Event: UFC Fight Night: Allen vs. Duncan
Date: 2026-10-10
Venue: Meta APEX, Las Vegas
Fights: 11 | Finishes: 9 (KO/TKO 7, submission 2) | Decisions: 2
What the card produced
Eleven fights produced nine finishes and two decisions. Seven of the nine finishes were recorded as KO/TKO and two as submission. The two decisions came in the women’s strawweight bout between Loopy Godínez and Ketlen Souza, and in the lightweight bout between Jai Herbert and Matheus Camilo.
The shape of the card is unusual for a modern UFC event. A card that resolves nine of eleven bouts inside the distance places a heavy load on the rating system’s update mechanism in a single night. Elo-style ratings move most when a result is unexpected relative to the pre-fight gap, and they move in proportion to how far the outcome sits from the expectation. A card with this many finishes does not, by itself, tell the model anything about how the fights were won — the model does not see method. What it sees is that nine bouts ended with a winner and a loser rather than a drawn-out decision, and that four of those outcomes ran against the pre-fight ratings by a wide margin.
The two decisions are worth noting separately. Decisions are the outcomes the model treats as the least informative per unit of rating movement, because they carry no additional signal beyond the win itself. The Godínez–Souza bout was the highest-rated pairing on the card going in (1560.1 vs 1532.1), and it resolved as a decision for the higher-rated fighter. The Herbert–Camilo bout resolved as a decision for the lower-rated fighter, but the gap (1500.9 vs 1525.2) was under the threshold used here to flag an upset, so it is recorded as a modest result rather than a major one.
The headline number for the card is the finish rate: nine of eleven. That is a card that generated a large volume of decisive results, and the ratings consequences are correspondingly large.
Main event
The main event was recorded as a submission win for Brendan Allen over Christian Leroy Duncan in round five at 2:46. Allen entered at 1676.7; Duncan entered at 1615.0. The pre-fight gap was 61.7 rating points in Allen’s favor.
The result is consistent with the ratings: the higher-rated fighter won. The model’s update for a result that matches expectation is small in relative terms, but the absolute movement is still meaningful because both fighters sit high in the middleweight distribution. Allen’s rating rises modestly; Duncan’s falls modestly. Neither fighter’s standing is transformed by the outcome, because the outcome was the one the ratings already favored.
What the result does establish is that Allen’s position as the higher-rated middleweight in this pairing is now backed by a head-to-head result rather than only by accumulated prior results. In a ratings system built on results alone, a win over a fighter rated 61.7 points below is a confirmation rather than a statement. It does not move Allen into a new tier; it holds him where the model already placed him. For Duncan, the loss is absorbed as an expected loss to a higher-rated opponent, and the rating decline is correspondingly limited. A loss to a fighter rated well above would cost less than a loss to a fighter rated below; that is the structure of the model, and it applies here.
The fifth-round submission is a detail the model does not use. The model records the win, the opponent, and the pre-fight ratings. It does not record how late the finish came, and it does not treat a late finish as more decisive than an early one. The analysis should be read with that limit in mind: the recorded result is a win for Allen, and the rating consequence follows from the win and the pre-fight gap, not from the round or the method.
Where the ratings were wrong
Four results on this card ran against the pre-fight ratings by a margin of at least 30 rating points. Each is listed with the gap as given.
Otari Tanzilovi (1453.7) beat Malcolm Wellmaker (1511.0) by KO/TKO, R2 2:39. Gap: −57.3. This was the largest upset on the card by rating margin. Wellmaker entered as the higher-rated bantamweight by a wide margin, and Tanzilovi won inside two rounds. The model will move Tanzilovi up substantially and Wellmaker down substantially, because a 57.3-point gap is a large expectation to overturn.
Francisco Prado (1429.3) beat Ismael Bonfim (1481.7) by KO/TKO, R1 4:23. Gap: −52.4. Prado entered as the lower-rated lightweight by more than 50 points and finished the fight in the first round. This is the second-largest upset by margin and the one with the earliest recorded finish among the flagged upsets.
Allen Frye Jr. (1476.0) beat RJ Harris (1526.0) by KO/TKO, R1 2:59. Gap: −50.0. Frye entered 50 points below Harris at heavyweight and won in the first round. The gap is large enough that the model will treat this as a significant correction to both fighters’ ratings.
