CageOracle is built and maintained by a quantitative researcher with a background in systematic trading — designing, backtesting, and risk-managing algorithmic strategies in public markets. This site applies that same discipline to mixed martial arts: rigorous feature engineering, layered machine learning models, strict out-of-sample validation, and continuous backtesting against the one benchmark that matters in any market — the price.
What the model does
CageOracle's win-probability model is an advanced machine learning system trained on career-level striking and grappling rates, round-by-round UFCStats data, Elo ratings computed per weight class, recency-weighted form, physical matchups (reach, height, age, stance), and layoff/ring-rust effects — roughly 50 signals kept from a candidate pool of nearly 300, selected by how much they actually improve predictions rather than by intuition.
One of the more distinctive pieces is a style-matchup layer, inspired by the “blade and chest” approach from competitive-game theory. Rather than reducing a fighter to a single skill number, each fighter carries two separate profiles — an offensive signature and a defensive one — so the model can represent match-ups that a simple rating can't: the durable pressure-fighter who struggles with elusive counter-strikers, the grappler who neutralizes power but not craft. It's a more formal way of capturing what fight fans already know instinctively — styles make fights, and the better fighter on paper isn't always the better fighter on the night.
The model never sees betting odds when evaluating a hypothetical matchup — there's no market price yet for a fight that hasn't been booked, so it has to work the way a scout would, from the fighters alone. For live cards, a second layer blends that fight-only read with the market's own opening and closing lines, the same way a trading desk treats a proprietary signal as one more input alongside price. That's what powers the Model vs. Vegas page.
How it's validated
Every model here is fit strictly on fights before a cutoff date and scored only on fights after it — a true holdout, never seen during training. See the Track Record page for the actual numbers, including where the model does and doesn't beat the closing line.
Data sources
Fight and round-level statistics from UFCStats.com, opening lines from BestFightOdds.com, live moneylines from The Odds API, and card structure from ufc.com. See the Disclaimer page for details on data reliability and betting risk.
What's next
Calibration improvements, fresh data through the current season, and — if it clears a bar for real edge rather than just parity with the market — tools for tracking your own bets against the model's picks.