How we rank
Every number on the rankings comes from here. If you think a weighting is wrong, this is the page to argue with.
The rankings
What goes in. Every documented bout we hold a source for: 21,196 of them since 1997, involving 4,008 wrestlers, from tournament brackets, federation and organiser results (6,231 of them from Sumo Cup, the European Sumo Federation's results service), the pro leagues' own bouts and bouts on video. Age-group bouts stay out; a bout inside a team match counts like any other. Only the wrestlers listed here are published, but everyone they have wrestled is rated too, so beating a strong man counts for more than beating a first-timer.
How it moves. Bout by bout, in date order. The winner takes points off the loser, more the more surprising the result. Each wrestler also carries an uncertainty: wide at a first bout, narrower with every bout after, wider again the longer a wrestler is out of action. The less certain a rating, the further one bout moves it.
Where a newcomer starts. At the average of the established wrestlers from the same country, less 150 points. A first-timer from a strong sumo country is not treated as a novice, and the men who keep competing are better than a typical first-timer, which starting everyone at the average ignored.
Size, where it applies. In an openweight bout the man from the heavier weight class gets 40 points for each class between them. In a weight class both men made the same weight, and nothing is added.
Who gets a number. Only wrestlers the rating can actually place. We tested where that stops: past an uncertainty of about ±350 a rating no longer predicts that wrestler's bouts reliably, and it runs too high. Most wrestlers are inside that line within about five bouts. Everyone else is still on the list, marked P and unnumbered, at the place our best estimate gives them: results so far weighed against published credentials, each by how far it can be trusted, or credentials alone for a wrestler with no documented bout. A wrestler with no credentials on record is placed by results alone: a record we have not found is not a weak one. The range shown with a P covers where results, credentials and the estimate each put them.
How often it is right. Every setting was chosen on bouts before 2024, then tested on the 6,317 men's bouts since, each predicted from only what was known before it: the favourite won 73.2%. On the 262 documented bouts between two men on the pre-season roster (the men the seed ratings covered), predicted the same way, it picked 68.9%, where the resume-based seed ratings this site published before the season picked 67.6%. The chances it states came closer to what happened (a log-loss of 0.591 against the seed ratings' 0.633; lower is better, and a coin scores 0.693), but on 262 bouts that gap is within the noise.
Only gaps mean anything. The numbers are centred so the middle of the list reads 1500. Rated 100 above an opponent, a wrestler is expected to win about 64% of the time, less when either rating is uncertain.
Credentials, where results are thin
A provisional wrestler's approximate place weighs results so far against published credentials. A credential scores by the level of the event (world or international, continental, national, regional) times the placement (gold, silver, bronze, finalist, took part), the best single result plus a little for depth. A result from another sport counts half; a title won in a light weight class counts for less, because the pro leagues have no weight classes. Credentials only fill gaps: where there is a real record, results decide.
Sources
A result counts when we can point you at it: a league's own pages, an official federation archive or results service, a published bracket or results sheet, or a recording that shows the bout. What an athlete reports to us is listed on the profile, marked, and left out until we find it in an archive. A few meets published nothing at all; where more than one competitor who was there agrees on the result, it can count, and we write up what was and was not established.
Age groups are left out: a bout in an under-18 division never counts (under-21 and under-23 divisions are adults, and count). A minor who enters an adult division has those bouts counted, but they are never shown here. Team matches are one wrestler against another and count like any other bout; a team medal is won by a squad and is never used as one wrestler's result.
Promotions that publish no results. World Elite Sumo puts out no results of its own, so its nights are rebuilt from the broadcast, uploads of it and press reports, and every bout on those pages links the moment or article it comes from. Where no source announces a result but the finishing order leaves only one possibility, the page says the bracket forced it. We never collect anything from a promotion's own site by script.
What we tested and threw out
Pulling idle wrestlers back toward the average made predictions worse. Widening their uncertainty while they are away is what works, and it is what the ± shows.
Mixing credentials into a real record added nothing at best, and at the weight a straight fit suggested it made predictions worse, so credentials only place the provisional.
A plain Elo rating tuned the same way came close but lost on the test years, and blending the two added nothing.
Counting bouts rebuilt from pool tables at half weight changed nothing, so they count in full.
Limits: we see only what is documented. A wrestler's record here can be shorter than the real one, which is what the ± and the provisional places are for. Corrections move ratings the day they land, and changes to how the ratings are weighted happen between events, never during one.