Every pick. On the record.
Win probabilities saved before the action, with the model’s main matchup edges and results updated during the card.
Connecting to fight-day results…
The live counter includes only forecasts saved before an event starts. Draws, no-contests, cancelled bouts and 50/50 no-picks do not affect accuracy. See the separate historical benchmark →
Loading fight cards…
Prediction record
Saved probabilities stay visible after a fight. Correct and incorrect picks both count.
How these probabilities work
Predictor 0.1 combines 24 pre-fight features from CMR components, opponent-adjusted striking and wrestling profiles, recent form, sample strength and Elo. Its coefficients were fitted on 2018–2022 bouts only and are frozen for the live record. No betting odds are used.
The explanation under each matchup comes from the model’s actual feature contributions, grouped into readable skill areas. It is not an AI-written reason added after the probability. Fighters with at least one prior rated UFC bout use Predictor 0.1; an unrated UFC debutant uses the documented neutral-start Elo fallback and is labeled as limited history.
Retrospective evaluation from 2023 through August 29, 2026 covered 1,516 matched fights: Predictor 0.1 picked 63.2% correctly with a 0.2282 Brier score, compared with 54.7% and 0.2459 for calibrated Elo. That is a historical test, not the live record.
We check UFC’s official card every two minutes during the event, and this page refreshes every 30 seconds. Result timing depends on the source publishing its decision. Original probabilities stay locked; detailed fighter ratings refresh after full statistics are verified. These probabilities do not include injury reports, camp changes or betting markets. See validation details.