That’s what I was thinking. But on the other hand, games often work better because they limit the player’s options. Take wordle for instance - always 5 letters long, no matter what, and only allows the input of words in a predetermined list
3 Likes
Yes, one color mode will remain, and we can make that the default (or even the only option).
Maybe a full fork would be better, since I think with such changes, we could maybe always leave the purple hints up, and then replace them with green as they are uncovered.
By the way, is it true that no 9x9 fuseki will involve captures??
2 Likes
I kinda like 9x9 fusekle from a game design perspective. You could probably even reasonably limit it so the fusekle length is always the same predetermined length, say 7 stones each time, whereas that’s really not possible with 19x19 joseki. After all, it’s not possible to tenuki on a 9x9 board other than to pass or resign, which I guess could arguably be called tenukis since they aren’t local moves
2 Likes
I know it seems trivial, but I guess some design work is needed to clarify when someone is trying to enter too short or too long of a sequence. Truncating hints (as was originally done) is possible as well, but maybe the interface itself should block clicking beyond the length limit altogether. Even doing something intuitive, like the interface shaking disapprovingly, would be something new to implement.
It looks like @antonTobi has already built up a nice library of varying length sequences, so I would want to be able to support that. Maybe some degree of variability is nice, and the length could be immediately revealed by a default one-color mode.
1 Like
Oh yeah, i just meant that was a design option, not as a necessarily ideal option. I just think it’s such a cool thing about 9x9, that with AI you can pretty much always find “ideal equal” sequences of any length whatsoever, since there’s not the ambiguity of whole-board positions that you have when studying 19x19 joseki
1 Like
Certainly there’s no hard rule like this, but it’s seldom acceptable to sacrifice a stone in the first few moves, so captures generally occur later in the game. Even allowing any legal sequence of moves away from the edges of the board, the first possible capture would be on move 7!
Here’s a more typical example of a sequence with a capture, and it doesn’t happen until move 15:
But that’s a bit longer than what I’m going for, I think 12 moves is a good upper limit (and I’ve only included two sequences so far of that length).
The only one I’ve exluded so far because it included a capture is this one:
So I figured if only 1 in a 100 potential variations include a capture, might as well exclude it to keep things simpler for the UI. But I guess the real trouble for the UI would be when two moves happen on the same intersection, which would be even rarer. If I end up stumbling on another good sequence with a capture I might reconsider and add them in 
2 Likes
Captures are probably okay from the UI perspective, but my real worry is this:
The current one-color mode would not properly indicate the actual length in such a case, and it could be confusing when trying to enter sequences that end up being too short.
However, I guess if length is indicated some other way (even if by rejecting too short sequences), then that would hint that overlapping moves are required. I guess the overall shape might also hint that some overlapping moves might occur.
It’s an interesting consideration when to end the variations; I’ve been trying to stop at a place when the position feels somewhat “settled” and there are multiple good moves next.
For instance stopping in this position here would feel wrong to me:
I guess this is mostly because of what I’m used to from joseki, but as you say tenuki doesn’t exist here so it doesn’t really matter in the same way.
One reason I still like it as a convention is that if the next move is an “only move” but also difficult to find, it would feel a bit unfair to “teach” it as an even variation without also showing the next move.
Here is one of the length-12 sequences I’ve included, where I didn’t want to stop at moves 8-11 because the next moves were kind of forced but also non-obvious:
But perhaps this sequence is actually too difficult, and I should remove it altogether.
This is the other length-12 sequence I’ve included so far, which I also extended to this length because there were not a lot of choices for the last moves, but at least they are natural-looking moves, so should be easier to guess:
So while I’m trying to keep most variations around 8 moves, some naturally end up a bit shorter or longer, and I think this will work fine for the guessing as long as the length is given as the first hint
2 Likes
I think it would make sense to exclude overlapping moves from joseki solutions as well, unless one can figure out a good way to make the hints not confusing. Although perhaps not super rare, there will be plenty of good joseki left even after excluding those, so it’s just an overall improvement to player experience (avoiding some players having a frustrating and unsatisfying time with the game on those days with duplicate moves).
I actually kind of wish Wordle would exclude multiple occurences of the same letter for similar reasons (although it arguably makes the game deeper, it’s so annoying having to explain extra unwritten rules to new players).
1 Like
That’s a really cool consideration. Do you have any idea, when there isn’t only one real move, how many reasonable moves there normally are? Is it normally just two or three? Or more like five or six? I can’t imagine it ever really being more than that
2 Likes
I don’t have a good idea for most moves - I’m not using a very high value of analysisWideRootNoise which means that most of the search is focused on the most promising moves:
Here is what it looks like if I crank up analysisWideRootNoise from the default 0.04 to 1.00:
That’s a lot of playable moves!
