UFC 333: Volkanovski vs. Evloev — The Premium Card Breakdown

UFC 333 lands at Etihad Arena in Abu Dhabi on 24 October 2026 with fourteen fights and almost no room to hide. This is not a card built around one lopsided title defence and a undercard of filler. It is a card of narrow rating gaps: four bouts the model classes as medium confidence, ten as low, and not a single bout it is willing to call high confidence. The headline fight — Alexander Volkanovski against Movsar Evloev at featherweight — is the fourth-closest call on the card by edge over a coin flip, not the first.

That makes this the most interesting kind of card to model, and the most dangerous kind to over-read. When a rating system has a real signal, wide mismatches are easy. When fourteen fights sit inside a 214-point band, the honest output is a set of modest probabilities and a clear statement of where the model is guessing. What follows is the full breakdown: every fight on the card, the model’s number, the reasoning behind it, and — just as importantly — the places where the number should not be trusted.

The card in one page

Fourteen bouts, model probabilities from the immutable Elo forecast (elo-v1.1), listed by the size of the model’s edge over a coin flip rather than by billing order.

Medium confidence (4)

• Bantamweight — Merab Dvalishvili (1702.8, 13-1 in window) vs Petr Yan (1607.0, 9-4): Dvalishvili 63.5%, edge 13.4 pts, finish 35.8%.
• Light Heavyweight — Azamat Murzakanov (1606.5, 7-1) vs Dominick Reyes (1525.8, 6-5): Murzakanov 61.4%, edge 11.4 pts, finish 50.7%.
• Light Heavyweight — Nikita Krylov (1510.5, 6-6) vs Abdul-Rakhman Yakhyaev (1589.2, 4-0): Yakhyaev 61.1%, edge 11.1 pts, finish 55.8%.
• Featherweight — Alexander Volkanovski (1620.5, 9-3) vs Movsar Evloev (1693.6, 10-0): Evloev 60.4%, edge 10.4 pts, finish 30.0%.

Low confidence (10)

• Lightweight — Grant Dawson (1664.5, 12-2) vs Nurullo Aliev (1605.5, 5-0): Dawson 58.4%, edge 8.4 pts, finish 42.5%.
• Featherweight — Daniel Santos (1545.4, 4-2) vs Bolaji Oki (1488.9, 3-3): Santos 58.1%, edge 8.1 pts, finish 44.6%.
• Middleweight — Abubakar Vagaev (1522.9, 1-0) vs Jacob Malkoun (1572.3, 6-3): Malkoun 57.1%, edge 7.1 pts, finish 30.0%.
• Heavyweight — Alexander Volkov (1634.5, 10-4) vs Rizvan Kuniev (1587.5, 4-1): Volkov 56.7%, edge 6.7 pts, finish 35.8%.
• W Strawweight — Nicolle Caliari (1499.9, 2-2) vs Iasmin Lucindo (1558.6, 5-2): Lucindo 55.5%, edge 5.5 pts, finish 42.5%.
• Flyweight — Ramazan Temirov (1566.4, 3-0) vs Lone’er Kavanagh (1545.2, 4-2): Temirov 53.0%, edge 3.0 pts, finish 55.0%.
• Bantamweight — Colby Thicknesse (1514.6, 2-1) vs Asaf Chopurov (1500.0, 0-0): Thicknesse 52.1%, edge 2.1 pts, finish 30.0%.
• Middleweight — Abus Magomedov (1562.3, 5-3) vs Cam Rowston (1575.1, 4-1): Rowston 51.8%, edge 1.8 pts, finish 51.4%.
• Featherweight — Aaron Pico (1509.2, 1-1) vs Losene Keita (1514.0, 1-1): Keita 50.7%, edge 0.7 pts, finish 46.3%.
• Lightweight — Noah Gugnon (1527.6, 1-0) vs Kody Steele (1525.2, 2-1): Gugnon 50.3%, edge 0.3 pts, finish 65.0%.

