How to make KGS better?

[xela]

Getting a bit off-topic now, but algorithm-as-circuit has come back into fashion with some surprising applications: see https://www.alexirpan.com/2016/02/11/se ... ation.html

[Kirby]

wms sold KGS three years ago.

Three years ago, when it was claimed that KGS had been sold, and people were speculating about who the new owners were, some admins pointed out that what WMS didn't say that he had sold KGS, only that ownership had been transferred. WMS might have donated KGS to the AGF, for all we know.

The official announcement says he ceded control.

I believe this post was also from wms: https://boardgamegeek.com/thread/175662 ... 3#25523313

I don’t know any specific details.

[mhlepore]

The American Go Foundation is pretty transparent about how they run their business. The 2018 Treasurer's report has a lot financial details relating to KGS:

http://agfgo.org/downloads/AGFTR18.pdf

[SpongeBob]

I used to be a KGS+ subscriber way back when. Now, KGS+ is based on donations, but I cannot stand to listen to it, due to the comically bad audio quality.
It would be really nice to have audio quality that does not sound like from the previous century.

I grew up (as a Go player) on KGS and back then, installing Java was not a problem for anybody. It is the native client and native clients have their merits, too. However, the native client today looks exactly the same as 10 years ago.
So my point is: do not invest all resources into new clients, but update the native one, too.

Biggest feature wish: Voice countdown in byo-yomi (it is standard now, every other server has it).

[SoDesuNe]

The American Go Foundation is pretty transparent about how they run their business. The 2018 Treasurer's report has a lot financial details relating to KGS:

http://agfgo.org/downloads/AGFTR18.pdf

For the lazy but interested:

Code: Select all

Per July 31, 2018 (USD)
Income
KGS transfer + income: 10,460.37

Expenses
KGS Web Maintenance:    4,488
KGS Admin Service:      3,600
KGS Pro Services:       9,541.56
                      -----------
        Total Expenses 17,629.56

Gain/(Loss)            (7,169.19)

For comparison: lichess.org costs as of today (I would have taken OGS but can't find any numbers)

Code: Select all

Per August 9, 2020 (yearly costs, USD)
Income
- donations not disclosed -

Expenses
Total Server Costs:    55,961
Main Developer Salary: 26,966,29
Site Administrators:   36,000
Tournament Payouts,
Chess Content &
Tournament Coverage:   30.000
Misc (incl taxes):     41,208.86
                      -----------
       Total Expenses 190,136.15

As of now:

Code: Select all

        KGS      Lichess      OGS    
Users   700      54,000      2,400
Games   220      22,000         90 (live)/29,000 (correspondence)

Some "easy" observations (KGS vs Lichess):
0) OGS is not even a contender if you want to play live games.
1) Lichess has roughly ten times the server costs (compared to web services) for roughly 90 times the games and users.
2) Lichess has ten times the admin costs for roughly 90 times the games and users.
3) Lichess' costs for extra services (tournament payouts & coverage and chess content) is roughly three times higher than KGS' pro services... for roughly 90 times the games and users.

"Easy" because on its server side Lichess also offers a lot more than KGS. First and foremost free analyses of all games, then opening explorer, problems/studies and other training tools and a community area...

Conclusions?

I still believe there is a niche for a western go server. OGS does not move its consumer experience forward, too. And OGS' player base seems to be primarily in the kyu-area (live games).

KGS has a healthier player base in my opinion and should have more pull name-wise. Since the underlying construct works, my first inclination would be to focus on making a pleasant (webbased) consumer experience (UI/UX design).

[Bill Spight]

Since the underlying construct works, my first inclination would be to focus on making a pleasant (webbased) consumer experience (UI/UX design).

Speaking as a contract bridge director, one important thing that I learned, not in the director's training course, but on the job, is that one of the most important jobs of the director is to provide a pleasant, welcoming atmosphere and an enjoyable experience to players. I think that goes double for online administrators, because of the lack of FTF human contact.

[SpongeBob]

I just spent way too much time trying to shrink my new avatar image down from 14K to the allowed 7K. Relaxing the size restriction to 15K would make life easier for users, I think.

[Nostromo]

The main reason I play mainly on the other servers is that I don't like KGS' "escaper-friendly" policy.
On the other servers, if you leave a game for any reason, your clock keeps running. If you lose internet connection (rare nowadays), you have 5 minutes to re-connect, and if you don't re-connect in that time, the game ends and you've lost.

Most people seem comfortable with that. It greatly reduces the "escaper" problem. Why can't KGS do that too?

[Harleqin]

The main reason I play mainly on the other servers is that I don't like KGS' "escaper-friendly" policy.
On the other servers, if you leave a game for any reason, your clock keeps running. If you lose internet connection (rare nowadays), you have 5 minutes to re-connect, and if you don't re-connect in that time, the game ends and you've lost.

Most people seem comfortable with that. It greatly reduces the "escaper" problem. Why can't KGS do that too?

This debate has now been going on for 20 years. I think that it is nice that KGS offers an alternative here, treating people as adults and games as games. If it were a tournament, then you'd obviously need to handle that, but otherwise I feel not very bothered even in the cases where I am sure that it was an intentional escape (what from?).

As another perspective, not everyone has a stable connection. I have been frustrated by losing games because of disconnects much more often than by other people disappearing.