In general the number of good moves will be higher in area scoring than in territory scoring due to the lower granularity (common scores jumping in steps of 2). Here is the same search for territory scoring, where I think we can see this effect a little bit:
You can see the effect more clearly by browsing through KataGo’s “solutions” of 7x7 and 8x8 at https://katagobooks.org/.
Chinese rules, 1-4 are all optimal:
Japanese rules, only 1 is optimal:
(edit: I should perhaps clarify that the numbers here are not sequences of moves, just different possible next moves, numbered in order from “best” to “worst”)
3 Likes
I’m thinking about indicating non-joseki moves in the share-results. Maybe by using squares instead of circles for such moves. This would provide some feedback on the quality of their sequences to them and to the people they share their result with, while not restricting players how to play the game (except those who want to hide that they aren’t playing joseki moves, but those can still cheat and manipulate their result reports).
What do you think about the idea?
3 Likes
so something like a white square instead of white circle? it’s a good idea, since right now you won’t get a hint which part of a completely new sequence (like your first guess) is not joseki, I hadn’t thought of that.
I turned on the “dictionary checker” as the default
2 Likes
I think that’s a great idea. It will make the “strict” players happy to see where they and others have deviated off of the dictionary tree, and more “casual” players can still play as normal.
So basically, a hint sequence might look like this
Basically, the symbols turn into squares as soon as the sequence falls off of the joseki dictionary tree. I’d need to figure out the exact graphics (maybe the red X on the board could be placed with a small red square?) Or maybe the graphics should be amend to just a small square on all invalid moves?
While working toward this, I’ll also think about how to refactor and generally redesign the hint system in order to support the later incorporation of forward propagating hints.
As another design note, I think we should reveal the length, with a note like “(X moves)” next to the puzzle number, and revert to truncating hints to that move limit. I don’t think I want to restrict the length of the input, so users can still enter shorter and longer sequences (and just get truncated hints), since some users might prefer playing with the board a bit, to explore continuations, or to help thinking about the general position.
4 Likes
These look more or less the same size
4 Likes
The little node button toggles on and off the letter variations for the next move. Maybe it could also toggle on and off persistent hints? When playing a joseki I’m unfamiliar with, I generally have to resort to rapidly flipping back and forth between move zero and the current move to get all the feedback, which is really impractical. I get the advantage of just visualizing the entire joseki before playing out a single move, but for somebody who doesn’t know the sequence, that’s really not possible at all, and the fact that the green circles shown at move 0 don’t actually give timing data makes it harder as well.
A lot of users expressed concern on Reddit about finding the first move, and my response was normally that they should pick a longer joseki to try to get more feedback first. Maybe if you’re truncating hints to the length of the solution, it would be helpful to show the first move? IDK I kinda hate the idea of showing the first move but love the idea of truncating hints to make it more Wordle-like 
maybe the other players just need to suck it up that they won’t necessarily know the starting move right away 
3 Likes
Doh, I didn’t realize that there was a emoji. Hopefully the Unicode conversion works as expected.
1 Like
Yep! People don’t ask for the first letter for free in wordle 
I’m also for restricting length, but I would prefer to only allow guesses of that specified length. I think allowing longer sequences but not giving hints for the end of the sequence would be confusing for new users. It would be like allowing longer words in wordle, but only giving yellow/green hints for the first 5 letters.
The only issue I see with this is that submitting incomplete joseki may feel a little strange. I think that’s something we can get used to though.
3 Likes
Yes, and if we give them one move, they’ll soon be asking for two.
I think that either way, there are some interface design considerations to be made. How to properly indicate that their submission was rejected for being too long, and that they need to shorten it? vs How to clearly indicate what is going on with hints being truncated?
At least since we are using explicit markers for all three classes of hints on the board, maybe some is already done for the latter case. We could perhaps modify the text to something like “too long, hints truncated”?
I don’t want to entirely block a user from putting a longer sequence on the board, since they might want to play around with the board just like an analysis tool to think about the joseki and continuations (to check if it makes sense). Maybe there could be some sort of indicator (perhaps with the submit button changing text and/or color) of whether the current sequence is too long or too short?
1 Like
I was also thinking just to not allow submission, with some explanation why. Analagous to how wordle says “not enough letters” if you submit too few letters or “not in word list” if you submit a non-word.
2 Likes