Guild Wars 2

(They fixed this issue 3 years ago when it was a problem and they have now reverted back to their old awful matchmaking algorithm. They need to do this again: https://forum-en.gw2archive.eu/forum/game/pvp/Matchmaking-Changes-MMR-Leaderboard-Reset/first )

Take a look at this screen shot.

1, They have 2 duo queues. We have none.

2, Then look at the ratings. I added everyone to friends to check the scores.

Team 1
2 duo queues
2070
2043
1988
1886
1725
Average = 1942

Team 2
0 duo queues. ALL SOLO
1810
1803
1778
1771
1764
Average rating = 1785

So they get 2 duo queues. We get none. They get average rating of 1942. We get 1785.

How is this possible. Why were the 2 duo queue of high rated people on the same roster?

This is evidence that gw2 matchmaking system is broken. Why would the system put those 2 duo queues of top 30 people on the same team to farm a bunch of solo queue platinum people?

This is not a salty post. I am just showing with proof that the system does not work at all. It is a post of science. I am just showing the numbers.

You only lost 7 rating

Abelisk.4527:

You only lost 7 rating

So why does the system not collect the 10 players then make fair matches?

I admit MMR is wonky, but it will adjust eventually, depending on w/l and other factors.

I notice that if I’m up against really good players and I stop trying, I get into games with worse players. Once I start tryharding again I get matched with better players.

It can also be match availability. Lots of players in NA are probably sleeping atm, so the rating swings are going to be huge.

Edit: if you’re EU then it’s the first point

This looks like a sorting problem. Looking at that particular team matchup I would have to guess that the system first assembled the teams with the aim of keeping the team average rating low-variance (this is a good thing) – and then matched two teams based on their average ratings. In the rating range in question, there may not be many teams close enough in their average ratings so that the likelihood of skewed matches increases a lot. But this ought to go both ways (not quite in equal measure though because of the gaussian distribution of player ratings) – so if you find yourself getting matched against the same higher-rated people a lot, maybe give this whole queue-dodge thing a try?

It might be because the rosters are defined and then an enemy roster is located. This would explain the issues. There is something wrong anyway, badly wrong.

Olterin Fire.5960:

This looks like a sorting problem. Looking at that particular team matchup I would have to guess that the system first assembled the teams with the aim of keeping the team average rating low-variance (this is a good thing) – and then matched two teams based on their average ratings. In the rating range in question, there may not be many teams close enough in their average ratings so that the likelihood of skewed matches increases a lot. But this ought to go both ways (not quite in equal measure though because of the gaussian distribution of player ratings) – so if you find yourself getting matched against the same higher-rated people a lot, maybe give this whole queue-dodge thing a try?

Yes, this will be the issue. What should happen is the game finds 10 players who are close in rating and then divides them into teams. The current system is worthless as this shows. They should remove ranked until this is fixed because it will just make people quit.

I had similar experience (during prime time!!!!), OP, posted numbers weeks ago as well. There was no acknowledgment so far.

Abelisk.4527:

You only lost 7 rating

This is a stupid argument because
1. game shouldn’t create such imbalanced matches on first place, guaranteed losses is no go. Guaranteed wins is no go.

2. he may have lost 7 rating, but i am sure the 1810 guy lost like 15 and he will need to win a match ot two to make up for it. There is also no guaranteed that that 1810 guy won’t be put in another such match again. Throws few matches like this (yes, it is a thing in higher ratings sadly) and you can easy drop out of plat.

They need to restrict how the MMR range expands to make matches after time in the queue more. If it means a longer wait for matches so be it. A 150 MMR over all difference per player per team is bad. This match should have been passed on by the rater. This deserves a direct response from dev team.

I’m having matches like these all the time. I’m currently doing the Year of the Ascention achievements and I’m on tier II, 4 wins in a day is like a dream come true. For the last 3 days I have 2 wins.
But no complains here, I’m just a noob Thief that can’t carry 4 ppl all the time!

