Simply put, they were added at first to reduce the number of matches required to determine a player’s strength (the original difference between 2 dan is 1 stone, then how to have even games for players with just 1 dan strength difference). The halfway point of the original thought of 2 dan different of 5 points (because from historical records till that point, even players seem to end up with lots of game with 5 points difference), 2.5 points is chosen to be “fair” komi for the 1-dan handicap. Even results were what they expect (play two games with this set komi of a 1-dan strength difference players, they would come out 50-50 (win one and lose one, instead of the older system require 3 or more games to determine the results)
However, Over time, it is then used to determine strength difference with just 1 game to speed up more and more competitions where single elimination can be applied.
I would be interested in seeing your evidence for this. Could be well be true, but I have been playing Go for more than 50 years, in the West and in Japan, and never encountered 0.5 komi for handicap games until Go servers arrived. Professionals in simuls would sometimes engineer a jigo if they wanted to be extra friendly or polite. I never played in a handicap tournament in Japan, where since they didn’t use Swiss tournaments there may well have been the rule that White advanced to the next round in case of a jigo, which is the same as using 0.5.
I wonder if it has something to do with the Point Rating System (a very fine grain ranking system for amateur players so to speak, and a rough ELO rating), where it already existed since 昭和５０年頃 (1975~1984), for amateur players to play fair games in Go clubs.
And from what I found from the Japanese wiki, the original table actually starts with 5 points (5目までのコミ出し), if two players have the same point(rank), and 12 points difference is a stone difference, matching the professional game komi at the time of 5.5 (the wording is 5 points komi and white wins if it’s jigo).
Also, I found Ing’s rules book since the 1980s, already stated 16 points komi equal to 1 stone difference (when jigo, black wins, hence an effective 7.5 komi). And Ing’s Foundation held many international games and tournaments all across the world and invited many programmers, scientists to participate in its own computer Go tournament since 1985. And the first Go server IGS came online in 1992. And the point ranking system got embedded into the online go servers ever since. Maybe the programmers of these servers heard of or even participated in Ing’s tournaments. hence adapted the point ranking system. And IGS-pandanet still uses a very similar point ranking system with the same ranking table as Point Ranking Scheme At Tokyo Go clubs at Sensei's Library. I don’t think this is a coincidence.
Could also be that my memory played me tricks.
You wrote: “internet Go servers unnecessarily introduced the idea of 0.5 komi for handicap games.” (unnecessarily is of course a matter of opinion).
In the December 1986 issue of the Dutch Go Magazine (Tijdschrift van de Nederlandse Go Bond) there is game with 5.5 komi for White. I just randomly picked an old issue (still enjoying the comfort of a warm bed in a dark world) so there may be even older 0,5 komi games, which I will check later
I don’t know when the first internet go server started. In the Netherlands internet was not a widespread phenomenon in 1986.
I joined the Nijmeegse Go Club in 1988. I am fairly sure that we already used a handicap system (which was different from the official one the Dutch Go Federation used): https://gobond.nl/clubsites/nijm/voorgift_tabel.html
This I picked up from the Dutch beginners book (translation from Japanese/Chinese? of a famous professional go player, don’t remember his name).
EDIT: the first go sever (IGS) started in 1992.
EDIT 2:
On Wikipedia NL it is stated that “Sinds 1970 speelt men daarom vrijwel altijd met komi.” (Since 1970 go is in almost all cases played with a komi.) Unfortunately there was not a source mentioned.
In the February 1973 issue of the Dutch Go Magazine a professional game is reviewed. This is a game between Takagawa (Meijin) and Kato (white, 4.5 komi, 5 dan) played on February 26-27 in 1969 in Tokyo. This is the oldest komi game I found in my Dutch Go Magazine collection (which is incomplete before 1974). There might be older komi games.
I do realise that you wrote “never encountered 0.5 komi for handicap games until Go servers arrived.”, but assume that you meant komi in general and that handicap games or even ranked games is not relevant here.
Thanks for giving me an excuse to dive into the history of go
Love to do that.
The time when the komi system was introduced for pro players in Japan is very clear and have records
1934年、全日本囲棋選手権大会において、初めてコミが導入されている。
In 1934, the komi system was introduced in “All Japanese (professional) Go Players Championship Tournaments” with the earliest komi system on record I mentioned above Why no Jigo? - #4 (half of the 5 points of 2 dan strength difference for 1 dan difference player)
1939年に始まった本因坊戦においては、段位を問わない総互先方式、コミは四目半と定められた。
In 1939, from the start of the first Honinbo title match, it set the precedent of switching from the old teai system match to a 4.5 komi match format.
まずは1964年ごろから、コミを五目半とする棋戦が現れた。しかしながらこうしたコミの改訂は、全棋戦で一斉に行われたわけではない。
In 1964, Nihokiin start to promote 5.5 komi to pro games, but it took about 10 years to gradually change all pro matches from 4.5 komi to 5.5 komi
Although the records for amateur players how they set up the handicap and various komi is less clear (instead of a specific integer komi and white/black wins as official pro games indicate, although more precisely white got a 0 point win of a lesser degree and we have title match results in that cause pro players to be upset). As I mentioned above, I can only found sources that Point Rating System existed in Japan since around 1975 for go clubs. Not too sure about the situations before that. Some no-komi 10 games match using the old system were still being played in the 1950s. And the Oteai system existed till the early 21th century.
