How accurate is the model?

Picked the winner in 19 of 25 fights

7 more bouts had no line available and aren't graded.

76% so far — still building

For every bout we have data on, the model publishes each fighter's percent chance to win and favors whichever it rates above 50% — locked in before the fight. This is how often that favored fighter went on to win, graded right-or-wrong after the result. Bouts with no line or stats anywhere are shown but not graded.

When research moves our number

The model starts from a baseline read of each fighter, then research signals — fight-week news, matchup analysis — can nudge that read up or down. For the graded fights where a nudge actually moved our number, here's whether it pointed toward the eventual winner.

Across 8 graded fights where research moved our number, the nudge pointed toward the eventual winner on 8 of 8.

n=8 · early sample — not enough yet to call it

Is the model's confidence trustworthy?

Still building — 25 of 30 fights graded so far. Check back once we have more results to show a meaningful pattern.

Gap to the closing market

How far the model's probability sat from the market's settled view just before each fight. The typical gap is 3.9 percentage points, and on average the model sat below the market by 3.0 points.

Across 2 resolved bouts with a captured closing line.

How this works

• Forecasts are saved before the fight and never changed after the fact.
• Results come from official sources — ESPN and UFC Stats — after each event ends.
• Fights with no usable pre-fight data plus draws and no-contests are not counted toward any of the numbers above.
• Small samples are flagged until enough fights have been graded to show a reliable pattern.