How MMA Index works

MMA Index combines historical UFC performance data, contextual fighter information, and current market pricing to estimate fight probabilities — who wins, how, and what a fighter’s statistical night looks like. The same analytical foundation is used across every Pro tool: the probability on a bout page is the probability in Composite Analysis, Pick EV, the Parlay Builder, and the projections.

Model inputs are drawn from broad, verifiable categories: striking and grappling output and defense, durability and finishing history, opponent quality using rankings as they stood at fight time, recent form and activity, physical and stylistic matchup context, and current market pricing. Every feature is computed as of the night of the fight — never with information that arrived later. The market is the strongest public forecast available, so our number is presented as a disciplined disagreement with it, never a replacement: every page shows the model number and the de-vigged market number side by side.

A model probability is published only when both fighters have prior UFC data and a real betting line exists. Debutants and unlined fights are shown market-only and labeled as such — we do not publish numbers the data cannot support.

Every market is de-vigged to a no-vig probability before comparison. Expected value is computed against the price source you select — a multi-book sportsbook consensus, or the Kalshi exchange — and never against a blend. Stale exchange quotes are gated out rather than allowed to manufacture phantom edges. Prices refresh every 15 minutes around fight week, and all Pro tools price from the same refresh, so no two pages disagree on a number. Prop and DraftKings projections come from a simulation of each fight validated against real historical results.

The model is evaluated on a pre-registered forward basis: the specification and the criteria that would count as failure were locked before the evaluation window began, and are measured against the de-vigged closing market — the hardest fair benchmark that exists. Results publish here as the sample becomes meaningful. We will not quote small-sample winning streaks as evidence; a handful of cards proves nothing in either direction.

Fight results and per-round statistics from UFCStats; card and athlete data from UFC.com; career histories from Sherdog; odds from a multi-book consensus feed with derivative markets from additional books; exchange prices from Kalshi; pick’em lines from PrizePicks and Underdog. The free database this all feeds — 877 UFC events dating to 1994, more than 9,000 fights, 8,600+ with per-round statistics — is browsable by anyone: the tools read the same data you can.

• Estimates are estimates. A 70% probability loses three times out of ten; a positive-EV position can lose many times in a row.
• Debutants and fighters with minimal UFC tape are unmodeled or heavily shrunk — the data honestly cannot separate them.
• Method and prop markets for undercard fights often post late in fight week; until they do, those rows show no price and no EV rather than a fabricated one.
• Short-notice replacements, weight misses, and late fight-week news move real probabilities faster than any statistical feature set can.