The full card average is a 56.4% favourite, a 44% finish probability, and an expected 6.2 stoppages across fourteen fights. The probability that every favourite wins — the “chalk” card — is 0.03%, or roughly one in three thousand. The model expects about six upsets by weight of probability alone. Anyone reading this card as a formality is reading it wrong.

Main event: Alexander Volkanovski vs Movsar Evloev

The number

Evloev 60.4%, Volkanovski 39.6%. Medium confidence. Ratings of 1693.6 and 1620.5 — a 73-point edge, the fourth-largest on the card. Model finish probability 30.0%, split into a 9.1% Evloev KO/TKO component, a 9.1% Evloev submission component, and 5.9% in each of the two Volkanovski finishing routes.

That 30% finish number is the lowest of the fourteen fights on this card. Read it as a statement about shape rather than about quality: the model assigns this fight the highest likelihood of reaching the judges of anything on the bill. A five-round main event with a 70% decision expectation is a fight the model thinks will be decided across rounds, not inside one exchange.

The case for Evloev

The rating gap is the whole case, and it is a real one. Evloev arrives at 1693.6 with a 10-0 record inside the rating window, and that combination — undefeated across a substantial sample — is exactly the pattern Elo rewards. The system is not evaluating highlight reels or title credentials; it is evaluating who beat whom, when, and by what method, with finish wins weighted 1.15× and K adjustments scaled by experience. Evloev’s profile under that rule set is the strongest on the card behind Dvalishvili’s.

What the model is not doing is calling him dominant. A 73-point rating gap produces a 60/40, not a 75/25, because the logistic mapping on the rating difference is deliberately conservative. The medium confidence label is the model’s own admission that a gap of this size is meaningful but not safe.

The case for Volkanovski

Volkanovski at 39.6% is the most consequential underdog number on the card. His rating of 1620.5 is the fifth-highest on the bill, and the gap to Evloev is smaller than the headline record comparison suggests — 9-3 in the window is a resume that keeps a fighter inside the top tier of the model even with three losses counted against him.

The upset routes are the ones the distribution already prices. A 5.9% KO/TKO share and a 5.9% submission share give Volkanovski roughly a 12% chance of ending the fight early in model terms; the remaining 28% of his win probability is a decision. That is the important structural point: the model does not think Volkanovski’s path runs through a stoppage. It thinks his most likely win looks exactly like Evloev’s most likely win — accumulated rounds, scorecards, a narrow margin.

Which makes this fight a test of a specific question: when two fighters are separated by 73 rating points and neither is projected to finish, how much does the gap actually show up on three scorecards? The model says: often enough to favour Evloev 60/40, not often enough to treat Volkanovski as a formality.

What would change the read

Three things, none of which are in the model. Strike volume and accuracy, takedown offence and defence, and control time are all absent from elo-v1.1 — ufcstats.com sits behind a JavaScript challenge, and the model refuses to guess at numbers it cannot verify. Fight-week news, camp changes and injuries are also unmodelled by design; the forecast is written once from results alone so that it cannot be quietly edited afterwards. Any of those factors could move a human read of this fight well away from 60/40. The model’s answer is the answer to a narrower question than the one fans ask.

Co-main: Merab Dvalishvili vs Petr Yan

This is the biggest rating gap on the card and the cleanest forecast on it: Dvalishvili 63.5%, Yan 36.5%, medium confidence, 13.4-point edge, 35.8% finish probability.

Dvalishvili’s 1702.8 is the highest rating of any fighter at UFC 333, and his 13-1 record inside the 2019-to-present window is the deepest winning sample on the bill. The 96-point edge to Yan is the largest in any medium-confidence bout. In plain terms, the model sees this as the fight where it has the most evidence and the widest separation — and even here it stops at 63.5%.

Yan’s 36.5% is not a courtesy number. His 9-4 window record keeps him at 1607.0, which is a top-five rating on this card, and the model’s finish expectation of 35.8% leaves real room for a fight that ends before the judges. The distribution splits Dvalishvili’s routes at 10.2% KO/TKO and 12.5% submission, Yan’s at 5.9% and 7.2%. If there is a spot on this card where the model’s 57.8% walk-forward accuracy is likeliest to be tested by a live underdog, this is it.

