How it works
Every fighter is rated 0–99 on the four phases of an MMA fight — striking, wrestling, grappling, and cardio + fight IQ — blended into one overall. The scores are derived, not hand-set: each fighter's public UFCStats career numbers are ranked against the rest of the roster and mapped onto the scale. It's a scout-style read to visualize and compare skill sets — not an official rating.
The inputs: UFCStats career averages
For all 175 ranked fighters we start from the eight career averages UFCStats publishes on every fighter's page, plus their win/finish split (KO-TKO / submission / decision) and record:
- Striking — strikes landed & absorbed per minute, accuracy, defense.
- Grappling — takedowns per 15 min, takedown accuracy & defense, submission attempts per 15 min.
- Outcomes — wins, losses, and how the wins came (KO / submission / decision).
These are counting stats: descriptive, opponent-blind, and shaped by style (a counter-striker posts low volume; a control grappler looks quiet on strikes). That's why we read all four axes, not just the composite.
Career numbers → four 0–99 axes
For each component we rank a fighter against the whole roster — a percentile — then blend the components into one axis and map it onto the 60–99 band you see (so the color ramp and bars spread across a visible range). Percentiles, not raw numbers, so a fighter is judged against the field:
The four axes → one Overall
The axes combine into an overall with a striking-heavy weighting. These weights aren't arbitrary — they're backtested against real fight outcomes (a time-aware, leak-free logistic model), tuned so the composite leans on the phases that actually predict who wins. Striking and durability carry the most signal:
A fixed pound-for-pound rank then orders everyone by overall, independent of how the table is sorted or filtered.
Opponent-adjusted (level of competition)
Counting stats are opponent-blind — five takedowns on a debutant look like five on a champion. So we also fit an opponent-adjusted model (FVOA-style) on fight-by-fight data: each phase is credited against the fighters actually faced, with partial pooling and small-sample shrinkage. That yields each fighter's strength of schedule and adjusted striking / wrestling / takedown-defense — shown on profiles once the model has run.
Win probability (Compare)
On the compare page a matchup model turns two fighters' stats into a win probability — takedowns vs. takedown defense, KO threat vs. durability, striking differential, and more. It's a logistic model trained on real fight results with time-aware (leak-free) features, so the reported accuracy is honest — currently ~68% on held-out fights. Not betting advice.
What this is — and isn't
This is a scout-style read from public career numbers — a useful lens, not a perfect rating. Counting stats are descriptive and style can beat stats, so the composite is opinionated by design. Read the four axes together, lean on the opponent-adjusted view for context, and treat the win probability as a model's best guess — not a lock.