This is a reason of many blowout matches. I don´t get it why the double duo is not split. I would implement a second round after finding the matches than analyses the outcome and then even out teams and classes with making the average closer.
In this case swaping the second team with the two lowest solo.
What also worries me. I am low silver and i think one of the names had been in a match with me …. As said i encounter 2/3 of the solo Q matches as 200+ diffrence outcome. This match imbalace makes the system random and makes bad match quality and it loos so easy to avoid that. You can never turn of human factors so blowouts will always happen but you can increase match quality and as result indirectly lower the rants and AFK´s which doubles up for match quality.

this been happening since the start, also this is the reason anet doesnt want us to see everyones mmr during match. coz they know MM sucks. so much for that spvp trailer they made with “matchmaking” word flash in the screen LOL

yep if they show the teaming and MMR of such a match, people will give up before start …. Would be even worse. I understand that and won´t do it. Showign it after the match would be the good thing.

Abelisk.4527:

You only lost 7 rating

I made a thread like this and i believe i had lost 12.

Either way, he is getting penalized for a match he cant win.

The rating system makes it so no one wants to play ranked.

You have top 25 players winning and getting +2-4 and when they lose -30-50. Then all of a sudden someone makes a alt account and goes to the very top of the leader board. Thats not right.

Lets be honest here, the OP is probably a plat-plat2 player. And he simply doesnt want to get put into gold by having 3-4-5 of these matches which can easily happen. Also its simply bad match making from a fun perspective.

Why not just keep him at a plat level and try to create fair matches instead of these lop sided teams to give these ESL players the best chance of winning.

This league system is not good and we are in Season 5.

They need to ditch the idea that personal rating depends on on match outcome…it’s not fair because you have zero control over other 4 players, it’s a matter of luck whether or not they make the right decisions.

I do not believe that a team based system with random pairing can give an actual representation of player skill….too many factors involved with very little influence by the player

mysterion.1276:

It’s called variance, theres also a chance that the legendary duo is in your team so matchmaking is fine.

You don’t understand. If i asked you, as a human, to split those 10 players into two rosters based on their MMR then you would not do it like this. So the system is broken. That is not variance. That is just a failure of the system. There was a way this game could have been fair (one of each of the legend duos on each team) but the system did NOT do this. I want to know why. And if anet cannot fix it then they need to remove ranked from the game because it will just make people quit.

mysterion.1276:

It’s called variance, theres also a chance that the legendary duo is in your team so matchmaking is fine.

There should not be a chance at all… that’s the point of this thread.

There needs to be a second sorting after players are picked for a game.

Protip, don’t play during nights when there are less players playing. That’s how the duoq guys even sit at so high mmr, they are afraid to play when all the esl players are on and can abuse matchmaking at night by getting easier opponents. System works fine there just isn’t enough players queueing during certain times and then this happens.

mysterion.1276:

It’s called variance, theres also a chance that the legendary duo is in your team so matchmaking is fine.

no, there is no chances:
mm takes 5 players of the aprox same mmr an makes a 5 man team….
after finds other team of the same aprox mmr and makes a match

the 1800+ guys have more math chances of geting the top ones than the 1700+ guys

mm have to put more filters than only mmr to work proper or be done a complete overhault in the form of geting firts a poll of players and after sorting them to balance things(not staking duos not staking classes and not staking high mmr in one team)… s1 complains of pros that they had to carry a lot lead to this, enjoy it former legends in other seasons, this system fault is yours for the continuos complains in the s1 system more based on pip ranges than mmr

Barry.3147:

You don’t understand. If i asked you, as a human, to split those 10 players into two rosters based on their MMR then you would not do it like this.

I’m not so sure about this. If I recall from previous seasons (don’t know if they changed it for S5) the matchmaker tries to find 4 players of your own skill level to create a team of 5. Then it tries to find another team with a similar average rating to fight your team. The matchmaker, to my knowledge at least, doesn’t work on 10 players but on 5 players, so even a human being would cause match ups like the one in the OP to happen.

Of course a human could easily see the difference between the teams and make the adjustments AFTER matchmaking, but before the match started. Is that even possible to do by an algorithm?

Spicy.2481:

Protip, don’t play during nights when there are less players playing. That’s how the duoq guys even sit at so high mmr, they are afraid to play when all the esl players are on and can abuse matchmaking at night by getting easier opponents. System works fine there just isn’t enough players queueing during certain times and then this happens.

Are people trolling me? lol. Clearly there was enough players. There was the 10 players available for a fun and close game. But it didnt happen because the matchmaker did not correctly divide the 10 players into two teams. It is nothing to do with lack of population in this case.

maddoctor.2738:

Barry.3147:

You don’t understand. If i asked you, as a human, to split those 10 players into two rosters based on their MMR then you would not do it like this.

