate across 426 historical fights. Sy also has seven inches of reach on him and an opponent who has absorbed 505 career minutes of damage to Sy's 121. He was submitted by Ion Cutelaba in March, which is exactly why he is available at -226 rather than -400.

Sygula is the price on this ticket and the reason it is worth building at all. The books have posted a literal coin flip, -108 against -109. The model has not: it gives Sygula 57.7 percent. She is 27, Cornolle is 36, and Cornolle has lost three of her last five including a submission to Joselyne Edwards in February.

The honest math. Combined price 7.76x, or +676. Our model puts the whole ticket at 16.19 percent, which prices fairly at 6.18x. The market's own implied probability for the same five legs is 10.60 percent. Model expectation is 1.26 times stake.

This ticket stops short of our usual 10x target on purpose. Adding Dias would have made it 15.2x, and I have already explained why his numbers cannot be trusted. Padding a ticket with a leg you do not believe in to reach a rounder multiplier is how these things become entertainment instead of analysis.

The upset ticket: 18.35x

Photo: SPORTS UZ / CC BY 3.0

Three underdogs, and not one of them is a lottery ticket. The model rates all three between 41 and 57 percent, which is the definition of a fight the books have priced more confidently than the numbers support.

Nursulton Ruziboev at +142 over Michael Page is the best-priced dog on the card that our model nearly favors outright, 47.5 percent against a market 39.7. Ruziboev has won three straight, submitted Andrey Pulyaev in the first round in June, and lands a knockdown per fifteen minutes where Page lands none. Page is 39, has gone to the scorecards in all three of his recent wins, and has not landed a single takedown in twenty five UFC fights, which is a specific problem against a man who completes 80 percent of his.

Sofia Montenegro at +123 is the only leg on either ticket where the model actively picks the underdog rather than merely liking her price: 57.2 percent against a market 43.2, a 14 point disagreement in the least-watched fight of the night. Both women are thin on UFC data, and I want to be precise about the difference between that and the Keita problem: Montenegro's numbers are sparse but plausible, off one recorded appearance, while Keita's are impossible. Sparse and impossible are not the same failure.

Pavel Andrusca at +240 over Nathaniel Wood needs the most caution and gets it here. Andrusca is 7-0, took the fight on short notice after Mairon Santos withdrew, and is 24 to Wood's 33. Our model has him at 41.4 against a market 28.2, and the honest reading of that 13 point gap is not that the model has spotted something in Andrusca. It is that Andrusca has zero UFC minutes on record, so there is nothing for the model to hold against him while Wood carries 189 minutes of wear. What is real, and what I would actually bet on: Wood has no knockouts in twelve wins in our records, so an undefeated 24 year old gets a full fifteen minutes to find something.

The honest math. Combined price 18.35x, or +1735. Model probability 11.24 percent, fair price 8.89x, market implied probability 4.84 percent. Model expectation is 2.06 times stake, which is the highest number on this page and the one you should trust least, because it is built on the two thinnest data profiles of the eight legs.

What is not on either ticket, and why

The main event is not here. Our model gives Salahdine Parnasse 80.4 percent and the market gives him 81.9, a gap of one and a half points, which is another way of saying there is nothing to disagree about. The co-main between Farès Ziam and Axel Sola does have an edge, 6.7 points on Sola at +129, and it is left out for a boring reason: adding it would push the upset ticket past 42x, and we cap tickets at 20x because past that point you are not expressing an opinion, you are buying a raffle entry. Same for Fabia Sintes at +106.

So: fourteen fights, three excluded on data quality, two excluded because the model and the market agree, one excluded by the price ceiling, eight actually played. At least 22 of the 24 books we track priced every fight on this card, so the minimum-book rule knocked out nothing this week. Probabilities are clipped to a floor of 5 percent and a ceiling of 90 so a single confident leg can never dominate a ticket.

The caveat that actually matters

Every combined number above assumes the legs are independent, and same-card legs are not. Fights on one card share a venue, a set of officials, a cage size, a night's judging tendencies and, in this case, the possibility that a crowd in Paris moves a close round for the French fighter. Correlation on a single card is real and it runs against you: the true probability of any of these tickets is somewhat lower than the number we printed. Treat 16.19 percent and 11.24 percent as ceilings, not estimates.

The full card, both tickets and every model probability are on the UFC Paris event page, and there is a full card breakdown with a pick for every fight if you want the ones that did not make a ticket. This is a model's opinion published as analysis, not betting advice, and it is not for anyone under 18.