Method
How we predict fights
Every pick on this site is our own estimate. It comes from a simple points model that uses only numbers supplied by our data source, the Boxing Data API. We don't invent stats, and when a fighter's profile is missing we don't make a pick.
The inputs (all API data)
- Professional record: wins, losses and draws
- Experience: total pro bouts
- Status: whether the fighter holds an IBF, WBA, WBC or WBO world title in the division, or their best top-10 ranking
- KO rate: KO wins ÷ wins, used only when the API gives it for both fighters
The points
| Factor | Points |
|---|---|
| Record | 2.0 × (smoothed win rate − 0.5). Smoothed win rate = (W + ½D + 2) ÷ (bouts + 4), so a 5-0 prospect isn't treated as “100%” |
| Experience | 0.30 × ln(1 + pro bouts). Diminishing returns. |
| Title / ranking | +0.40 for a world champion in the division, otherwise 0.25 × (11 − best rank) ÷ 10 for a top-10 ranking |
| KO rate | 0.6 × (KO rate − 0.5), only when both fighters have KO data |
From points to a win chance
We turn the points gap into a probability with a logistic curve: chance = 1 ÷ (1 + e−1.3 × gap). We then cap it between 25% and 75%, because records alone can't capture styles, age, inactivity, injuries or the weight cut.
Confidence labels: under 55% = Toss-up, 55–62% = Slight edge, 62–70% = Clear edge, 70%+ = Strong edge. When we have KO data, the favourite's KO rate gives a method lean: 60%+ suggests a stoppage, 40% or less a decision.
Limits
This is a deliberately simple, transparent model, not a crystal ball. Our data plan only shows fights from about the past week, so recent-form and head-to-head sections fill up over time as we store results. Fight times are shown as the API lists them, and the API doesn't state a time zone.