I’m not so sure about this. If I recall from previous seasons (don’t know if they changed it for S5) the matchmaker tries to find 4 players of your own skill level to create a team of 5. Then it tries to find another team with a similar average rating to fight your team. The matchmaker, to my knowledge at least, doesn’t work on 10 players but on 5 players, so even a human being would cause match ups like the one in the OP to happen.

Of course a human could easily see the difference between the teams and make the adjustments AFTER matchmaking, but before the match started. Is that even possible to do by an algorithm?

Well then this is a flaw in the matchmaker. Again which is the point of the thread. If what you say is true then the ranked system should be removed. The system would be much much better to pool 10 players and then create rosters that are fair.

It is possible to do via algorithm. It isn’t even hard to do. You have the system put players into buckets. When a bucket is full of 10 players THEN the system creates rosters based on average MMR on each team. This is the only logical way to actually implement a matchmaking system. You wait for 10 players to create rosters, not 5.

It is so simple that the fact anet do not do it means they have no idea what they are doing.

They never learn. They used to use this broken system and they fixed it 3 years ago. Now it is back.

https://forum-en.gw2archive.eu/forum/game/pvp/Matchmaking-Changes-MMR-Leaderboard-Reset/first

saturday nigth i played a match on low time, 4 am and i get one player in team who played dh the way of a pew pew only uses lb on range, no traps no blocks no nothing only pew pew AA, i added to friendlist, the system not showed their mmr not showed me on rankings i cant know what is their mmr on the exact moment of match, but day after at evening when i come back on game system showed me, im around 1000 they had 550.
now few moments ago i played a match, doble duo on enemy and players that make plof in firts teamfight in my team i take home and before i cap it i had 4 enemys on the point after how match is in temple that 4 enemys camping the only exit and …. blowout

Evan Lesh.3295:

10 players are collected before making teams.
You are looking at ratings after the match which are going to be further apart than the ratings that were used to make the match.

The two duo queues on one team is annoying, and something could be done to improve that.

And as always (though slightly out of date) the matchmaking process has been documented here:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

I thought people might say that to try and invalidate what I was saying…..which is why I also took a screenshot of the ratings before the game was over. The leaderboard for friends takes time to update. Again I include the evidence below as a new screenshot. This shows the ratings before the game was over, which shows the 1942 rating average vs the 1785 average rating. AND the 1942 rating team had 2 duo queue and the 1785 team had none.

If the system does collect 10 players then it is failing to divide teams fairly. You have a bug in your algorithm or something. This happens all the time btw. All the time those of us rated around 1800 in solo queue end up vs double duo queue of people at rating 2k+. Why are these duo queues not split up always to give fair matches?

Something is badly wrong. This happens all the time at plat 2 and above. It just puts all the legendary plays in 1 team for example. And plays them vs a platinum team. These issues aren’t made up. Which is why i presented the evidence. The algorithm does not work to make fair matches. Testing has shown that, as I presented here as evidence.

This game is moving in the right direction and you all do a great job. But I think it is obvious that something went wrong here. And this happens ALL the time.

Whether anyone believes me or not is irrelevant as I have proven it can and does happen in this thread.

Evan, could you add an average mmr of both teams at the end (mmr before the game) paradoxical, this should protect the individual mmr (like you can see in here) and help people understand

Evan Lesh.3295:

10 players are collected before making teams.
You are looking at ratings after the match which are going to be further apart than the ratings that were used to make the match.

The two duo queues on one team is annoying, and something could be done to improve that.

And as always (though slightly out of date) the matchmaking process has been documented here:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

Oh and when it starts with 10 players and makes teams later… could you divide professions equally?

(There where many matches when we (enemy and my team) wanted to exchange eg one duo thief for eg one duo warrior – because the other team had none)

Tiefsee.3647:

Evan Lesh.3295:

10 players are collected before making teams.
You are looking at ratings after the match which are going to be further apart than the ratings that were used to make the match.

The two duo queues on one team is annoying, and something could be done to improve that.

And as always (though slightly out of date) the matchmaking process has been documented here:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

Oh and when it starts with 10 players and makes teams later… could you divide professions equally?

