Having read the entire discussion before posting, I assure you that I have all the necessary context. In fact it just earned me the Reader badge. I swear I’m not trying to misrepresent or undermine you, I just have a different opinion.
No one has proposed eliminating CG in this thread. The most severe change to CG that was suggested is making all CG unranked. But even this most extreme option would not prevent anyone from starting a CG to play with their friends.
If I’m just trying to get games, I don’t want to unintentionally play the same person over and over again. I take Regenwasser to mean that a well functioning AF can prevent this, compared to CG where I might accept many of the same person’s challenges if we often play at the same time. He can correct me if I’m misinterpreting him!
I guess I would prefer not to wait in two queues for a game at the same time (AF and CG). Visually AF is emphasized on the Play screen as the ‘default’ option for finding a standard game so I assume my best chance of getting into a game quickly is in that queue. But I admittedly haven’t tried to get games by making my own CGs so I don’t know if this would be faster on average.
I mostly just think that there shouldn’t be two separate queuing areas to pick between. Maybe a solution in the opposite direction would be to get rid of AF entirely and have those buttons just create a CG with predefined settings, or accept a game that fits those settings if one already exists. It might start to really clutter up the CG list with too many active players, but I suppose that’s a nice problem to have!
Generally, I think it would be an almost perfect solution. However, it’s not clear how to solve the problem with ratings. Custom games could have all sorts of time and rule settings, affecting the variability of rating across different settings inside the same larger class of games (e.g., “live” or “blitz”) significantly. E.g., byo-yomi of 20s and 30s are both “live” games, but those are rather different types of games. Hence the idea of standardized and rated quick finder games, and unrated custom games with any types of settings. This would bring a clear separator between completely different use cases.
Another bad news is that we really need some data on user journey, where and how newcomers drop from the website, in particular, and I’m not sure if there is a way to get it. Because there could be all sorts of things, like newcomers are totally fine with quick finder / custom games, but drop after they can’t score the game properly. Or maybe there is indeed a problem with quick finder, and they drop because they don’t understand the interface and can’t find a game quickly. Only with the data we will be able to truly identify the bottleneck.
Yes and also the problem with random pairing, which would be a huge problem in my opinion.
Do we really need that to get moving at all? I made a feature proposal that would probably enhance the AF experience without affecting the CG experience too much. Why not start from there and see if we can make AF better little by little.
True but I think the ability to get games is the number one factor. That is pretty clear just by thinking about it and by reading the posts here that say so.
I also play on Fox mainly because I get games there and I don’t on OGS.
I agree with you, but to be completely fair, this is a hypothesis, an idea of an bottleneck on which to focus our efforts. And because people are reluctant to change, I want to have a more strong argument, with data, ideally. This is a quite common situation – that we “fight” with the status quo, and need to put more investments into our suggestion to change the status quo, rather then defenders of status quo need to put into their defence. I mean, OGS works somehow, right? So it is not that bad. But I do believe that it might be much, much better.
Another benefit coming from the “data research” strategy is that we could identify bottlenecks and increase the number of active players, potentially leading to removing the need to rework the automatch quick finder. I mean, with lots of players in the queue you would get a game in no time regardless of how many settings modes there are.
I think OGS has a very strong position in two areas. First, OGS is great for correspondence games. OGS might even be the best place there is for correspondence games. Second, OGS is very strong with Go variants, kind of like a more polished version of govariants.com. You can try out Rengo or play according to the NZ Go rules and more.
But OGS is lacking in the area that is most interesting to me personally and that is competitive live games. OGS does not seem to understand the concept of competitiveness and what that comes with. They also don’t seem to understand how important queueing times are for a queueing system, or how to decrease them. And they do not really seem to focus on the whole topic of live games at all.
I believe this to be a paradox. I believe that OGS will not be able to significantly increase the number of active players without reworking the matchmaking.
But I think you’re right in general. If we could prove that the lacking live game experience causes players to move to other servers, instead of me just rambling about it once in a while, then the issue might attract some more interest.
Except isn’t that just automatch, being phrased differently?
Automatch creates a challenge with predefined settings (it’s just not visible in the CG list) and it accepts (matches with) a challenge that already exists that fits the settings.
No, it’s not. One essential difference, and downside in my opinion, would be that you would not have random pairings. For AF the algorithm decides who plays against each other. In the CG list some random dude decides that he will play against you.
You might be slightly missing my point. If one person clicks a “formerly automatch” button which makes a “custom” challenge with predefined settings, and another person clicks the same button which then accepts that “automatch” custom match since it exists with the right settings - this is just the automatch.
So it’s not really doing away with the automatch, it’s just in disguise.
I actually think as well having the automatch create custom matches which show in the list will scale very terribly with more players. Imagine if 50 people click the automatch, 50 new challenges appear in existence at one which completely jumbled the custom match list you were just looking at, and then some of them match instantly or near instantly, again jumbling the list.
It would probably ruin the custom game list after a point.
