Prop record

Round totals, inside the distance and the method board: what the model makes them, against BetOnline's own quote. What would still make this wrong

Edge over 2% Held out since 2021-01-01
Walk-forward, distance lines +9.3% on 5,167 bets over 6 years
Years in profit 6 of 6 walk-forward, distance lines
Held-out record +11.75% 6,628 bets · t = +5.50
On the cards ahead 15 of 23 priced markets clear 2%

Upcoming

The props that clear 2% against BetOnline's own quote Every market →
BoutBetPrice ModelMarketEdge
Contender Series 2026: Week 7 2026-09-22
Marcos Degli vs Paris Moran Over 2.5 rounds +128 59% 42% +15.1%
Marcos Degli vs Paris Moran Over 1.5 rounds +109 63% 46% +14.7%
Callum Connor vs Piero Guaylupo Over 1.5 rounds -165 71% 60% +8.4%
Alvi Dasuyev vs Jaden Ortega Over 1.5 rounds +111 52% 45% +4.7%
Damian Piwowarczyk vs Emilio Quissua Over 1.5 rounds -101 55% 48% +4.3%
UFC Fight Night 289: Rosas Jr. vs. Barcelos 2026-09-26
Montel Jackson vs Ricky Simón Over 2.5 rounds -180 78% 62% +13.7%
Robert Bryczek vs Rodolfo Vieira Over 1.5 rounds -120 67% 53% +12.9%
Alatengheili vs John Castañeda Over 2.5 rounds -300 86% 73% +11.0%
Brady Hiestand vs Rinya Nakamura Over 1.5 rounds -167 72% 61% +9.6%
Melissa Amaya vs Tina Black Over 2.5 rounds -285 83% 72% +9.2%
Elves Brener vs Josiah Harrell Over 2.5 rounds -130 64% 55% +7.7%
Ailín Pérez vs Norma Dumont Over 2.5 rounds -500 90% 81% +6.4%
Christian Edwards vs Rodolfo Bellato Over 1.5 rounds -160 66% 60% +4.7%
Mickey Gall vs Sedriques Dumas Over 1.5 rounds -125 59% 54% +3.6%
Ilimbek Akylbek Uulu vs Mehemmedeli Osmanli Over 1.5 rounds -170 66% 61% +2.7%

Walk-forward

The honest test: each year's model is rebuilt from scratch on only what came before, then bet on that year alone

Distance

YearBetsUnitsROIt
2026 588 +32.5 +5.52% +1.19
2025 790 +52.5 +6.65% +1.56
2024 911 +157.1 +17.24% +4.44
2023 1,107 +110.1 +9.95% +2.68
2022 1,011 +89.9 +8.89% +2.69
2021 760 +35.9 +4.72% +1.31
All5,167+478.0 +9.25%

Method

YearBetsUnitsROIt
2026 97 +11.0 +11.34% +0.55
2025 94 +15.1 +16.01% +0.71
2024 100 +68.0 +67.95% +2.55
2023 202 +57.8 +28.61% +1.81
2022 88 +24.2 +27.50% +1.04
2021 74 +12.5 +16.89% +0.61
All655+188.6 +28.79%

The record, by market

Held out since 2021-01-01
MarketBetsStrikeROIt
the distance2,00149% +12.43% +4.15
2.5 rounds1,84861% +5.36% +2.45
method board1,44733% +15.36% +3.14
1.5 rounds1,10562% +11.96% +4.25
inside the distance22748% +33.61% +3.11
Everything6,628 +11.75%+5.50

By year

Every market together
YearBetsROIt
2021890 +2.79% +0.62
20221,279 +8.09% +1.91
20231,547 +18.09% +3.61
20241,138 +20.55% +3.62
2025971 +9.48% +1.74
2026803 +5.52% +0.91
How this works and what would still make it wrong

Two stages. The first learns how a fight ends from every graded fight in the database — 186,139 in the fit window, with no market input at all. It knows the weight class (43.9% of strawweight fights reach the cards against 19.8% of heavyweight ones), whether it is a women's bout (42.7% against 29.7%, and the gap survives controlling for weight), how often each fighter has finished an opponent, how often he has been finished, and what share of the scheduled clock his fights normally use. The second stage blends that single number with the book's own de-vigged price, two coefficients fitted on the priced fights.

Why it is built that way. The previous version used the market line as an input, so it could only be fitted where a price existed — 2,232 fights. Eleven coefficients on 2,232 fights is mostly noise, and an earlier evaluation duly concluded that every term added to the closing line made the forecast worse. Fitted on 186,139 instead and compressed to one number, it beats the closing line on all three distance lines and on the method split.

The three distance lines are one survival curve read at three points, so they are forced into order: a fight that reaches the cards has already passed 1.5 rounds. Fitted independently they violated that on 62% of fights, and the page was briefly offering "under 1.5 rounds" and "goes the distance" on the same bout. Method is P(he wins) × P(that method, given he wins) — the moneyline model, then the split.

What would still make this wrong. The walk-forward refits the coefficients each year but not the choice of features, and those were chosen while looking at this same period. Prices are BetOnline's own pre-fight closes, which are real, but prop limits are a fraction of moneyline limits — the constraint is how much you can get down, not the edge. Only the distance lines are walk-forwarded; the method and inside-the-distance records use the fixed 2021 cutoff. And t is computed across fights rather than bets, because the three distance lines on one card land together.