Multiple mistakes in OGS Joseki position

According to Waltheri ( local search, the treatment of has three mistakes.

  1. R18 is marked as ideal. It has never been played professionally.

  2. Q18 is marked as ideal and has also never been played.

  3. Ironically Q17, which has been played 24 times (compared to 56 plays of R15), is only marked as good.

If Waltheri is missing new, relevant pro games then that’s one thing, but to add ideal-marked options onto OGS Joseki which have never even been played is misinformation.


There must be some confusion, because I see on Waltheri that R18 has been played 20 times in the database and that Q18 has been played 69 times. Indeed, none of your numbers seem to match what I’m seeing. R15 was played in 1208 games, not 56, and is definitely the main line of this joseki.

You’ve moved O16 to O17.


Ah, yes, that’s my mistake. I would say that the josekis are mostly analogous, and that if R18 and Q18 are joseki for the more popular low knight’s move, then it’s also joseki for the high one space jump. Indeed, I suspect that they are missing from the database because the high one space jump is more rare.

This is an example of my issue with joseki explorers, and why I usually advise people to use Waltheri instead. You’ve seamlessly shifted from the idea of providing a medium by which the student can access professional knowledge, to the idea of the players of OGS (all amateurs) dispensing their own judgements as fact. The latter initiative should be very clearly separate from the former.


I can see that perspective, but the issue is that there are far too many variations for each ideal variation to see extensive play by pros. If you don’t want to trust me, which is probably a wise decision, you can ask katago which is stronger than all pros.

All the options here are acceptable, R18 and Q18.

The explorer also listed the source (the position with one stone moved) and that listed as its source a video by a professional.


So KataGo should be the arbiter of what goes in OGS Joseki? I wholeheartedly disagree with that.

In that case, all OGS Joseki would be is an inferior fixed-state imitation of KataGo itself, which could be run on the student’s computer.

1 Like

No, no, no … you can’t just shift the position of a stone in joseki and say that it still works.

I don’t think it was ever the intention for amateurs to be dispensing their own judgements as fact. I think it was the aim all along to catalog professionally verified knowledge, and this may just be an honest clerical error.

What video are you referring to? I did not notice any link on the page. Are you looking at


You’re right, you can’t say that in general. As I have demonstrated, you can say that in this case because of the similarity of the positions. I don’t think katago should be the arbiter of what goes on the OGS joseki dictionary, but it seems reasonable to use it to compare similar shapes to see if something has been missed.

Edit: Indeed, it seems that all pros have moved beyond the need for joseki dictionaries and instead constantly develop new joseki with the help of the AI. It seems strange to link to a pro’s youtube video, mostly developed with the help of strong AI’s, then insist that your joseki explorer doesn’t use the AI as a source.

The high one space jump links to the knight’s move position, which lists a source.

1 Like

How did you demonstrate this? By using KataGo?

I’m sorry, but where is the link on the joseki page

I’m having trouble finding how that page links to any other joseki page, and still don’t know which source video you are referring to. Could you just post a link to the source video here?

In the same line as R18, links to another joseki position:, which links to A video based in part on study with AI.

And yes, I demonstrated it using Katago, which is actually a more authoritative reference than the video made by Yeonwoo.

This question of whether AI alone or AI in the hands of an amateur is sufficient to decide upon the joseki that goes into the dictionary has been discussed before. See Does the AI winrate simply tell us what is "Joseki"

I think there were some good points made in that thread, like here:


Surely when the OGS Joseki project was drawn up, there was a proper discussion held over what were to be deemed acceptable influences and a set of guidelines made? Or was no such discussion held?

I don’t think this particular thread is the place to be (re-)discussing it too deeply. This thread is primarily about the mistakes on this joseki page.

It seems like there were, and if it was held that only positions that show up on waltheri are acceptable, then you’re right that the R18 and Q18 should not be in the explorer. From what I’ve seen, it seems like the system is to use trusted players who cite sources. One of those trusted players, mark5000, contributed the joseki, and he linked it to a specific source. It seems like the contributor must have made some level of analogy between the two positions. I agree that the positions are similar enough to apply results from one to the other, and I am confident that nothing catastrophic was missed because I checked with katago.