(There where many matches when we (enemy and my team) wanted to exchange eg one duo thief for eg one duo warrior – because the other team had none)

Profession imbalance is secondary. If you start allocating teams based on professions then you get more blow out matches as the system will be matching high rated players togther vs low rated players together just because of classes they selected. The matchmaking needs to match people to have fair MMR skill levels for each team. Anything else is counter productive.

Even accounting for match results, there’s more than 100pts in rating separation between these team’s average ratings. Even looking past the Duo queuers (which I don’t believe is a factor at all, but that’s another argument for another time), how is that an acceptable matchup?

Evan Lesh.3295:

Barry.3147:

I’m definitely going to be looking up the match. I need to see what the ‘real’ before/after ratings are without things like decay and make sure those are the right players.

please seperate the duos in between team.

I looked through and didn’t see where you said it, but I’m curious how many points you lost and how many points the other people gained after the match to MMR. I think that while being mis-matched is annoying, it is less so if the losses aren’t huge.

After looking at the algorithm a few months back it looks like the matchmaker tries to put you sometimes equally matched, and sometimes outmatched. . .but there is an expectation of win and loss that goes with that.

That’s probably confusing, but like I mentioned I’d be curious how much you lost for losing that fight. I’ve been up against hard duo Q fights before and lost very little, and been up against hard Duo Q fights before that we won and got a yuge payoff.

Glad to see evan in this discussion. :-). thanx.

Evan Lesh.3295:

I have good news and bad news.

This example is an extreme case so it is not screwing over everybody. There were multiple people that failed to ready up, which meant multiple rounds of substitutes were needed.

Bad news. Teams are not rebalanced after finding substitutes and it has been that way for years. There are complicated computer-science reasons why, but it is solvable and I will attempt to fix it. In this case since it happened multiple times, it compounded the inaccuracies.

Ahhh, the curse of the edge conditions.

Evan Lesh.3295:

I have good news and bad news.

This example is an extreme case so it is not screwing over everybody. There were multiple people that failed to ready up, which meant multiple rounds of substitutes were needed.

Bad news. Teams are not rebalanced after finding substitutes and it has been that way for years. There are complicated computer-science reasons why, but it is solvable and I will attempt to fix it. In this case since it happened multiple times, it compounded the inaccuracies.

If I’m understanding correctly, the matchmaker does something like this (A):
- 10 players are chosen for match, balanced between teams.
- 5 fail to ready up, matchmaker finds replacements for those from the queue, while remaining 5 remain in the “match” while it’s being formed.
- repeat X amount of times.
- finally all ready up, match progresses.

(That’s the impression I seem to be getting, at least).

Unless it does something like this instead (B):
- 10 players matched up.
- 5 fail to ready up, so everyone is back in the queue.
- match re-formed, again, 5 fail to queue up.
- Repeat.
- Eventually match formed.

Obviously over time, players in the queue are changing, so the matchmaker has to work with what it has access to, occasionally resulting in weird matchups.

IMO, Scenario B should be what happens, not A, since Scenario A is working with a partial match and filling the gaps with as close to suitable players as possible. It’s entirely possible that players in the queue (substitutes) who were closer to the rating of players already in the match are already in another match of their own. Scenario B effectively resets the matchmaking and starts over (I think).

Correct me if I’m mistaken.

Wow, I’m impressed. Thanks for the response, Evan.

Dev of the year already won in this thread.

Better luck next time to the rest. Communication and proactive problem solving will do it

Ive posted the same things except you threw out every ones names to see.

The ratings are slightly different but the rating system is still the same. Ive made 2 threads and have moved onto the leader board but since its up again.

This match Team A ratings are currently

2024
1862
1730
1667
?

And the team i was on

1661- My current rating
1634
1629
1628
1608

based on there 4 players they have a average rating of 1820, my teams ave rating was 1638.

Obviously my match picture was taken about 2-3 weeks ago but ive had maybe 4-5 matches this season like this. And i believe ive only lost single digits in rating once. In this match i believe i lost 12.

Evan Lesh.3295:

I have good news and bad news.

This example is an extreme case so it is not screwing over everybody. There were multiple people that failed to ready up, which meant multiple rounds of substitutes were needed.

