Makhachev Moves to Welterweight and the Board Has a Rough Night

Islam Makhachev won the welterweight title fight over Ian Machado Garry by decision — scorecards 49-46, 48-47, 49-46 — at UFC 330 on August 15. Five rounds, no drama about the outcome from two of the three judges. The market had Makhachev at 74%; the board had him at 54%. He was the favorite on both numbers, and he delivered.

Then the rest of the card happened, and it was something. All 12 bouts were decisive — not a single contest — and the finishes came in waves. Jalin Turner put the exclamation mark on his night 39 seconds into the first round, KO/TKO over Kauê Fernandes. Neil Magny — a veteran who has seen everything — finished Ramiz Brahimaj by punches in Round 2. Esteban Ribovics stopped Edson Barboza in Round 2 as well. Lucas Fernando ended Rafael Tobias with a knee in Round 3.

The submission game matched the knockouts. Jeremiah Wells guillotined Myktybek Orolbai in Round 3 — one of the night's biggest upsets. Dustin Stoltzfus tapped Mansur Abdul-Malik with a rear naked choke in Round 2. Charles Johnson pulled off a twister — a finish you might see once in a career — to defeat Eduardo Chapolin in Round 3.

Four fights went to the scorecards: Makhachev over Garry, Chidi Njokuani over Joel Álvarez, Mackenzie Dern over Gillian Robertson in a five-rounder, and Tresean Gore over Vicente Luque. Three of those four decisions went to the fighter the board liked less. That tells you something about the night's general vibe.

Donte Johnson handled Eric McConico in 1:38 of Round 1 to round out a card where the final scoreboard read: five KO/TKOs, three submissions, four decisions. Twelve fights, twelve winners.

The gate verdict is posted right here at the top: this is a v0 model accuracy receipt. Nothing on this card was ever a pick. What follows is an honest grade of the journaled published probabilities — where they were right, where they were badly wrong.

Model accuracy on the night: 33.3% (4 of 12). Market accuracy: 66.7% (8 of 12). Model log-loss 1.0885 versus market 0.8135. Model Brier score 0.3998 versus market 0.2877. By every calibration measure, the market had a significantly better night. That's the receipt.

The board flagged five divergence cases — spots where the model's number and the market price diverged by 8+ percentage points. The model's side went 0-for-5 in those fights. Own it.

• Gillian Robertson vs. Mackenzie Dern — the model liked Robertson at 65%; the market liked Dern at 64%. They disagreed by 29.3 points, and Dern won. Biggest divergence on the card, model side lost.
• Ian Machado Garry vs. Islam Makhachev — the model had Makhachev at 54%, the market had him at 74%, a 19.9-point gap. The model's lean was toward Garry in spirit (54 to the winner is a slim margin, but the divergence flag is on Garry at 46% model vs. 26% market). Makhachev won. Market was closer to the truth here.
• Joel Álvarez vs. Chidi Njokuani — model had Álvarez at 88%, market at 73%, divergence 15 points. Njokuani won. Model side lost.
• Ramiz Brahimaj vs. Neil Magny — model had Brahimaj at 71%, market at 57%, divergence 13.1 points. Magny won. Model side lost.
• Tresean Gore vs. Vicente Luque — model had Luque at 64%, market had Gore at 51%. Model side lost.

Where the board aligned with the market and the favorite delivered: Makhachev won (both liked him), Ribovics won at 90% model / 85% market, and Johnson won at 85% / 73%. Those landed. The card's story, though, is the 0-for-5 on divergence flags and the overall accuracy gap versus the market. For a v0 model, that's the data we needed to collect. We collected it.

Five KO/TKOs, three submissions, four decisions — that's how UFC 330 settled. Eight of twelve bouts ended before the final horn.

Turner's 39-second finish of Fernandes was the fastest of the night. Fernando's knee on Tobias at 1:10 of Round 3, Johnson's punch on McConico at 1:38 of Round 1, Ribovics stopping Barboza at 1:32 of Round 2, and Magny finishing Brahimaj at 3:20 of Round 2 rounded out the KO/TKO column. Stoltzfus tapped Abdul-Malik at 4:25 of Round 2, Wells guillotined Orolbai at 1:24 of Round 3, and Johnson submitted Chapolin via twister at 1:36 of Round 3 for the submissions. The four decisions all went the full distance.

On the sim layer: the board journaled DK-point bands (p10 floor / median / p90 ceiling) for all 24 fighter slots. Band coverage landed at 83.3% — 20 of 24 inside the p10-p90 window, which sits above the model's validated 78.1% walk-forward target. Mean projection was 61.5 DK points; mean actual was 60.7, a bias of 0.9 points. The aggregate numbers were clean. The individual band misses were not — more on those below.