No I was referring specifically to handicap games with a komi of 0.5. I knew of course that in even games komi with values of 4.5, 5.5 etc have been in use for a very long time - Clare’s researches gives us dates.
Fair komi on 9x9 according to KataGo was also part of the L19 topic that @hexahedron linked to a couple of days ago.
With integer komi on 9x9, 6 points is fairest under Japanese rules (48% winrate for black) and 7 points is fairest under Chinese rules (42% winrate for black).
With half-integer komi on 9x9, 5.5 points is fairest under Japanese rules (70% winrate for black) and 5.5/6.5 points in fairest under Chinese rules (both 70% winrate for black).
Sorry Ive not read the linked topic but wow, this seems interesting. In that Komi of 6.5 under Japanese rules and 9x9 gives black a win rate of less than 30%. Quite a drop for a single point. Wouldn’t be so surprising I suppose but my conclusion is that 9x9 can never really approach a “fair” game.
Well, I think the main difference is that AI are closer to perfect play on 9x9 than on 19x19.
As skill becomes higher, the drop from a single point komi becomes sharper.
A perfect player would give 0% or 100% winrate for a game with unfair komi.
KataGo is not perfect yet, but from KataGo’s evaluation, I would conclude that 6 is fair komi under Japanese rules and 7 is fair komi under Chinese rules on odd board sizes from 9x9 to 19x19.
As AI get stronger, they may find that jigo is the expected result in those cases. The jigo may be harder to hold for one color, causing current AI to be slightly off 50% with fair komi. But in the limit of perfect play, I expect it to come to approach 50%.
Just re-upping this as it is still an issue and the previous thread was closed.
It came up in this game in the BGA congress small board tournament with AGA rules. IMO the result does matter and the voice message should not tell one of the players “you have won”.
KataGo thinks for odd board sizes between 9 and 19, 6-7 komi is fair for area scoring (such as NZ rules) and 6 komi is fair for territory scoring (such as Japanese rules).
I think this bug is actually really important, even if jigo is rare and can only occur under certain rulesets (so is double rare). For example, I quite like the idea of playing with integer komi so that a draw is possible, as that could be considered an especially nice outcome in an even game. However, with this bug still there, it undermines integer komi such that I don’t see the point in trying rulesets with integer komi.
And as the bug is related to scoring, which is so fundamental to the game, it kind of undermines confidence in the whole site. Like these bugs about sounds going off at the wrong time or something, one can kind of accept as “Yeah, ok, tricky browser compatibility issue or something. It’s kind of annoying but I can live with it.” But if the result of a game is wrong, even if it is an edge case, then it makes you wonder what other crucial aspects of the game have bugs in.
Incidentally, one of the original arguments against fixing this bug was this:
So custom komi is only possible in unranked games? I think this should be changed. For any game, custom or not, you can work out the probability of who should win based on rank, handicap, komi, etc. and then adjust ranks based on the actual vs expected outcome. Custom komi is basically adjusting the handicap by some fraction of a stone, so that should not exclude such games from being ranked. Clearly, this is a somewhat separate issue but posting here as I thought of it when reading this thread.
From the api and the game before it for https://online-go.com/player/742504 with termination api ratings v5.
ended
game_id
played_black
handicap
rating
deviation
volatility
opponent_id
opponent_rating
opponent_deviation
outcome
extra
annulled
result
1611135704
30286104
1
0
1841.90
61.55
0.060019
746359
1805.88
64.09
1
null
0
0 points
1610527607
30064526
1
0
1832.00
61.62
0.060020
745737
1740.02
61.93
1
null
0
Resignation
while for the opponent
ended
game_id
played_black
handicap
rating
deviation
volatility
opponent_id
opponent_rating
opponent_deviation
outcome
extra
annulled
result
1611135704
30286104
0
0
1795.20
63.87
0.059958
742504
1832.00
61.62
0
null
0
0 points
1608713795
29418232
0
0
1805.88
64.09
0.059958
899832
1692.64
105.40
0
null
0
19.5 points
So the higher rated player (Black) won points (1832.00->1841.90) and the lower rated lost points (1805.88 ->1795.20), while the game is recorded as a win (1) for Black and a loss (0) for white.
I don’t know why I thought it was only a display issue…
@anoek It would be really nice if jigo didn’t get scored as B+0 I called a moderator during the scoring phase to manually score a draw, but the report didnt get handled in time and black “won” our drawn game.
(I dont care about this result in particular getting changed, but the bug in general still persisting after all these years)
As far as I am aware There is literally no such outcome as “Draw” or “Jigo” on OGS.
EDIT: But: I was wrong, see later posts
That’s what it says earlier in this thread.
It’s not a “bug” per-se, it’s a “feature that we don’t have”.
If I recall correctly, the reason why it’s not an “easy fix” is because all sorts of things (tournaments, ladders etc) would suddenly need to understand that there’s a second “kind” of outcome - a way that games can legitimately be scored, that does not have a “winner”.
“Revamping the whole tournament system” is definitely on “the list”, although it keeps getting bumped by other things. Possibly this could be looked at as part of thing, but in the mean time - for the foreseeable future - I don’t see this changing.
But wait…
Curiously, @benjito 's observation that Mods can end games as “Tie” is true - so maybe I’m simply mistaken. Someone would need to dig into “so what happens in that case”, and see where it leads.
I called a moderator during the scoring phase to manually score a draw, but the report didnt get handled in time and black “won” our drawn game.