Board Game Arena

sprockitz wrote: ↑04 March 2020, 19:53 True but I have a sneaking suspicion this isn’t working quite right...

Yeah that's the trouble with algorithms. It happened to us a lot you know to think "Is this right?", because it's complex and there can be many reasons for which there is no match since there are a lot of options that can be incompatible. When we digged it checked out as correct. So my informed opinion is that it works, and now I trust it unless proven otherwise, because at some point we can't spend our days checking on algorithms ^^

sprockitz wrote: ↑04 March 2020, 19:53 With some work you can find the actual open tables and my settings have been such that I should have joined those tables (as they were listed as either first choice or okay to join). However I would start a new table instead of joining the already open one. As an example, After failing to join the open (2 turns per day) table the system instead started a new (1 turn per day) table (despite both being okay or 1st choice). I quit looking for game, change settings so only first choice is two turns per day and look for a table again and am immediately added to that table.

If I understand correctly what you are saying, it's what's intended. The reasion is that first choice is not the same as OK. If there is a first choice, it's the only valid one for the initial period of search, and the OK choices will match only after a while when the matchmaking expands the search for second period of search (if you think about it it's logical, otherwise you would get your first choice only by being very lucky).

dschingis27 wrote: ↑04 March 2020, 20:18 Maybe it's the same bug causing all the hassle with the filters in real-time Arena? Oh would that be nice if that bug were caught finally.

The wait times when applying the filters are absurd, I think many users are already convinced they MUST be bugged.

Sorry to say, but this is a little unclear for me

I would say that about any matchmaking process on BGA or anywhere else, if you set filters that get you in the same pool than a lot of other people, then you'll get a match quickly. If you set restrictive filters, then the pool of people with compatible wishes is of course small, and so wait times will be longer. So if you don't get matches as quickly as you like, be less restrictive or use the defaults to be in the largest pool.

It's always a compromise between getting the exact options you want and the time you are willing to wait for someone with the exact same wish. Matchmaking cannot create an opponent for you (yet... )

Een wrote: ↑04 March 2020, 22:17

sprockitz wrote: ↑04 March 2020, 19:53 With some work you can find the actual open tables and my settings have been such that I should have joined those tables (as they were listed as either first choice or okay to join). However I would start a new table instead of joining the already open one. As an example, After failing to join the open (2 turns per day) table the system instead started a new (1 turn per day) table (despite both being okay or 1st choice). I quit looking for game, change settings so only first choice is two turns per day and look for a table again and am immediately added to that table.

If I understand correctly what you are saying, it's what's intended. The reasion is that first choice is not the same as OK. If there is a first choice, it's the only valid one for the initial period of search, and the OK choices will match only after a while when the matchmaking expands the search for second period of search (if you think about it it's logical, otherwise you would get your first choice only by being very lucky).

Without seeing tables and without trying to recreate the scenario it is impossible for me to be sure...as I can't say for certain that I had 2 things as first choice (but I thought I did) when it failed to join as it may have been okay...and again kinda hard to test on our end without easily visible tables (which would be nice just don't list who the players are ). Not knowing how often the algorithm actually searches for tables created by someone else in all the categories outside the 3 where your tables are created also makes it hard to tell...hopefully more often than just at start, +4 hr and +12 hr searches for other tables are being done. I don't understand why only 1st choice, prefer not, and reluctant create tables, it seems like it would make more sense to make a table for 'okay' in the second pass versus skipping over that table and going to the prefer not choice. b/c okay was being skipped over I was under the assumption that it was being treated equitably with first choice.

sprockitz wrote: ↑05 March 2020, 04:48 Without seeing tables and without trying to recreate the scenario it is impossible for me to be sure...as I can't say for certain that I had 2 things as first choice (but I thought I did) when it failed to join as it may have been okay...and again kinda hard to test on our end without easily visible tables (which would be nice just don't list who the players are ).

I understand how you feel, but actually seeing the tables would just create more confusion, since all the info is not immediately visible (especially if not showing who the players are). I can't count the times when sourisdudesert told me "this is not working, look at this" and I went "oh, no, I have to check again" then checked and answered "it works, you just forgot to take that into account" (that being player levels, red thumbs or whatever). So we would just be drowned in false "this is not working" reports, and we don't want that, when we spend time on false positive we are not using our time to move the site forward

This is a case where more visibility would actually undermine trust more than help it, because it needs real expertise and a lot of time to evaluate properly what happens.