One note that belongs here plainly: per-bout finish-method probabilities were never published on the board. The sim journals DK-point bands only. The KO/TKO and submission results above are reported as results — there is no per-bout method number to grade them against.

Six fighters cracked 92 DK points. Here's how the board looked versus what actually hit the sheets.

1. Jalin Turner — 128.8 DK points. Pre-event projection: 60.7. Band: 3.6–118.8. He finished above his ceiling by 10 points in 39 seconds. A low-data bout, so the band was wide, but he still broke through the top. Did not cover the band.
2. Islam Makhachev — 121.9 DK points. Pre-event projection: 90.6. Band: 12.3–152. Five rounds of work, landed inside the band. Covered.
3. Mackenzie Dern — 106.8 DK points. Pre-event projection: 81.8. Band: 12–144. Five-round decision, beat her projection by 25, stayed inside the band. Covered.
4. Dustin Stoltzfus — 106.3 DK points. Pre-event projection: 27.5. Band: 1.6–92. Biggest ceiling bust on the card — exceeded his ceiling by 14.3 points. The board had him at 11% to win; he finished Abdul-Malik in Round 2 and put up a number the sim never contemplated. Did not cover the band.
5. Donte Johnson — 104.4 DK points. Pre-event projection: 77.9. Band: 12.7–123.3. Clean finish, landed inside the band. Covered.
6. Esteban Ribovics — 97.2 DK points. Pre-event projection: 82.9. Band: 35.2–127.8. Inside the band, close to projection. Covered.

Makhachev, Dern, Johnson, and Ribovics all covered their bands and performed near or above projection. Turner and Stoltzfus were the two that broke the ceiling — for very different reasons.

This is the Bad Beats segment. No spin.

Biggest Model Misses

• Myktybek Orolbai vs. Jeremiah Wells — the board had Orolbai at 90% and the market had him at 88%. Wells guillotined him in Round 3. Both numbers were badly wrong; the board was just slightly more committed to being wrong. Not a low-data bout — there was information here, and both the model and the market misread it badly.
• Dustin Stoltzfus vs. Mansur Abdul-Malik — model had Abdul-Malik at 89%, market at 85%. Stoltzfus submitted him in Round 2. Low-data bout, so the model was deferring to the market — the market was also wrong.
• Joel Álvarez vs. Chidi Njokuani — model had Álvarez at 88%, market at 73%. Njokuani won by decision. Not a low-data bout. The board was more confident in the wrong fighter than the market, and the market was also wrong.
• Ramiz Brahimaj vs. Neil Magny — model had Brahimaj at 71%, market at 57%. Magny finished him in Round 2. Both were leaning Brahimaj; neither saw the finish coming.
• Gillian Robertson vs. Mackenzie Dern — model had Robertson at 65%. Dern won a five-round decision.

Divergence Flags — 0-for-5

Every fight where the model's number diverged from the market by 8+ percentage points went against the model's side. Five flags, zero wins. That's a rough night, and it's in the ledger.

Band Busts

• Dustin Stoltzfus — biggest ceiling bust. Actual: 106.3 DK points. Ceiling: 92. Over by 14.3. He wasn't supposed to win, so the sim built a ceiling around a loss. He won and finished. The band had no room for that outcome at that scoring level.
• Myktybek Orolbai — biggest floor bust. Actual: 50 DK points. Floor: 60.2. Under by 10.2. The sim built a floor around a fighter it expected to win; he lost by submission in Round 3 and scored below the lowest projection.
• Jalin Turner — above ceiling by 10. Actual: 128.8. Ceiling: 118.8. Low-data bout, wide band, still broke through.
• Kauê Fernandes — below floor by 2.8. Actual: 0.8. Floor: 3.6. Lost in 39 seconds. Low-data bout.

Three things UFC 330 put on the record.

Makhachev is Makhachev. Moving to welterweight and winning a five-round title fight by lopsided scores on two of three cards is a statement. The board had him at 54% — a coin flip with a slight lean. The market at 74% was much closer to how the fight actually went. File that gap away.

The model had its worst night on the fights where it diverged most from the market. Five divergence flags, zero wins for the model's side. The accuracy number — 33.3% against the market's 66.7% — is a v0 data point, and it's an honest one. The log-loss and Brier scores tell the same story. The market was better calibrated on this card across the board. That's the whole point of publishing these numbers: to know where you stand.

The sim's aggregate was fine; the tails were the problem. A 0.9-point mean bias across 24 fighter slots is clean. But Stoltzfus busting his ceiling by 14.3 and Orolbai falling 10.2 below his floor are reminders that when a big underdog wins — especially by finish — the band structure breaks down by design. The model builds bands around expected outcomes. When the unexpected outcome happens, the unexpected DK total follows. That's not an excuse; it's how the math works, and it's worth understanding before you build a lineup around a projected loser.