Track Record

Track Record

A real backtest, scored honestly, shown in full — including where it doesn't win.

CageOracle hasn't taken a public bet yet — this site is pre-launch, so there's no live betting record to point to. What you'll find below instead is a backtest: the exact model currently running this site, scored against real UFC results it never saw during training. Every number here is reproducible from the data in this repo.

Full holdout

716 fights, 2+ years — the long-run validation

The model trains only on fights before 2024-06-01, and is scored on every fight since with market odds available — genuinely never-seen data.

Model accuracy
69.1%
495/716 favorites correct
Vegas accuracy
69.0%
494/716 favorites correct
Model AUC
0.750
Vegas AUC
0.750
Model log-loss
0.592
Vegas log-loss
0.591

These numbers are effectively tied. That's the honest headline: CageOracle's model matches the accuracy of the closing betting line — which is itself one of the hardest numbers to beat in sports, since it already prices in the market's collective information. We do not claim to reliably beat the market, and we'd rather show a tied number than a cherry-picked winning streak.

What about the value bets?

Over the same 716 fights, 141 were flagged a 5%+ edge (same threshold as the “Value” badge on the Model vs. Vegas page). Hit rate was 77.3% (109/141) — but a high win rate on mostly-favorite picks doesn't automatically mean profit once the bookmaker's vig is priced in: flat $1-unit staking on every flagged edge returned -3.2% over this period. That ROI number is shown deliberately, not just the flattering win rate, because win rate alone is easy to make a model look good with.

Illustration, not proof — small sample

Most recent 2 events

The last two UFC cards, scored exactly the same way: UFC 329: McGregor vs Holloway 2 (July 11, 2026) and UFC Fight Night: Fiziev vs Torres (June 27, 2026, Baku) — 24 fights total. n=24 is too small to be meaningful on its own; the 716-fight number above is the one that matters. This section is a worked example of the methodology, not a second proof point.

Model accuracy
70.8%
17/24 favorites correct
Vegas accuracy
79.2%
19/24 favorites correct
Model AUC
0.867
Vegas AUC
0.874
Model log-loss
0.501
Vegas log-loss
0.500
MatchupModel Pick (%)Vegas Pick (%) Actual WinnerResultEdge
UFC 329: McGregor vs Holloway 2 · 2026-07-11
Adrian Yanez vs Cody GarbrandtAdrian Yanez (83.1%)Adrian Yanez (77.2%)Adrian Yanez✓ Correct5.9% Value
Alessandro Costa vs Cody DurdenAlessandro Costa (71.5%)Alessandro Costa (66.7%)Alessandro Costa✓ Correct4.8%
Cesar Almeida vs Damian PinasDamian Pinas (67.7%)Damian Pinas (68.9%)Damian Pinas✓ Correct1.2%
Conor McGregor vs Max HollowayMax Holloway (71.5%)Max Holloway (71.4%)Max Holloway✓ Correct0.1%
Cory Sandhagen vs Mario BautistaCory Sandhagen (60.8%)Cory Sandhagen (56.9%)Mario Bautista✗ Wrong3.9%
Elisha Ellison vs Gable StevesonGable Steveson (91.1%)Gable Steveson (89.9%)Gable Steveson✓ Correct1.3%
Farid Basharat vs John GarzaFarid Basharat (89.5%)Farid Basharat (84.0%)Farid Basharat✓ Correct5.5% Value
Kai Kamaka III vs Luke RileyLuke Riley (68.1%)Luke Riley (68.8%)Luke Riley✓ Correct0.7%
King Green vs Terrance McKinneyKing Green (54.7%)King Green (51.5%)King Green✓ Correct3.2%
Nikita Krylov vs Robert WhittakerNikita Krylov (50.0%)Robert Whittaker (52.9%)Robert Whittaker✗ Wrong2.9%
Tracy Cortez vs Wang CongWang Cong (50.2%)Wang Cong (52.2%)Wang Cong✓ Correct2.0%
UFC Fight Night: Fiziev vs Torres · 2026-06-27
Abdul Rakhman Yakhyaev vs Julius WalkerAbdul Rakhman Yakhyaev (88.5%)Abdul Rakhman Yakhyaev (82.7%)Abdul Rakhman Yakhyaev✓ Correct5.9% Value
Abus Magomedov vs Michal OleksiejczukMichal Oleksiejczuk (50.8%)Michal Oleksiejczuk (52.4%)Abus Magomedov✗ Wrong1.6%
Andrey Pulyaev vs Nursulton RuziboevNursulton Ruziboev (60.1%)Nursulton Ruziboev (61.3%)Nursulton Ruziboev✓ Correct1.2%
Asu Almabayev vs Charles JohnsonAsu Almabayev (79.4%)Asu Almabayev (73.9%)Asu Almabayev✓ Correct5.5% Value
Bekzat Almakhan vs Jean MatsumotoJean Matsumoto (60.3%)Jean Matsumoto (61.3%)Jean Matsumoto✓ Correct1.0%
Brunno Ferreira vs Ikram AliskerovIkram Aliskerov (70.7%)Ikram Aliskerov (70.7%)Ikram Aliskerov✓ Correct0.1%
Daniil Donchenko vs Theodor BerggrenDaniil Donchenko (85.0%)Daniil Donchenko (79.8%)Daniil Donchenko✓ Correct5.1% Value
Eric Nolan vs Farman HasanovFarman Hasanov (59.0%)Farman Hasanov (60.4%)Farman Hasanov✓ Correct1.5%
Javier Reyes vs Kaan OfliJavier Reyes (71.0%)Javier Reyes (66.1%)Kaan Ofli✗ Wrong4.9%
Jefferson Nascimento vs Tahir AbdullayevJefferson Nascimento (62.0%)Jefferson Nascimento (57.9%)Tahir Abdullayev✗ Wrong4.2%
Manuel Torres vs Rafael FizievManuel Torres (52.0%)Rafael Fiziev (50.0%)Rafael Fiziev✗ Wrong2.0%
Matheus Camilo vs Nazim SadykhovNazim Sadykhov (61.9%)Nazim Sadykhov (62.8%)Matheus Camilo✗ Wrong0.8%
Michel Pereira vs Shara MagomedovShara Magomedov (77.6%)Shara Magomedov (76.8%)Shara Magomedov✓ Correct0.7%

Every fight here is scored using CageOracle's live model, evaluated strictly on fights after the training cutoff — a true holdout the model never saw during development. “Vegas” is the median closing decimal odds across non-exchange sportsbooks, de-vigged proportionally within each matchup, so the comparison is apples-to-apples rather than raw market price. A “value” fight is one where the model's edge over the de-vigged market price is 5% or more, the same threshold used for the Value badge on the Model vs. Vegas page. ROI reflects flat $1-unit staking at raw decimal odds — vig included, not stripped out for payout.

Historical and backtested performance is not a guarantee of future results — see the Disclaimer for the full picture.