I would add that this system generates 7 Gbytes of logs every day, so you can see that it's not a volume where we can examine everything manually. We have built some tools for checking to make reasonably sure that it's going well, but overall the main thing we look at now is the number of matches, with alerts if it drops below a certain level, so that we can check if something is going wrong.

sprockitz wrote: ↑05 March 2020, 04:48 Not knowing how often the algorithm actually searches for tables created by someone else in all the categories outside the 3 where your tables are created also makes it hard to tell...hopefully more often than just at start, +4 hr and +12 hr searches for other tables are being done.

Not sure if that's what you want to know, but currently matchmaking is processed every 5 seconds for realtime and every minute for turn based; the search expands options being considered after 2 minutes / 4 minutes for real-time and 4 hours / 12 hours for turn-based (note that this may change anytime for technical or functional reasons so those values should not be considered final in any way).

sprockitz wrote: ↑05 March 2020, 04:48I don't understand why only 1st choice, prefer not, and reluctant create tables, it seems like it would make more sense to make a table for 'okay' in the second pass versus skipping over that table and going to the prefer not choice. b/c okay was being skipped over I was under the assumption that it was being treated equitably with first choice.

Well, I think it may be a question of semantics. It's not easy to express clearly and concisely the concept behind these preferences, we had a lot of back and forth on this before settling on the terminology explained in the FAQ (https://boardgamearena.com/faq?anchor=f ... implelobby). Maybe it would help to think of OK as "no opinion expressed". It's used only if there is not a preference explicitly expressed for the matchmaking expansion being considered. If we have an explicit preference, we always use it rather than the absence of preference.

Hope this thread helped shed some light and trust in the process and did not just add confusion about the complexity

Thank you, Een, that definitely helped. I admit that my first post was not really helpful, sorry for that. I totally agree that I should expect very long wait times for specific settings or unpopular games.
Still I think there is some bug with the filters but I will write a proper bug report for that when I have the time.

Bug or not, I would like to describe two experiences that can help you understand why users may continue to report that the matchmaking feels strange.

1. For 7Wonders in manual simple mode, usually, it takes just one or two minutes or at most 5 minutes to fill a table with "strong or higher" players (Elo>300). Yet in Arena, when trying to find "Gold or higher" players, it often took 30 minutes or more or didn't work. (This was the situation a few weeks ago, I haven't really tried this recently). These two experiences taken together just felt totally odd.

2. I am Silver in Can't stop and tried to find +/-1 opponents. I waited 40 minutes and nothing happened which felt ridiculous to me because Can't stop is such a popular game on this site. I checked and quite some Arena games started during my waittime but far most were only Elite players with occasional diamond players. So no obvious bug here but this is an odd phenomenon in itself, that there seem to be only very few active lower-league players in such a popular game.

I am not sure what you can do about it but I wanted to report it here. Thanks agaiin for the extensive replies.

Here is my specific bug report: https://boardgamearena.com/bug?id=14597
If it cannot convince that there is a bug then I don't know what can.

dschingis27 wrote: ↑08 March 2020, 19:52 Here is my specific bug report: https://boardgamearena.com/bug?id=14597
If it cannot convince that there is a bug then I don't know what can.

Thanks, that's clear and specific. I'll try to dig and see if there is an explanation or if it's a bug.

And... It was a bug. You were right, thanks a lot, indeed it was a general failure when setting the level filter in arena mode (it worked fine with the automatic lobby in simple mode, but there was a small difference...)

We just made a release fixing this (and other things). Thanks again! Now it should work as you expect it to!

Nice! Thanks for digging and reacting. I just renewed premium.

Maybe you make some kind of release note because I think this bug was annoying quite some users.

dschingis27 wrote: ↑10 March 2020, 08:18 Maybe you make some kind of release note because I think this bug was annoying quite some users.

Yes, we checked on who was affected by this and will send an email to notify them about the problem and that it's now fixed