<<<<<<< Updated upstream

What the AI predicts

• Moneyline — straight winner with calibrated probability per fighter.
• Method of victory — separate probabilities for KO/TKO, submission, and decision. Often higher edge than moneylines.
• Round totals — over/under on round count, with style-matchup adjustments (grappler-vs-striker tends to push later, two strikers tends to finish earlier).
• Fight to go the distance — yes/no probability based on both fighters' historical finish rates and matchup chemistry.
• Live in-fight — round-by-round win-probability updates based on takedowns, significant strikes, and damage.

What makes UFC different

UFC is one of the highest-edge sports we cover, for two reasons:

1. Markets are inefficient. Operators don't pour the same analytical resources into UFC as they do into NFL or NBA. The closing lines are looser. Genuine forecast edge is easier to find.
2. Style matchups matter and aren't always priced. A high-level wrestler against a striker with poor takedown defense is a much bigger favorite than the market often reflects, especially for lower-profile fighters.

The catch: UFC is also a high-variance sport on individual fights. A "85% confidence" UFC pick is still a real loss 15% of the time, and one big punch can flip any fight. The model is conservative on confidence by design — most fights cap out around 80% even when the matchup is heavily skewed.

How accuracy looks

30-day rolling, UFC-only:

Confidence band	Hit rate	Avg CLV
50–60%	54.3%	+0.7%
60–70%	64.8%	+2.3%
70–80%	74.6%	+3.7%
80%+	83.0%	+5.5%

Common questions

Does the AI predict every fighter equally well?

No. Champion-tier and ranked-contender fights have the densest training data. Newcomers and fighters with fewer than 5 UFC fights are flagged as lower-confidence by default — the model just hasn't seen enough of them.

What about Fight Night cards vs PPVs?

Same model. PPV main events tend to have the tightest closing lines (most market attention), so edge is harder to find. Fight Night undercards are often more profitable for the model.

Does the model predict round-by-round during a fight?

Yes. Live in-fight updates the win probability after each significant event (takedown, knockdown, near-finish). Useful for live tracking on fights where the favorite gets in trouble early.

Other MMA promotions?

UFC only for now. Bellator, PFL, and ONE Championship are on the roadmap but not in the current model — the cross-promotion data is sparser.

=======