How the rating is calculated

Every athlete has a rating, recalculated after each tournament.

01Principle

A win raises the rating, a loss lowers it. Beating a higher-rated opponent earns more than beating a lower-rated one.

The calculation uses Glicko-2, a variant of the chess Elo system, which gives every rating a margin of error: large at first, it shrinks with every match. On the site, the rating is called “Elo”, and the margin is shown as reliability, from 0 to 100%: the smaller the margin, the higher the reliability.

The same match, 1950 vs 1750
The favourite wins (74 %)+7−18The underdog wins (26 %)−20+57

The upset is worth 8 times more than the expected result.

02Match simulator

Athlete A

Athlete B

Chance of winning
74 %26 %
A winsA +7B −18
B winsA −20B +57

With equal margins, a 200-point gap gives the higher-rated athlete about a 75% chance of winning. The larger the margin (lower reliability), the more the rating moves after a match.

03Starting rating

A new competitor starts at their belt's starting rating with a margin of error of ± 250 (reliability 46%). The gaps between belts were measured on thousands of matches between different belts (open class, mixed divisions): from one belt to the next, the higher belt wins about 6 times out of 10. A competitor whose belt is never known starts from the level of their divisions when they have one (beginner: white, intermediate: blue, advanced: purple), otherwise slightly above a white belt.

Black belt1975± 250
Brown belt1875± 250
Purple belt1800± 250
Blue belt1725± 250
White belt1600± 250

Promotion

On promotion, the rating gains 50 points and the margin of error goes back up to at least ± 100 (reliability 88% at most): the first matches at the new belt place it quickly. There is no forced jump to the new belt's starting rating: a recent promotee isn't yet at their new belt's average level.

BeforeBlue belt1855± 90
After promotion to purple beltPurple belt1905± 100

04Example season

Rating of a fictional blue belt over six tournaments. Clicking a match flips its result and recalculates the rating.

140015001600170018001900200021008 months without competing
margin of errorranked (reliability 70%, 6 matches)not ranked
StartBlue belt1725± 250
Open de ParisFeb 2023
1744 +19± 179
Coupe Île-de-FranceApr 2023
1810 +67± 137
Open de LyonJun 2023
1823 +13± 122
Open de FranceFeb 2024
1899 +76± 116
Grand Prix de NiceApr 2024
1922 +23± 103
Championnat de FranceJun 2024
1926 +5± 97

Ranked from: Open de Lyon.

Matches from the same tournament are calculated together, from the ratings before the tournament. Without matches, the margin of error slowly grows and reliability drops.

05Matches taken into account

Win method

A win by submission, points or advantages has the same effect on the rating.

Walkovers and disqualifications

Walkovers are not counted. Disqualifications count as losses.

Open class

Absolute matches count like any other match.

Gi, No-Gi and combined

Three ratings are calculated separately: Gi, No-Gi, and a combined rating that includes all matches.

Age groups

Only adult and master matches are used. Kids and juvenile matches are not collected.

CFJJB results

CFJJB publishes brackets and podiums but not individual matches. Matches are reconstructed from this information.

06Conditions for being ranked

Everyone has a rating from their first match. To appear in a category's rankings (Gi, No-Gi or combined), all three are needed:

70%reliability at leastit grows with every match
6matches at leastin that category
2 yearssince the last match, at mostwithout matches, reliability slowly drops

For black belts, the world rankings also require a connection to the international circuit (IBJJF, AJP, UAEJJF), Gi and No-Gi combined: 3 international matches, or 3 different opponents with at least 5. A closed circuit inflates its regulars' ratings, for lack of a comparison with the rest of the world. Country rankings don't have this condition.

07How accurate the ratings are

To check the ratings, we replay every match of the last 12 months knowing only what was known before it, and compare the prediction with the result.

73 %of matches correctly predicted between established competitorsat least 10 matches at their belt and a reliable rating each
63.5 %across all matchesmany of them involve a newcomer we know little about yet
546,873matches checkedsince November 11, 2025

Announced chances come true

When a favourite is given 70%, they should win about 7 times out of 10. Each line compares the chances announced before the match (shaded area) with the favourite’s actual share of wins.

Announced 50–60%actual 54%149,743 matches
Announced 60–70%actual 62.4%142,806 matches
Announced 70–80%actual 72.5%100,105 matches
Announced 80–90%actual 84.2%56,823 matches
Announced 90–100%actual 94.2%18,121 matches

Jiu-jitsu keeps a share of the unexpected: a mistake, a surprise submission, a referee call. Even well tuned, a rating can’t call every match; it tells who is more likely to win, and those chances are accurate.

Measured in the November 11, 2026 computation, recomputed every night.

08Club rankings

A club's score is the average rating of its 10 best ranked athletes in the chosen category (Gi, No-Gi or combined). A club needs at least 5 ranked athletes to appear. Athletes count for the club of their latest competition, and only ranked athletes count: an uncertain rating can't lift a club.

09Sources

Results come from IBJJF, AJP and UAEJJF (Abu Dhabi), Smoothcomp and CFJJB publications. When an athlete appears more than once, their profiles are combined if the name and club match, if the profile photo is the same, or if the name is identical, rare and from the same country.

IBJJF removes brackets from older tournaments, so some older matches may be missing. All ratings, including past ones, are recalculated whenever data is added.

An error in the results, a missing event?

Legal notice · Privacy