Bad news. Teams are not rebalanced after finding substitutes and it has been that way for years. There are complicated computer-science reasons why, but it is solvable and I will attempt to fix it. In this case since it happened multiple times, it compounded the inaccuracies.

That.. makes a lot of sense on so many levels. Though everyone may not have experienced the fault you explained, fixinf it makes things that much better.

Evan Lesh.3295:

I have good news and bad news.

This example is an extreme case so it is not screwing over everybody. There were multiple people that failed to ready up, which meant multiple rounds of substitutes were needed.

Bad news. Teams are not rebalanced after finding substitutes and it has been that way for years. There are complicated computer-science reasons why, but it is solvable and I will attempt to fix it. In this case since it happened multiple times, it compounded the inaccuracies.

Nice evan. Thanks. You are awesome.

I actually wondered this because the queue was an insta pop for me and only i had to ready up. Secondly my friends had a,ready told me that the insta pops were often thekr worst matches

Abelisk.4527:

I admit MMR is wonky, but it will adjust eventually, depending on w/l and other factors.

I notice that if I’m up against really good players and I stop trying, I get into games with worse players. Once I start tryharding again I get matched with better players.

It can also be match availability. Lots of players in NA are probably sleeping atm, so the rating swings are going to be huge.

Edit: if you’re EU then it’s the first point

HAHAHA no it won’t adjust. MMR is so bad right now, whatever they did broke it.

I mean the best I can do is assume what happens by looking at the MMR model, but no wonder my games were SO OFF when people failed to ready up. That is a major fault. Thanks for the response Evan!

Barry.3147:

Evan Lesh.3295:

I have good news and bad news.

This example is an extreme case so it is not screwing over everybody. There were multiple people that failed to ready up, which meant multiple rounds of substitutes were needed.

Bad news. Teams are not rebalanced after finding substitutes and it has been that way for years. There are complicated computer-science reasons why, but it is solvable and I will attempt to fix it. In this case since it happened multiple times, it compounded the inaccuracies.

Nice evan. Thanks. You are awesome.

I actually wondered this because the queue was an insta pop for me and only i had to ready up. Secondly my friends had a,ready told me that the insta pops were often thekr worst matches

I understand why it has been made that way. It makes for a good(smooth) customer experience in match preparation and is good design. Unfortunately it lowers match quality. And this to a higher extent then some might think. There are 10 humans and i always wondered why such a low number of match breakoffs gettting back to Q happen. It is unhuman ^^. The above example is propably one of the more unbalaced but i say it happens in most match preparations just to a much lower extent like one failed and a lower on gets added. The more fail and maybe even some in multiple cycles the more inacurate MMR matching gets. And with 10 humans to syncronize it´s significant.
A full fix to re Q will result in very high preparation breakoffs and i fear to an extent where experience is so bad that it impacts the player base. Imagine your Q aborts an AVERAGE of 2-3 times per match. Maybe you get one that restarts 9 times …
While it will signifficantly increase match quality I don´t think this is a good sollution.
The matchmaker should rebalance both sides after the match is confirmed. I think this is quite doable swaping side for 1-2 players. Especially important if you get those double duo on one side.

Ross Biddle.2367:

EU off hours are usually way worse than NA though O_o

https://puu.sh/tjEsP/c8cf2be3a7.jpg

My friend lost 15 for this match against 2 legendary tier players as a 1.8k MMR guy. Can you please explain that match please Evan thanks.

Evan Lesh.3295:

10 players are collected before making teams.
You are looking at ratings after the match which are going to be further apart than the ratings that were used to make the match.

The two duo queues on one team is annoying, and something could be done to improve that.

And as always (though slightly out of date) the matchmaking process has been documented here:
https://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

Your matchmaking system needs refinement. Why can’t you acknowledge that? How could you not have the foresight of knowing 2-2-1 vs 1-1-1-1-1 split doesn’t work? Why do you still hide averagebteam rating? Need less transparency in this fickle PVP system.

1.) Redistribute players after match preparation to get optimal team, MMR and class distribution between sides. So system gets 10 players, looks at the MMR diffrence, teams on both sides and classes then optimizes by swaping players(teams) from each side.

2.) Show player MMR after match with the top stats.

So you don´t need to touch the current match preparation and keep is smooth. A redistribution of the 10 players won´t need massive resources. You have 10 playes with three variables, MMR, Class and team binding.