The gap uses whichever model you've selected. Value always comes from Market Margin, the one model whose disagreements with the betting line were backtested (flagged sides returned about +8% for favorites and +18% for underdogs on flat bets at the opening price across ~2,800 past fights). Other models' disagreements with the market haven't held up, so they don't get a value flag. It's measured against the opening line — if the price has moved, the edge will be different.

Real UFC history since 2006, using each fighter's record, form and stats as they stood before every past fight. Only patterns that hold up statistically (corrected for testing many ideas at once) and point the same way in both halves of history are shown — these describe the past, they are not predictions.

Click a model to view the fight through its eyes — probabilities, prediction and explanation update.

"Over 1.5" means the fight lasts past the middle of round 2, the way sportsbooks settle it. Fair odds are the break-even price with no vig, so compare them with a book's line. On unseen fights this model beat the plain base rate by about 10% (log loss; 46% of 5-round fights go the distance), so treat it as a tilt away from the base rate, not a certainty.

Bars show each factor's SHAP contribution to this specific prediction — how much it pushed the win probability toward one fighter, given everything else the model knows. Percentages are each factor's share of the total push across the top 30 factors.