It might be nice in theory to join the two in some way, but it could just be better to
make the matching algorithm faster in automatch
lower the number of options as another way to make the matching faster
be able to automatch from the home page, or make the home page the play page or something useful for non correspondence players
be able to automatch again right after a game finishes from the game page
Less useful but some other ideas anyway
show the activity in some way so people can tell if the automatch is creating games at your level frequently and recently (I don’t think this will work perfectly, but it might help encourage people initially that are discouraged by currently slow enough times at certain times of the day)
something to do while queuing? daily puzzle, spectate etc
In the scenario you describe that is true. But in another scenario where someone picks the game out of the CG list before a second person clicks on automatch again then what I said will happen.
Also depending on the implementation, matchmaking algorithms can consider several factors for machmaking and do not necessarily need to match the first two entities that enter the queue.
I personally think we should focus on the low hanging fruit which is a UI overhaul before touching other items on the list.
Regarding “random pairings”: To my knowledge random pairings are generally recommended in these types of situations because they provide a better accuracy, fairness and integrity in a competitive matchmaking scenario.
List-selection based games enable players to artificially boost their ratings by focusing on weaker players. They also enable players to “target” others which might have a negative effect on the targeted player’s experience. These are general shortcomings of list-selection based games. This is independent from the question of how “noteworthy” the overall effect of this is on the rating.
But there is an even more important factor here when it comes to “random pairing”. And I realized just now that I did a bad job by calling this concept “random” because it is not random. It is actually “rating based” pairing. Which leads us to the second large shortcomin of list-selections: They are not actually based on the exact ratings. So by using list-selections you are undermining all the heavy lifting the “AF algorithm” and the “Glicko2 rating” would have provided. We basically flat out ignore all the data we have and all the considerations the algorithm would have enabled us to make, and instead just overwrite it manually.
I most agree with your statement about focusing here, and focusing on UX of AF sounds like a very good direction in particular. What can we do incrementally to make AF better?
There random pairings thing feels like a distraction - we might have a hunch that current pairings are not good, but that requires it’s own investigation and discussion - belongs to another thread.
There are plenty of people who would like to just come to the site and click a single button (or maybe one of a small set of poplar options for blitz/live and board sizes) to reliably get a game. And AF isn’t working for that right now
Yes, I agree with step-by-step approach. Again, ideally, I want to see a data on how beginners drop from the website due to the issue with AF, but seems like many people agree that redesign is needed. Clossius already did a great job on the redesign: New Landing Page Idea
But I too think we need to cut options even more. Having more options is what scares people, beginners especially. One blitz, one bullet, one live. Correspondence are going to custom games. But we need to decide what to focus on, the problem we are having is that there is too much. We should decide on time, on handicap, on something else I forgot about.
For the next stage, I would try to implement automatic pairing between AF and custom games if the rules are similar enough. What we need to do is to decide what rules are similar to not break the rating. Is 20m+5x30s the same as 40+5x30s? 40+3x20s? But I don’t have anything against putting more strict rules, I think it would only benefit as people would have more understandable rules to play by – and expectations from the server, which is the most important part. Especially considering most of the customer games are correspondence games.
Because the community won’t drop the settings, we could try to implement a fallback scenario for cases when there is no game with your specific settings, but with the closer settings. E.g., you ask for fischer time, but only byo-yomi is available. You could have a pop-up or something with the suggestion that you could either continue waiting or agree to the another settings. It’s a little bit weird because otherwise what the point of settings is, but what can we do, right.
We could just analyse the data to see what the most popular formats are. I would do it, but I need a database dump. I tried to do it already, but I was, sorry, but, suffering from getting the dataset from API.
I did a quick analysis on the data. The results show us that live games and correspondence games are the most popular game types players choose, haha. If I were a product manager, I would say that let’s cut chinese rules and fischer for automatch, and focus on making the ideal experience for the major population of players - the population the uses japanese rules and byo-yomi. But people wouldn’t agree with me, of course. The question stays: how to make controls that won’t confuse people due to the diversity of results (handicap/non, different time controlsm etc.), which is critical for new players. People want to see same results from the same inputs, in our case – time controls, rules, etc.
Byo-yomi
X axis – seconds; Y axis - number of games in millions
I think it’s really great to have found some allies with similar goals in this endeavor. But I think we could achieve more if we talked to each other instead of talking to ourselves while being in the same room. You literally just repeated everything I laid out in my feature proposal, a few comments earlier, in your own words.
What do you think about my proposal? Is there something you would change?
And now we’re jumping ahead again. Which is not necessarily bad cause we should at some point think about some kind of roadmap but I think for now we should try to get the UI overhaul going.
To be fair those options are also the default. This could also show that most people don’t care at all.
Yes, they would because we’re only reworking the AF. The players that love customization play CG anyhow. If we can’t get rid of overcustomization of AF then we can’t fix the AF and all this talk won’t get us anywhere anyways.
I don’t understand the question. I think this has already been solved in my proposal or what exactly is unclear?
Fischer and byo-yomi both have pros and cons for Go, but Chinese as implemented on OGS is far better than Japanese, and should be the default (I think NZD should be the default, but Chinese is a more reasonable compromise candidate), as it allows playing out to resolve scoring disputes and avoids the insanity that Japanese rules is in extremis
I swear I’m not trying to misrepresent or undermine you, I just have a different opinion.