Niko Price (1411.1) beat Leon Shahbazyan (1450.3) by KO/TKO, R1 1:58. Gap: −39.2. Price entered as the lower-rated welterweight by just under 40 points and won in the first round. This is the smallest of the four flagged upsets but still well above the 30-point threshold.
The pattern across the four is consistent: in each case the lower-rated fighter won by KO/TKO, and in three of the four the finish came in the first round. The model does not weight first-round finishes more heavily than later ones, but the clustering is worth recording as a description of the card rather than as a model input.
The remaining seven results went to the higher-rated fighter, including the main event. That is a card on which the ratings were right more often than wrong, but wrong by large margins when they were wrong. Four upsets of 39 to 57 points is a substantial correction load for a single event.
Method patterns
The card’s method distribution is seven KO/TKO, two submissions, and two decisions. The model does not use method as an input, so this section describes the card rather than the ratings.
The KO/TKO count is the dominant feature. Seven of eleven bouts ended by KO/TKO, and among the four flagged upsets all four were KO/TKO. The two submissions were the main event (Allen over Duncan) and the light heavyweight bout between Brendson Ribeiro and Felipe Franco, which Ribeiro won by submission in round one at 1:49. Ribeiro entered at 1449.6 against Franco’s 1486.2, a gap of 36.6 — below the 30-point threshold used to flag upsets in this analysis, so it is recorded as a result for the lower-rated fighter but not counted among the four flagged upsets. It is worth noting because it is the only submission on the card outside the main event and it also went against the pre-fight ratings, though by a smaller margin.
The two decisions were Godínez–Souza and Herbert–Camilo. Godínez was the higher-rated fighter and won; Herbert was the lower-rated fighter and won, but by a margin under the flag threshold. Neither decision produced a large rating correction.
The overall picture is a card on which the decisive outcomes clustered in the heavier and lower-weight men’s divisions, with the two decisions falling in the strawweight and lightweight bouts. The model does not distinguish divisions in its update rule beyond the ratings themselves, so the division clustering is descriptive only.
Get the free foundational lesson
The MCMA System starter lesson, plus new MCMA articles by email. No spam, unsubscribe any time.
What it means going forward
The rating system absorbs this card through eleven individual updates, each driven by the pre-fight gap and the recorded result. The four flagged upsets will produce the largest movements: Tanzilovi up and Wellmaker down by the largest amount, followed by Prado and Bonfim, then Frye and Harris, then Price and Shahbazyan. The main event produces a smaller, confirmatory update for Allen and Duncan.
The practical consequence is that the bantamweight, lightweight, heavyweight, and welterweight divisions each receive a correction at the points where these fighters sit. Tanzilovi’s win over Wellmaker is the single largest correction on the card and will move him toward Wellmaker’s pre-fight rating while moving Wellmaker down toward Tanzilovi’s. The same logic applies to the other three flagged pairs, in proportion to their gaps.
The card also illustrates a structural feature of a results-only model: it cannot distinguish between a fighter who was underrated and a fighter who had a single good night. The model will treat the four upsets as evidence that the pre-fight ratings were too far apart, and it will narrow those gaps. Whether that narrowing is correct depends on results the model has not yet seen. The next appearances of Tanzilovi, Wellmaker, Prado, Bonfim, Frye, Harris, Price, and Shahbazyan will test whether the corrections hold.
For the main event, the consequence is stability rather than movement. Allen remains the higher-rated middleweight in the pairing and now has a head-to-head result to support that position. Duncan’s rating declines by the amount the model assigns to an expected loss, which is limited. Neither fighter’s standing is redefined by this card.
Methodology and limits
The ratings referenced here are Elo-style ratings derived from UFC results from 2019 onward, updated after every fight. The model uses only results as input. It does not incorporate striking statistics, takedown statistics, control time, or any other in-fight metric. It does not use method of victory, round, or time as inputs to the rating update; those details are reported here as description of the card, not as model inputs.
The pre-fight ratings quoted are the ratings going into each bout. The upset threshold used in this analysis is a pre-fight gap of at least 30 rating points in favor of the fighter who lost. Four results met that threshold on this card. The Ribeiro–Franco result, at a gap of 36.6 in favor of Franco, is noted in the method section but is not counted among the four flagged upsets because the analysis applies the 30-point threshold consistently and the gap is stated as given; readers should treat the flag list as the four results explicitly identified in the verified payload.
No pre-fight prediction was made for this card. This is a retrospective analysis of recorded results and their implications for the rating system.
