Make Tsumego Great Again!

Oh, how did I forget that? :dizzy_face: Sorry.

And please pardon my ignorance when it comes to the programming. I should probably confine my opinions to the building of wooden boards, rather than virtual ones.

1 Like

Haha, please excuse my harsh tone, it’s entirely aimed at the questionable use of the misleading word ‘intelligence’ in the term AI. It’s already questionable in a human context and I much prefer the terms “bot” for ‘robot’ (from the Russian word for “work/er”) to “AI” and the term ‘algorithm’ for a set of instructions to transform one set of data to another set of data. :slight_smile:


No worries. I’ve been treated far more harshly by people much closer to me, and yet I survive :grin:


I had this same idea and created an app that does exactly this (but using glicko rather than elo).

It’s very simple at the moment, so would love some feedback on what direction I should take it :slight_smile:

@smurph Thanks for the link to GoTools. Auto-generating them is a very interesting idea due to no human error being involved, and the possibility of more branches.


Hello @cameronmartin123 and welcome to ogs forums :slight_smile:

I am loving the idea, and think a good, simple, but complete tsumego app is still missing in the west.

But seriously…

✓ Password must contain a lower case letter
✖ Password must contain an upper case letter
✖ Password must contain a special character
✖ Password must contain a number
✖ Password must contain at least 8 characters

God forbit, somebody hacks my tsumego account, right?.. :roll_eyes: are you actively trying to discouradge people from using the service? :smiley: Throw us a bone a make a guest use avilable too, especially in the alpha stages.

Not sure what feedback you are looking for, as there is very little to comment on currently and you are probably planning any features I am about to recommend. But:

  • I would differentiate a bit between wrong and correct. Be it a color or a symbol
  • There seem to be no variations for the “wrong” last moves. (even for the expected mistakes), I think it is good to showcase why the move does not work and/or let users replay the puzzle to try and find it.
  • This is one of my personal wishes, but I find every tsumego app/web (apart from hactarGo) I see, is solely life or dead situations. If we had also endgame and tesuji options, I would love that.

This restrictions are causing more trouble then they are worth.
The password will have 8 characters. The first is uppercase followed by lower case letters followed by 4 or 2 digits, or even only the digit 1 followed by either a !,.


Neat app! It’s quite simple, but I really like the minimalist approach.

Here are some thoughts/suggestions:

  • Dark mode (maybe via just a simple toggle button in the corner)
  • Work on the aesthetic design of a bit:
    • Add hoshi to the board
    • Play with colors for the board and stones (e.g., the stones might look a bit better if they were both made a little bit more grayish rather than pure black/white).
    • Decrease the spacing between the stones (i.e., make the stones larger with respect to the grid).
    • The line widths appear a bit uneven on my device. Maybe this is a rendering issue for the very thin lines. See the example below.
  • I think sometimes the solutions are declared correct a little bit too soon. The example below shows a problem declared correct after just one move. I would have expected the app to attempt at least one response to make me prove why that move works.

  • It seems that the problems are always presented with black to play and set up in the top-left corner of the board. Maybe add options to flip color, orientation, or randomize those.
  • I second @AdamR’s suggestions to broaden the problem set beyond just life/death, to increase the visual distinction between wrong/correct, and to show a refuting move after the user puts in a wrong move.
  • An option to replay a failed puzzle would be nice, but I think any replay attempts would need to be ignored when computing Elo updates.
  • It would be nice to have some sort of indication of progress or the difficulty of the problems being faced. However, I don’t think these numbers should be made too prominent, since I do like the current minimalist style. Maybe put this information somewhere out of way, like behind something that you need to click on to show.

Everyone, thanks for the feedback.

The kind of feedback I’m looking for is what the additions/changes that I could make that’d get this furthest to being the best tsumego app.

Regarding the password rules, these are cognito’s defaults. I’ve relaxed the rules a lot to just 6 chars minimum length.

And @flovo, I do really like passphrases, and the above change should support them. But I also think using SSO or a password manager is best, because then you only have to remember one master password.

Good idea. I’ve opened an issue for this.

There should be for some wrong branches, although not all. I think using computer-generated life and death problems would solve this a lot, since all branches could be enumerated.

I also am planning to allow users to review puzzles, which would among other things show you the correct answer. I haven’t worked out all the details, but one thing I want to highlight is where your branch deviates from any correct branch, i.e: “This is where you went wrong”.

This is something I really want to expand to. Endgame should be easy to generate by using pro matches, possibly validating them using a bot, or just using self-play matches.