The rest of the main card

Azamat Murzakanov vs Dominick Reyes (Light Heavyweight). Murzakanov 61.4%, medium confidence, 11.4-point edge, and the second-highest finish probability on the card at 50.7%. A coin-flip-ish stoppage expectation is what a light heavyweight bout between a 7-1 fighter and a 6-5 fighter looks like in this system: real separation in the ratings, real volatility in the method distribution. Murzakanov’s KO/TKO share is 15.6%, Reyes’s 9.8%.

Nikita Krylov vs Abdul-Rakhman Yakhyaev (Light Heavyweight). The model takes the 4-0 fighter over the 6-6 one: Yakhyaev 61.1% off a 78.7-point rating edge, medium confidence, 55.8% finish — the highest finish expectation of any medium-confidence bout. This is the fight where the model’s blind spot is most visible. A 4-0 sample is small, and Elo is generous to unbeaten runs; the low number of data points is not a reason to distrust the direction, but it is a reason to distrust the precision.

Grant Dawson vs Nurullo Aliev (Lightweight). Dawson 58.4%, low confidence, 8.4-point edge, 42.5% finish. Dawson’s 1664.5 is the third-highest rating on the card; Aliev’s 5-0 window record keeps him at 1605.5. The model likes Dawson, flags the confidence as low, and offers no further opinion.

Daniel Santos vs Bolaji Oki (Featherweight). Santos 58.1%, low confidence, 8.1-point edge, 44.6% finish. Oki’s 1488.9 is the lowest rating of any fighter on the bill; Santos at 1545.4 is mid-card. The model’s edge is real but modest, and the low-confidence tag reflects how little separates the two outside the rating.

The prelims

Alexander Volkov vs Rizvan Kuniev (Heavyweight). Volkov 56.7%, low confidence, 6.7-point edge, 35.8% finish. Volkov’s 1634.5 is the fourth-highest rating at UFC 333, but the model’s gap to Kuniev (1587.5) is smaller than the name recognition would suggest.

Abubakar Vagaev vs Jacob Malkoun (Middleweight). Malkoun 57.1%, low confidence, 7.1-point edge, 30.0% finish. Vagaev has a single bout in the rating window, which is the thinnest possible basis for a rating; treat the 57.1% as a statement about Malkoun’s 6-3 sample more than about Vagaev.

Nicolle Caliari vs Iasmin Lucindo (Women’s Strawweight). Lucindo 55.5%, low confidence, 5.5-point edge, 42.5% finish. This is the one fight on the card carrying a documented adjustment: Lucindo has a 440-day layoff, and the model docks her rating by 20 points for inactivity before calculating. Without that adjustment her edge would be larger. It is applied because the model treats ring rust as a measurable cost, not an anecdote.

Ramazan Temirov vs Lone’er Kavanagh (Flyweight). Temirov 53.0%, low confidence, 3.0-point edge, 55.0% finish. A near coin flip with a high stoppage expectation — the flyweight version of a fight the model simply cannot separate.

Colby Thicknesse vs Asaf Chopurov (Bantamweight). Thicknesse 52.1%, low confidence, 2.1-point edge, 30.0% finish. Chopurov has no UFC results inside the 2023–2026 window, so he sits at the 1500.0 default rating. The model flags that explicitly. When one side of a fight is a default value, the number is close to meaningless as a prediction and useful only as a marker of how little verified history exists.

Abus Magomedov vs Cam Rowston (Middleweight). Rowston 51.8%, low confidence, 1.8-point edge, 51.4% finish. Rowston’s 1575.1 edges Magomedov’s 1562.3 by twelve points. The model is not really picking a winner here; it is reporting a rounding difference.

