About the ratings

How the ratings are calculated

A rating represents a player's strength. We use standard Elo, calculated from every domestic official game. Games in international tournaments are not used for the ratings, because the opponents are not domestic players (they are still counted in a player's record).

Ratings are updated after every game, and the movement per game (the K value) is 32. Every player starts at 1500, and the calculation runs over games from 1960 onward.

Nihon Ki-in and Kansai Ki-in players are handled in a single rating list.

About playing Black

No adjustment is made for playing Black. In Go the advantage of playing first is already offset by komi. Measuring more than 50,000 games from 1960 onward, Black wins about 50.5% of games at the current komi of 6.5, and adding an adjustment did not improve the predictions.

In games with a komi of 5.5 or less, Black won 53–55% of the time, but separating out that period did not improve the predictions either.

How the K value and the Black adjustment were chosen

The K value and the Black adjustment were chosen by testing against past results.

We compared many settings on games from 1960 to 2002, and then measured how well each setting predicted the results of games from 2003 onward.

By measuring on a different period from the one used to choose the settings, we avoid fitting too closely to any one stretch of history.

We have not taken numbers from rating sites for other games. These values were measured on Go data.

Professional and women's tournament ratings

The professional rating and the rating calculated from women's tournaments alone are computed independently.

A player who appears in women's tournaments therefore has two ratings. Both are shown on their page.

This means that 1800 in the professional list and 1800 in the women's tournament list do not represent the same level of strength. The two ratings cannot be compared directly.

The period records cover

The records shown on player and comparison pages count official games from 1927 onward. Note that the further back you go, the patchier the surviving records become, and not every game played is included (About the data).

For this reason these records do not match the career records published by the Nihon Ki-in. The Ki-in's official tally does not include games by Kansai Ki-in players, while this site puts both Ki-in in a single rating list, so the two are counted differently to begin with.

For the same reason, a head-to-head record here counts only the games held on this site.

What counts towards a record, and what counts towards a rating

The record shown on a player's page and the games used for the rating are not quite the same.

For records we follow the way the Ki-in count:

• wins and losses by default are counted
• jigo and no-result games are not counted as wins or losses (but are counted as games)
• games in international tournaments are counted
• when both players lose, each is given one loss

Ratings, on the other hand, are calculated only from domestic official games that were actually played and decided, so a win or loss by default does not move a rating.

This is why the number of wins and losses does not match the number of games that moved a rating.

Changes to the method

Ratings are updated as new results come in, but the method itself is not changed often.

As a rule we check the current settings once a year against recent results, and change them only if the evidence says the predictions would improve.

If the method changes, it will be published on this page.

Current method: v1

Technical details

The rest of this page is for readers who want to check the finer assumptions.

The burn-in period

The calculation starts in 1960, but since everyone starts at 1500, ratings in the first few years do not yet reflect real strength. Records such as peak rating therefore use only values from 2006 onward.

Which games count towards a rating

• Counted: domestic official games that were actually played and decided (including replays)
• Not counted: jigo and no-result games (the replay is counted instead), wins and losses by default, games where both players lost, handicap games, international tournaments, unofficial events, and games that were cancelled or whose result is unknown

Oteai and team events are treated as unofficial and count towards neither ratings nor records.

Games where the colours were not published are still counted towards ratings.

Handicap games

A game with handicap stones is played with the difference in strength already offset by the stones, so it is not used for ratings. It is counted in a player's record.

Games against amateurs and foreign players

Amateurs who appear in official tournaments, and foreign players who do not belong to a Japanese Ki-in, do not hold individual ratings; they are pooled as “Amateurs”, “Korea”, “China” and “other countries”, because there are too few games to rate them individually.

Players belonging to the Nihon Ki-in or the Kansai Ki-in hold an individual rating whatever their country of origin.

When a player meets one of these pools, the player's own rating moves as usual. When two pools meet each other, neither rating moves.

What these ratings cannot tell you

• A rating is an estimate derived from results, not strength itself. The fewer games a player has, the more uncertain the estimate is
• Records from the more distant past are patchier, so ratings for older periods are less reliable than current ones
• Professional and women's tournament ratings are calculated independently and cannot be compared with each other
• A rating does not change while a player is not playing, however long the gap