[RobertJasiek]

It is hard to relate KGS ranks to real world ranks. When KGS 3d as real world 5d, I was surprised to see that I am not alone: another KGS 3d is real world 6d, yet another player is KGS 2d but real world 4d etc. KGS 5d+ ranks can be inflated for some players; KGS high dan ranks increase faster than real world ranks. Many years ago, I was KGS 5d at times but bots have made competition tougher so many players have decreased 1 or 2 ranks, although a few real world 5d (those close to real world 6d) are KGS 5d. Ideally, KGS ranks should be independent of the impact of bots to system while in reality the presence of bots has caused overall rank deflation at least in the mid to low dans.

Recently, I have been between KGS 3d and 4d. I play stronger now than years ago, both times as real world 5d. My lower KGS ranks have two reasons: 1) the overall impact of bots, 2) for a period, I played badly, lost too much, deservedly went down in rank but could not recover the KGS rank because increasing KGS rank is very much harder than decreasing it.

Nevertheless, I have overcome my temporary inability to improve a KGS rank. Now, when I fall back to KGS 3d, I can fight my way to KGS 4d within reasonable time. I can do so even when it requires winning 80%+. In earlier years, increasing a dan rank required ca. 69% wins. That now 80%+ can be necessary is absurdly high and does not conform to ELO style rating systems. KGS should be improved by avoiding such high occasions of required winning percentages for some (not all though) planned rank incremenets.

That I have regained my ability to increase back from KGS 3d to 4d regularly and reasonably quickly has the following causes, of which each is necessary:

1) My play has improved as a consequence of writing my recent books, because I play more disciplined and because I have changed my playing style to more realism (no uncertain moyos but preference for certain territory), a bit more AI-like and better understanding of the opening.

2) Against opponents with equal rank or 1 rank below, I only play even games with komi. Against opponents with 1 rank above, I play even games with or without komi (the KGS default for "without komi" makes this more frequent against such opponents but I would always accept even games with komi against them). I only play against opponents +-1 KGS ranks. This range adapts when my KGS rank changes. Within the lowest 1/3 of a rank's rating range, I do not play against opponents 1 KGS rank stronger. Escapers etc. immediately make it to my ignore list, where they remain for years. With this policy, most unpredictable rating effects (such as from opponents a) with higher ranks when playing them that quickly lose several ranks afterwards or b) 2 ranks below mine) are avoided. The policy is necessary to enable KGS rank improvements after losing a rank temporarily. That this policy is necessary means that KGS ratings must be improved so that playing against opponents +-2 KGS ranks in even or no komi games would have reasonable impact on one's own rating.

3) Usually, I avoid lightning games (10s byoyomi) unless I am in the proper mood to have about the same winning chances as in slow games.

Among KGS 3d opponents, quite a few are clearly weaker, and I lose significantly more close games than my 3d opponents lose close games. KGS 4d opponents feel more like providing an equal challenge. Currently, I do not know how KGS 5d would feel because I am below 4 1/3 dan rating and play against too few of them; presumably KGS 4d is my currently best fitting rank. That I am KGS 3d for about the same duration on average in recent times is caused by the fact that increasing one rank is very much harder than losing one rank. When at a lower boundary of a deserved rank, chances are great to pass that boundary and then the fight upwards has to repeat.

How to make KGS better? The number one reason has always been and continues to be: make rank improvement as easy as rank decrement! So far, there is no symmetry.

[jlt]

I didn't have the impression that getting promoted was more difficult than getting demoted, but maybe the situation is different at kyu levels. Here is my rank graph:

Capture.JPG (29.96 KiB) Viewed 29634 times

I play rated games quite infrequently, only 31 games since last June, sometimes I don't play any rated game for 1 month so the rank graph jumps up and down wildly. The size of upward jumps looks similar to the size of downward jumps.

[RobertJasiek]

Playing infrequently explains a lot:)

[Harleqin]

I just had the thought that this may be a general problem with (all?) current rating systems that rely on statistical calculations for rating adjustments. They have basic assumptions that (1) winning and losing have the same “probability” and (2) thus the same “severeness”. These might not actually be the case. As an intuition: games are more often decided by blunders than by brilliant play, so maybe there is no real symmetry in winning and losing. How to model this?

The other source of statistical noise is of course that there is no randomness in opponent choice, but only (limited) arbitrariness. There are likely wafts of sub-populations drifting around, which more often play among each other, and whose ratings as a whole drift only loosely coupled to those of other sub-populations. Depending on the rating area, playing time (timezone), and settings (fast/slow, ruleset…), you might encounter one or the other of such populations and then have seemingly random strong or weak effects of results on your rating. This is again not necessarily symmetric, as you may win more against one population than the other, but their strength-to-rating correlation might not reflect that.

I think that the “right” general way to counteract the latter effect is to restrict rated games to a very pre-deterrmined format with actual random pairings, fixed schedule, and fixed settings. This is of course a rather large step and akin to a soul transplant for KGS, so it might not be the right way for KGS.

[Boidhre]

Yeah without tight controls on game type, playing conditions etc you can't expect KGS, OGS ratings etc to necessarily represent a players strength accurately. But so long as the handicaps feel right I'm not there's a problem from the perspective of the server working well. The issue is more people think 1d has some kind of intrinsic meaning in terms of strength that should be shared with ratings coming from say tournament play or correspondence play rather than a KGS 1d just being what the server guesses someone who can play KGS 1 dans even should be.

[RobertJasiek]

There is another reason why it is hard to improve in rank: quite a few players with equal rank or one rank lower decline matches because they know that their winning chances are 0 - 10%.