Tesuji is more difficult to auto-generate. Maybe I can just use an existing collection for this.

What’s with this (seemingly) recent fascination with dark mode?

I wanted to avoid the skeuomorphic designs of most go boards, but yeah there’s also a lot that can be done with the design. I’ll have a play.

This is probably due to the way your browser renders fractional pixels. I think I need a way of forcing the lines to fall on pixel boundaries. What browser are you using out of interest?

This would also be fixed by using computer-generated go problems. Is GoTools actually the only research/software that’s done this?


Given what the website is currently doing, maybe logging in should be optional? You can still track and store data for individuals on a per device basis using cookies and local storage.

Having people log in might make the data slightly better by being able to associate data for an individual across multiple devices, but I think it may also drastically reduce the overall quantity of data and traffic to the site, since the process of making a login is an extra burden that might turn some away.

I think it’s become popular with a lot of people, simply because they don’t like brightness of white backgrounds. It’s bit a easier on the eyes, and I think it feels nicer to use in the dark. Even just an easy way to make the background black would be nice.

I’m definitely not advocating for a skeumorphic design.

I like the simplicity of plain flat colors, but even in that realm, there’s a lot that can be done just by playing with the palette. For example, here is an example with the stones and lines being pure black/white, followed by the same example with the stones made slightly grayer, and the lines shifted to a muted brownish.



It’s a subtle difference, but I think small adjustments to color design can produce a more visually pleasing result.

Firefox 72.0.1 on a Linux desktop


You should wash your stones. Maybe it’s only the contrast with the forums dark mode, but the black stones look dirty on my display.


Yeah, that previous example might have overdone it a bit.

How about this comparison?

bold flat

Ultimately, my point isn’t to say use this specific off-white/off-black palette, but rather just the general point that the colors could be played with a bit.


I like the board and the lines of the left board, and the color of the stones of the right board…



I’ve changed the colour of the stones and lines, and decreased the spacing between the stones (although not as much as you suggested). Is that better?

The next thing I’m looking at is computer-generating/solving the problems. I’ll update this thread when I’m done, although that may be a while because this is a reasonable involved problem…

So I created a solver/generator, and posted about it here: I created a life and death solver and generator


Here’s a post to revive the original discussion 2 years later…

So, I’m currently in the process of adding some high quality puzzles to OGS from a book written by a pro. The problem I have is accurately judging the difficulty. Therefore, in line with the original suggestion and subsequent agreements, I think it would be better to have: (a) coarser categories (e.g. elementary, easy, medium, hard, very hard) to give an initial rough guide to the difficulty and these can probably be set fairly reliably by the puzzle creator; and/or (b) automatic ranking of puzzles based on successful / unsuccessful solving of the puzzle and the user’s rank at the time of attempting the puzzle.

I agree with the other posters that this is an important change for the puzzle section because the current system of puzzle creators having to assign the difficulty of a puzzle from a list of ~ 40 ranks is clearly resulting in many problems with misleading difficulty settings.

I wonder, 2 years after the original post, if there is any progress or plans on this idea?

Many thanks :smiley:

1 Like

I think it would be better to have … coarser categories

I agree. I’d suggest something like:

  • 5d+
  • 1-4d
  • 1-5k
  • 6-9k
  • 10-15k
  • 16-19k
  • 20k-

Or we could reduce the granularity even further, to just

  • High dan / pro
  • Low dan
  • SDK
  • DDK
  • TPK


you ask the players who tackle a go puzzle to score the puzzle on a scale that goes from very hard / hard / neutral / easy to very easy


you combine this with their rank


in time (if enough players of diverse ranking participate) I suspect that you will get good impression of the difficulty of a go puzzle.

Great that you do this. Love it.


Not sure if this will work, because if a good player sees an easy puzzle and decides to skip it, this will influence the result in an undesirable way.

Yes, I like your idea of players essentially voting on the difficulty of a puzzle and combining this with their rank to determine the displayed difficulty. After all, just because someone gets it right, doesn’t mean it was necessarily easy - it might have taken them a lot of hard work to get the right answer.

I do think the categories should be elementary, easy, medium, hard and then one of very hard, advanced or pro as these are in line with existing tsumego conventions, e.g. Cho Chikun encyclopedia of life and death.

Maybe time spent on solving the puzzle could be taken into account?

Agree with you.

1 Like

I believe chesstempo would be a good model to emulate.

1 Like