Descriptive Outcomes

So, Mark’s thread kinda forced me to post this…

There’s been some discussion on how to emulate Josekipedia’s “good sequence” “ideal sequence” and whatnot… and I just think that’s all misguided. Here’s my 2 cents, semi-copypasted from my longwinded discussion with Eugene on the topic of keywords/outcome summaries. Slightly edited. Enjoy.

Could you suggest what the actual list of outcome keywords might sensibly be?

Sure. When we think about keywords in this context, we can see that they belong to two categories.

The kind that becomes more useful and the kind that becomes less useful. The appreciating keywords belong to a class that improves with the amount of data we gather. They are merely descriptive. [edit: e.g.: “R17: 33% of all recorded variations (N=17645) lead to B getting the corner & top”]

The depreciating keywords can only improve with better understanding/analysis - they are inferential (“good for [color]”). I see lots of inferential keywords on josekipedia, but now that I’ve given this idea some thought, I’m not so sure we should adopt that. It would be impossible to redact any changes in inferential data as people’s (or bots’) analysis of positions advances. It is trivial to present us with descriptives of what’s in the database.

I think the most important summary would ideally be of the format

results in [color] getting {corner, outside} with access to {top, center, right}

That would give us a 5-factor descriptor (2x2x2x2x2) which could inform our play. If black wants to maximize the probability of getting the right side while retaining some access to the center, we should pick subtree x.

If we don’t mind giving up both sides adjacent to the corner as well as center access, say because we trade it for some cash, we choose subtree y.

When we tenuki from a standard knight’s move approach to 4-4, we might plan to give up both sides and the corner in order to retain center access.

If I had to pick a second keyword, it would be

requires [stone somewhere] or [ladder maker/breaker]
mostly because it would be cool to be able to search specifically for, say, ladder variations. Non-essential.

Another benefit of this approach would be that it’s very simple to see why the descriptor is accurate (for the data we have!!! As I said, the value of this descriptor depends almost entirely on the amount of data we can gather), whereas it can be impossible for even pros to explain why a sequence is regarded as “good for black”.

OH! It could also allow for a new kind of joseki search: by (5-factor) result. :slight_smile:

If we really wanted to include inferential keywords, I would restrict myself to patently obvious “bad for [color]” in the sense of blundering or getting tricked.

Hey, why did you describe it as 5 factor? I interpreted it as 3 factors: color(2) x getting(2) x access(3)

Ah, yes, it’s a bit confusing I see. I only wrote [color] because I expected the viewer to look for a certain result for themselves and the color would be interchangeable (only leaving corner x outside x top x center x right). This way if you didn’t care about what outside you get, you could just specify outside , though I admit that’s a bit fuzzy still. Moreover, the case could be made that you’d want to use two colors in a request to look for a way to secure the right while closing off center access for white.

Looking at it in yet more depth, it’s hard to argue we’ll end up with independent factors, but my idea was that no combination of two is mutually exclusive. You can get the corner and some outside access.

Perhaps it is the most concise if we use these 5 factors:

  • color(2) x corner (2) x top (2) x center (2) x right (2)

with [outside] being implicit.

Ah, one more thing. It would actually be very useful to be able to filter out sequences that require “your opponent” to blunder/get tricked because those sequences would confound the summaries.

[something something]

The point being that corner and outside aren’t mutually exclusive and top-center-right imply outside access. Conversely, getting not-the-corner also implies outside, because not getting the corner and also not getting any outside would have to be a blunder (i.e. death).

You can reasonably get (as black here):

And pretty much the same for no-corner. It is, however, difficult to produce

  • no-corner, center

Because if you have access to the center, you most likely have access to at least a single side as well. This means {black-corner-top-right} isn’t mutually exclusive with {white-center-top-right}. Both players can retain access to a region (in some cases they might split a corner, too).

Anyway, this is why I’d vote to dismiss the outside keyword altogether.

I’m not at all sure that “a joseki dictionary” is a place for recording “interesting though crude moves that a respected professional played once”? What do you think about this?

As I’ve stated above, I think a dictionary is a place for reference, not inference.

The mere fact that a sequence isn’t played very often does not mean that the sequence itself is bad. It is just as likely that it is simply only good under rare circumstances. And indeed, as we would expect from a sequence that results in access to the corner as well as the center but none to the sides, it can only be beneficial if you already have access to the sides - as was the case in both Cho’s game ( as well as O Rissei’s ( 16 years prior. Cho won that game, by the way. And did I mention that early 3-3 are really crude? :wink:

The main question the dictionary should help anwer is “Starting from a double low kakari, what sequence of moves will let me retain access to the corner AND the center?”.

If you really really want to express that a sequence hasn’t been seen much, you could include a waltheri (or homebrew waltheri) cross-reference so you could say “2/90’000 recorded pro matches contain this sequence = incredibly rare”. No value judgement, but valuable additional data.

Well - here’s the thing though. Same question in a way: is this a joseki dictionary or something broader?
Is it
“what sequence of moves will…” or
“what _well established as even outcome” sequence of moves will…" ?

It can be anything you want it to be :sunflower:

I’m just saying it would be beneficial to

  • avoid value judgements
  • be descriptive
  • take advantage of growing sample size

That said, sequences of moves need not result in a locally equal outcome in order to be joseki, mostly because we have to consider the whole board position. It is entirely possible to get a locally better but globally worse result.

Moreover, it is obviously good for marketing purposes to use the term everybody knows. If you want to give it a more descriptive and less loaded name, you could obviously call it the “OGS Corner Sequence Database”. If you don’t want to restrict yourself to the corner, you can call it the “OGS Pattern Database”; if you want to give it some more adventurous flair, call it the “OGS Pattern Explorer”.

Hey, it’s great that you are offering names - this was on my list to ask about.

No kidding, I need a better name.

All those are good - how about give yourself license to extend the thinking and make other suggestions, unless those are already your faves…

[…] I think it would be good to emphasize the community aspect of it, so I’d perhaps use something like:

apex - the alternative pattern exchange. - “The apex of OGS.”


FWIW my current take on this (totally subject to the ongoing discussion) is that descriptive outcome labels make perfect sense.

I’m currently going with a set along the lines of

Black gets corner and top

etc (there are 12 for each colour based on smurphs enumeration above)

But at the gross level, I still feel the need for “Good” and “Bad”. And when you have “Good”, there’s still room for “Not as Good but Worth Mentioning”.

I’ve realised that all this needs to be in the context of “what are we trying to achieve”.

I will write some words about that shortly, which ought to help guide what to do (assuming that “what we are trying to achieve” can be agreed :slight_smile: )


Lately there’s some discussion about what the role and function of OGS Joseki is. This thread has some interesting points and ideas.

This concept of “multi-factor description” is something to think about.

Hooray, welcome back @bugcat :slight_smile:


One challenge is that I’m not sure that the availability and utility of these cool things is very clear or obvious, and another is that I don’t have good instrumentation to tell if people are actually using them :s or how.

1 Like