Aaron Pico vs Losene Keita (Featherweight). Keita 50.7%, low confidence, 0.7-point edge, 46.3% finish. Two fighters at 1-1 in the window, four rating points apart in the underlying Elo but effectively level after the mapping. The closest thing to a true coin flip on the card.

Noah Gugnon vs Kody Steele (Lightweight). Gugnon 50.3%, low confidence, 0.3-point edge, and the highest finish probability of the entire card at 65.0%. A pick’em with a two-in-three chance of ending early is the model’s way of saying the outcome is open and the method is not.

Card-level reads

Three aggregates are worth carrying into fight night, all computed directly from the fourteen model outputs.

First, the favourites are not a bloc. Multiplying the fourteen favourite probabilities gives a 0.03% chance of a clean sweep. Expect roughly six underdog wins by expectation alone (6.1 summed underdog probability). A card that goes eight-of-fourteen to the favourites is a normal card, not a bad model.

Second, the card is decision-heavy by design. Average finish probability across fourteen fights is 44%, which implies about 6.2 stoppages on the night. Four fights sit at 30% or lower on finish, including the main event. If you are watching for a card full of finishes, the model says this is not that card.

Third, confidence is doing real work. The four medium-confidence fights carry an average favourite probability of 61.6%; the ten low-confidence fights average 54.3%. When the model says medium, it means the numbers separate; when it says low, it means they barely do.

Methodology and limits

The model is called elo-v1.1 and it is deliberately narrow. It ingests verified UFC results from the ESPN official scoreboard from 1 January 2019 to the present — 4,331 decided fights across 1,780 fighters at last build — and produces an Elo rating per fighter. K is set at 48, 40 or 32 depending on experience, finish wins are weighted 1.15×, and the win probability is a logistic function of the rating gap clamped between 5% and 95%. Inactivity is penalised by 35 points after 540 days idle and 20 points after 365. Method distributions come from each fighter’s own verified finish record, KO/TKO and submission split separately.

Measured performance, from walk-forward backtesting rather than in-sample fitting: 57.8% accuracy across 3,145 fights where both fighters had prior UFC form, a Brier score of 0.2415 against 0.25 for a coin flip, and log loss of 0.6759. Calibration is monotone — 53.8% accuracy in the 50–55% band, 61.7% in the 55–65% band, 73.4% in the 65–80% band. By year: 58.6% in 2024, 60.2% in 2025, 59.3% in 2026.

The model does not include striking volume, significant strike accuracy, takedown accuracy, control time, submission attempts per fifteen minutes, betting odds, injuries, camp changes or weight-cut history. It has no opinion on who looks better in the gym. It is not a betting model, it produces no odds advice, and a 60% forecast is a 60% forecast — it will be wrong four times in ten, by construction.

The forecast file for this card was written once and is immutable by design; there are no hindsight edits and no post-fight revisions. Sources are recorded in the forecast payload, which cites the ESPN UFC fight centre for this event, retrieved 20 October 2026.

How to read this before Saturday

Take the medium-confidence fights seriously and the low-confidence ones lightly. Dvalishvili–Yan, Murzakanov–Reyes, Krylov–Yakhyaev and Volkanovski–Evloev are the four bouts where the model has both separation and evidence. Everything below them on the confidence ladder is a coin flip with extra arithmetic.

And keep the main event in proportion. Evloev at 60.4% is the model’s fourth-closest call on a fourteen-fight card; it is not a coronation, and Volkanovski at 39.6% is a live number, not a formality. The most likely single outcome on Saturday night, in model terms, is a decision in a fight the system thinks is closer than the billing suggests.

Every forecast on this card will be scored after the event against verified results, in public, with the full ledger and the running Brier score available. That is the point of writing the number down first.

Forecasts are model probabilities, not guarantees, and not betting advice. All ratings, probabilities, method splits and backtest figures above come from the immutable forecast payload for event 600061267 and the project’s published walk-forward backtest. No invented statistics, quotes, injuries or betting odds appear in this article; every number comes from the immutable forecast payload.

Card artwork: UFC.