Don’t forget 9 komi on 7x7! The default 5.5 really makes a difference there.
2 Likes
It does make a difference when you have seki, Seki is real. If you have area scoring and an odd number of neutral points due to Seki, you could come up with a game where black has 43 points and white has 37 points. If komi is 5.5 then white loses a game that he/she should have won.
Please be careful about statements like this because it seems people here are really easily misdirected. Lets work towards the common good here.
2 Likes
5.1 komi is no different from 6.9 if there is no seki. So if a tiebreaker is desired then 5.5 is okay
As you mention, seki is very real, not a theoretical thing only. If there is an odd number of neutral points due to seki and black has 43 points and white 37, 5.1 komi would mean white losing a game that he should have won and 6.9 would mean white winning a game he should have won.
Due to the above, 6.5 and 7.5 are the only acceptable tiebreakers (the former favoring black slightly and the latter favoring white slightly).
In a perfect world, komi for the three ranked board sizes would be 7.0 …
4 Likes
What do you think is the fair komi on 5x5? It’s about a quarter of 9x9.
3 Likes
I’ve merged the two threads discussing komi on smaller boards, in particular for 9x9.
@terrific apologies I seem to have missed your post when moving them and now it’s (slightly) out of order 
1 Like
Remember Koreans playing 9x9 with big stones? I remember. Uberdude obviously remembers. Who else?
Relevant because komi wasn’t the same.
Apologies for necroing this thread, but has anything been decided in the end?
This seems to be a pretty consensual matter with AI providing a clear answer.
4 Likes
anoek is agreeable to change komi, but wants to change handicap at the same time, and wants to have enough time blocked out to do that properly.
1 Like
I’ve heard that before and I don’t understand.
- The question of komi on smaller boards has been answered years ago.
- The question of correct handicap on smaller boards is not settled and probably won’t be for at least several more years.
I applaud the desire to improve the handicap system, but it’s mostly unrelated to just changing the komi from 5.5 to 6.5 (Japanese) / 7.5 (Chinese).
Why keep using the wrong komi?
As long as the komi is wrong, the games are unbalanced.
Furthermore, the AI analysis provided at the end of 9x9 games is mostly useless because of the wrong komi. When I play a 9x9 game with White and I lose, the AI graph is a flat line that says “Black was already winning at move 0, no move in the game changed anything”. It’s disheartening.
5 Likes
There are people who use win percentage not score? 
fascinating…
It is kind of funny though that some people, not necessarily you above, will talk about how it can be hard to understand why AI says X or Y, why AI scores a position as B+(score) and how that score might only be true if you can play precisely like the AI, while on the other hand others might be willing to accept that katago’s estimate of fair komi is perfectly accurate and doesn’t require one to play perfectly like an AI also.
Maybe 5.5 is fine for a large number of players, maybe it’s even too large for others.
I personally don’t really care what the komi is - so the above is not a comment coming from an OGS team member or anything, just from me personally - so whether we set it to 5.5, 6.5, 7.5 it probably won’t impact the 9x9 games I play from time to time noticeably.
I think if you want the official stance then it’s as @BHydden says
and probably wants to consider some actual data and analysis also from games on the site.
1 Like
But unless perfect play leads to a draw or a no result or something, this is just how it should be if you played well enough for example.
It will just be a flat line for two players playing perfectly and who is the winner won’t change, and it’ll be set by the komi.
So maybe you adjust the komi a bit, and now it instead says
“White was already winning at move 0, no move in the game changed anything”.
and it’s disheartening for your opponent instead.
It’s not to say you and your opponent were definitely playing perfectly for instance, but maybe you could pick different or more complicated or less explored openings, ones your opponents aren’t familiar with and the score/winrate won’t be a flatline.
But there’s a huge difference between “assuming perfect play, one player would win by 0.5” and “assuming perfect play, one player would win by 2.5”.
2 points is a lot on 9x9.
In many situations on this small board, we have to choose between a “pacific” or “submissive” strategy, and a more aggressive strategy. With a balanced komi, both players get to make this choice. But the too-small komi on OGS gives Black the option to always play “pacifically”, not offering any opportunity for White to overturn the game.
On GoQuest I have 54% wins in games I am White, and 62% wins in games I am Black. The games always feel balanced. If I win it is because I played better than my opponent. If I lose it is because my opponent played better than me.
On OGS I don’t know how to get the statistics, but my feeling is that when I have White on 9x9, I only win against weaker players; against a player of my own rank or stronger, I simply never win with White. And then the AI analysis simply tells me “Of course you lost, you were White with 5.5 komi, there was nothing you could have done”.
2 Likes
One place to get some stats would be to use
https://avavt.github.io/gotstats/#/
but I think because OGS is struggling a bit, the api calls that that github page is using also aren’t loading.
When it’s back up though, I think I remember it being able to tell you how often you win as black and white on different board sizes and possibly by how much.
This would be the thread anyway for further suggestions and things for that site/github page.
Once again into the breach….
As I understand it, komi is based on pro games. The idea that this standard is proper for players at all other levels is unproven so far as I know. To believe it is the same would require us to believe that kyus can exploit the advantage of the first move as well as pros, which strikes me as absurd on its face. There could be a gradient of value through the ranks or there could be a stepwise change, where players at some unknown rank have a lightbulb moment that allows them to make good use of the first move—or it could be anything in-between.
1 Like
I think the reason it’s tied tightly to handicap is that right now the rating system it’s very strict about only ranking games with the “right” settings. But once we increase the komi, we’ll still want to be able to retroactively rank all those past games with the “wrong” komi. Or do they all become unranked?
So maybe it should know the value of different komi settings and rank those past games with a slight advantage to black. But knowing the value of komi is almost the same as knowing the proper handicap settings, and then you might as well make an attempt at getting everything right at once so you don’t have to keep tweaking it.
Personally I think we should allow any handicap and komi (and board size!) to be ranked and just get comfortable with the idea that we’ll have to keep tweaking the system and retroactively changing ranks over time.
1 Like
For players that cannot capitalize on the first move advantage (i.e. players that play essentially randomly in the opening) the games are generally not decided by komi anyways. The relatively small amount of games that are won because of komi should be roughly balanced by the amount of games where a player inadvertently uses the first move advantage.
I expect the perfect komi to be more or less stable throughout the ranks, but there is an easy way to check if we have access to the data. We can just check if the game results are biased in favor of one color or the other at a given rank.
Edit:
I’ve checked my own games and the results might be interesting, even if it is very little data. I do in fact win 55% of my games with white and lose 55% of my games as black on the 19x19 board, so maybe komi is too high at my level of play (about 9 kyu). Roughly 10% have been decided by komi. Also interesting is that I have the inverse effect on the smaller boards, suggesting that komi is indeed too low for 9x9 and 13x13, even at my level.
That is a reasonable opinion, which has been stated many times on OGS in the past, but it misses my main point: it is unproven, and at this point is only an opinion. Earlier parts of this discussion questioned why anoek needs to examine the issue. The answer is that statistical analysis of the game data is needed, and that takes time.
Edit: My comment above was before LKSFRZ edited his post.