The easiest mistake in this kind of check is to take today's ratings and apply them to old bouts. Do that and you are predicting with the answer already in hand, and the numbers can be made to look as good as you like.

What is used here is only the rating each fighter carried into that bout. A 2015 bout is judged on the 2015 values, a 2026 bout on the 2026 ones. The rating is updated one bout at a time, so what was known at each point is still there, exactly as it was. This is a record of going back, predicting, and then checking the answer.

The subject is every decided bout between 1980 and 2026: 65,639 of them. Draws and no-contests are excluded.

First, only bouts where both men had at least five prior fights (19,337 of them). The reason comes later.

The eight bands mostly run the way you would expect: the wider the gap, the more often the higher-rated fighter wins, and past 400 it is 82.2%. The first two bands, though, are the wrong way round (51.4% against 51.1%). The difference is 0.3 of a point, and 1,931 and 2,003 bouts cannot tell that apart. Below a gap of 50, read it as the rating barely predicting the result at all.

What should not be missed: even at a gap of 200 the favourite loses one time in three. On this scale 200 is roughly the distance between No. 10 and No. 90 on the active pound-for-pound list -- the top is denser, so 200 below the top is about No. 17. And still 33.3% of them turn over. That is the sport.

Drop the experience condition and look at all 65,639, and it changes.

150-200 sits below 100-150. It is not monotonic. And 400+ holds 32,366 bouts, half of the whole set.

The cause is debutants. The rating this site shows is a conservative one: the estimate minus how unsure of it we are. A fighter with no record at all is completely unknown, so a great deal is subtracted and he lands somewhere around 650 to 800. The result is that much of "a gap of 400 or more" is not a gap in ability but a gap in information -- an unknown taking on a veteran.

So this number is no evidence that the rating is good. It measures the obvious: a fighter nobody knows anything about usually loses. The earlier table, restricted to experienced fighters, is the one that says something about the rating itself.

Not just the ordering -- the numbers themselves can be checked. Did the bouts called 75% actually win 75% of the time? All 64,271 bouts are in what follows, with no filter on experience.

The 1,368 bouts in which the two ratings are exactly equal are left out of this table. Neither man is the favourite, so there is nothing to count as “the favourite won”. In 1,225 of them both were making their debut, the rating still sitting untouched at its starting value.

The table lines up reasonably well. The gap that stands out is at the end: bouts given 90% or better came in 4.4 points below. The three middle bands are within 1.9 points and the close band within 1.0, with no consistent direction. This rating overrates the favourite in lopsided bouts; elsewhere it carries no large bias.

Overall accuracy as a Brier score (lower is better; always saying 50% gives 0.25) is 0.1911. Clearly better than knowing nothing, and a long way from perfect.

The same check was run on GOAT Index Boxing, over 90,562 bouts (from 1904, bouts with exactly equal ratings removed). It is on the same basis as the MMA table.

In boxing too, the actual win rate is above the prediction in every band, by around nine points in the middle ones. 79.7% of decisive bouts have a rating gap of 400 or more, and most of those are against an opponent who appears in the record exactly once.

At a gap of 200 to 300 the higher-rated fighter wins 66.7% of the time in MMA and 70.7% in boxing (both with five or more prior bouts on each side, on the same basis). Boxing is the higher of the two, by 4.0 points. These numbers do not say why.

Restricted to bouts where both men had already built up five or more, the five boxing bands are out by at most 1.0 of a point (+0.7 / -0.2 / 0.0 / +0.4 / +1.0). The timidity was a property of the population, not of the rating. An opponent who appears once is given the starting value, which is the site treating him as an average fighter. He is not one, and that difference is what was lifting the all-bouts table.

Among bouts where both men had five or more prior fights, the seven with the largest gap where the lower-rated fighter won.

Stated plainly.

These bouts are the same ones the rating is computed from. Only information available before each bout is used, but the design of the model was settled while looking at the whole period. This is not a fully independent test.

The records come from what Wikipedia says. Errors in the articles come through unchanged.

The further back you go the thinner the data: the 1990s contribute only 2,369 bouts.

What was measured is ordering and calibration, and nothing else. The question of who is the best has not been answered here.

How the numbers are produced is set out in how the rankings are built. The hit rate of the fight predictions is published here.

[Correction] 20 September 2026 — the “50-60%” row of the calibration table was wrong. The 1,371 bouts in which the two ratings are exactly equal were being counted as bouts the favourite won, which pushed the actual win rate up. They have been excluded and the figure recomputed: 58.0% (+3.5) is corrected to 53.8% (-1.2). The surrounding text has been revised to match.

[Correction] 21 September 2026 — the boxing figures were wrong. Bouts in which the two ratings are exactly equal were being counted as bouts the favourite won, which put the actual win rate in the 50-60% band at 72.0% (it is 57.4%). This is the same mistake that was corrected on the MMA side first. The bout count, the starting year and the share of bouts with a gap of 400 or more have also been updated to the current data, and the 70-80% band, which was missing, has been added. The 200-300 comparison changes from “MMA 66.4%, boxing 66.8%” to “MMA 66.4%, boxing 76.4%”. This correction changes the conclusion.

[Correction] 23 September 2026 - the MMA table and the boxing table in this article were built on rating gaps measured on different bases (MMA on the value with the uncertainty subtracted, boxing on the value before subtracting it). The prose argued from the subtracted reading, so the tables and the text disagreed. The boxing table has been rebuilt on the same basis and the two-sport comparison rewritten. The "50-60%" row of the calibration table had its mean prediction computed on a different population from the rest of its own row, which is fixed; a table restricted to experienced fighters has been added; and the data is brought up to date.

The figures in this article are based on data as of 23 September 2026. The data is updated weekly, so the last digit may differ from the current tally.