There is a free public record of what every UFC fighter actually does in the cage: how many significant strikes they land a minute, how many they take, how often they finish a takedown and how often they stop one.
It lives at UFCStats, the promotion’s own database, and we pulled 292 fighters out of it: significant strikes landed and absorbed per minute, takedowns per fifteen minutes, takedown defence, submission attempts. Our engine carries its own numbers for the same fighters — hand-written, argued over, and used by every game on this site. The obvious question was whether the two lists agree.
Mostly they do not, and we think that is worth explaining rather than quietly fixing.
What we compared, and how
Comparing values directly would have been meaningless. Ours sit on a 0-to-99 game scale where the low 90s is elite; theirs are percentages and per-minute rates. So we compared the ordering instead: if our ratings are picking up something real, the fighters we rank highest for takedown defence should broadly be the ones with the best measured takedown defence.
Rank correlation, where 1.0 is perfect agreement and 0 is none:
- Takedowns — 0.42
- Striking defence — 0.35
- Takedown defence — 0.27
- Striking accuracy — 0.14
The first is a moderate relationship. The last is barely a relationship at all — our accuracy rating and real striking accuracy are close to two unrelated lists.
Some of the gap is our fault
Curtis Blaydes is the one that stung. We had his takedown defence in the top tenth of the roster; his measured figure is 35%, near the bottom. Part of that is the sample problem that makes defensive percentages unreliable for wrestlers who are always the one shooting — but not all of it. We had quietly conflated "excellent wrestler" with "hard to take down", and those are different skills. Aljamain Sterling shows the same shape.
Reach was worse, and much simpler, because reach is not a judgement call. We were carrying 218 wrong figures, several off by seven inches or more. Paulo Costa was listed at 82 inches against a real 72. Israel Adesanya at 74 against a real 80. Those are just errors, and they are fixed.
So are the records. Of 304 fighters, 248 had no win-loss on file at all, and of the 56 that did, 40 were wrong — Andre Fili was carrying 11-2 against an actual 25-14. Every fighter on the roster now has a verified record, and the AI matchup analysis on the homepage reads it.
Some of the gap is not
Striking accuracy is the clearest case for leaving a rating exactly where it is. Cruz records 33% because his entire game is feints, angles and volume from distance — a style that misses more by design and wins anyway. A cautious pressure fighter who throws only when certain will out-measure him on that column and be the lesser striker by a distance. Rebuilding our accuracy rating to chase the real numbers would encode "throws less" as "better", and every game on this site would get worse for it.
The same goes for chin, cardio and fight IQ, which are not counted anywhere. There is no statistic for a chin. There is a statistic for how often somebody gets hit, which is a different thing wearing similar clothes.
What changed and what did not
We fixed the facts — reach, records, and the handful of ratings that encoded a plain misunderstanding rather than a stylistic judgement. We did not rescale the attribute ratings to chase the correlations, because those ratings exist to make a fight resolve believably when you simulate it, not to predict a rate statistic. A game in which Dominick Cruz is a poor striker would agree better with one column of a spreadsheet and be less true to the sport.
Where the measured data is straightforwardly better, we now use it directly instead of approximating it. Five question types in Higher or Lower are built on real career figures, and those rounds reveal the actual number rather than a letter grade — because something that was counted can be shown as fact, and our own rating cannot.
Why publish the bad result
The footer of every page on this site says the ratings are our own model rather than official statistics. That line is worth something only if we are willing to say what it means in practice: that a lot of them do not line up with the counted record, that we now know which ones and by how much, and that in several cases we looked at the gap and decided the rating was still the more useful number for a game.
If you disagree about a particular fighter, you are probably right about at least one of them. The simulator is free and the roster is real — go and find the one we got wrong.
Sources & method
The measured figures in this article were taken from ufcstats.com, the UFC’s official public fight database, on 28 August 2026, covering 292 fighters on our roster; fighters with too few UFC appearances for a stable per-minute rate were excluded. Agreement was measured as Spearman rank correlation between our rating for an attribute and the corresponding UFCStats figure, across all fighters holding both. Fight records and reach were checked against UFCStats and Wikipedia and corrected where they disagreed with our file. The attribute ratings themselves are our own model and are not published by, derived from, or endorsed by the UFC.