I am not arguing that anything with a good winrate should be joseki. I am simply supporting the analogy that was made by the contributor of the joseki in question, which I don’t believe to be mistaken.

It’s also possible that a simple clerical error was made. Of course, we should ask @mark5000 to clarify and weigh in.

The cited source is talking about a different joseki, right? Assuming that the professional opinion about one joseki analogously holds for a different position (even if it is just shifting a stone by one space) seems risky and would root the genesis of the joseki in the judgement of the amateur rather the professional, even if it is later checked with KataGo.


There could be a mistake, but a link to a very similar position was purposely added. But I agree that this involves some level of judgement by amateurs, even if aided by computer. If your standard is that amateur judgment should never be involved, then you might as well just use waltheri’s or another similar database. I don’t think it is necessarily a good idea to rely on historical databases, especially in an era of rapid joseki change.

1 Like

I’d be happy for Mark5000 (because I’d isolate him as the “guy most clearly in charge of OGS Joseki”) to make a thread tomorrow or sometime soon to discuss this properly, under his “aegis”, in a format where he can – if he wants – open with an introduction about previous debates and his own thoughts about where the project its going and its function. I reiterate, I don’t think this is the place for it. And if he doesn’t want to make such a thread but simply state the guidelines in here, that’s also fine by me.

And yeah, I agree that you may as well use Waltheri. I was never a big supporter of the OGS Joseki project precisely because Waltheri was already available; but since I know that people are using the tool, I like to keep a handle on what’s on it – so that, at the least, it doesn’t become “Josekipedia II” with all the associated issues of bloat and misleading content.

I think you might be misinterpreting the purpose of OJE here. I would agree that josekipedia is far more likely to have users submit their ideas of ideal moves as fact. In the OGS joseki explorer, a large number of joseki variations come from pro sources and are cited. At least this is one of the goals, to cite books and more recently lectures/youtube videos from professional players.

On the other hand, I believe anything marked as source ‘Traditional’ should probably be eventually changed to some actual source, book video etc. I think this might’ve just been the original source when getting the functionality of OJE working.

I think the objections you give about Katago can more or less be translated to your use of Waltheri to define ‘mistakes’ in joseki. If you disagree with having a pro strength bot decide what is joseki, then similarly just using a local search feature on a website with a limited pro database also should be objectionable.

Appearing in a local search also doesn’t really define a joseki in and of itself. ‘Real’ josekis probably should appear in thousands of games but then again those are really the popular ones that get played regularly at pro level. However ‘real’ joseki should also give an even result in the corner independent ignoring the rest of the board. In real games you can’t actually ignore the rest of the board, so even the fuseki choices will affect the frequency with which joseki appear in a local search*. This is much the same with pro strength bots, they’re always considering the whole board position when playing out local sequences leading to many early tenukis which can give a slight local loss for instance.

*(Imagine a move appearing, but in game it’s played at move 150 rather than move 5/6 - does that make it joseki or are there more stones all over the board that influence whether this move is playable)

Back on topic though, this could be true and requires attention I believe. Definitely find a source for it and maybe change it to ‘good’ depending. I think though if there is a source saying ‘This is/was joseki’ from a professional player(s), even if it hasn’t been played in waltheris database, I don’t think that is reason to say it’s not joseki in and of itself.

At least this one cites AlphaGo teach. It does make me think there’s probably an AlphaGo vs AlphaGo game where it could be played, and possibly a video by Michael Redmond discussing it :slight_smile: (I’m also not sure how to find it in alphago teach)

This could be a fair point. I think @Eugene can also help answer what the idea behind ‘Ideal’, ‘Good’ etc mean and should mean.


I think that these words are still holding up to the test of time:


  • “Ideal” means “we consider this to be Joseki, in the full sense of our (OGS OJE) definition of that”.’
  • “Good” means "not Joseki, as defined, for many reasons, but are arguably worth including - common “situational sequences” are an example.


How was r18 never played? What am I missing?

Were you talking about an earlier move? Has OJE been changed in the meantime? This was the only time r18 was a potential play in OJE when I checked just now.