Guild Wars 2

Go figure that an unofficial site had incorrect information.
Also, here is what Justin actually said:

Justin ODell.9517:

Trevor Boyer.6524:

After hundreds and thousands of matches, it is very visible that your MMR system functions as this:

• Ten players que at the same time
• High MMRs are 10, low MMRs are 1
• The ten players qued look like this: 10, 6, 5, 4, 4, 3, 3, 2, 2, 1
• System tries to balance two teams from these MMRs
• Result looks like this -> RED: 10, 4, 3, 2, 1 --- BLUE: 6, 5, 4, 3, 2

This is not how the system works. The old solo queue worked like this, sort of.

You can find more details on matchmaking here:
http://wiki.guildwars2.com/wiki/PvP_Matchmaking_Algorithm

The current system is more likely to produce a result like this…
Red: 10, 5, 4, 3, 2
Blue: 6, 4, 3, 2, 1

Which could be argued is less fair, but really because MMR isn’t the only factor this is an oversimplification.

What he said about team splits looks exactly like my original assumption, if you didn’t notice that. The only difference is that he pointed out how the system creates a Team A and a Team B, where the Team B are lower MMRs players that have been out performing their current MMRs and are being given a chance to face a team of higher MMRs.

And to the guy who said axe wasn’t viable:
Watch the video on page #1 if you haven’t already.

well… i am stuck with the mmr i have. i go on a losing streak where i have teams that don’t attack target and flop about off point like headless chickens. After the losing streak its all good some wins some losses until bam back to the headless chickens flopping about off point vs teams with 2+ eles and premades. why?!?!??!?

i have an account in the eu and i don’t have this problem its quite consistent this one is in na..

Quality of matchmaking aside, even a poorly matched team is not without opportunity for you. When things are going poorly for me, these sentiments recenter me on the path of improvement and enjoyment:

“Every game you play there is at least one thing you can learn from to do better. Equally important, there is at least one thing that you did well, and you need to remind yourself to do it again. Maybe it’s big, maybe it’s small, but learning what went right and what went wrong after the game will influence the outcome of the next one. Just because you played seemingly well individually, but lost, and have some nice stats to support you doesn’t mean that you are right, but more importantly, it doesn’t mean that you can’t learn from it.”

“You’re playing solo queue for the wrong reason. You don’t play to gain elo, you play to learn to play better, and when you play better, your elo rises accordingly."

Source, worth a full read. https://www.reddit.com/r/heroesofthestorm/comments/33yi46/the_elo_ladder/

So I’ve had further thoughts on all of this
Arena net please read these!

A system that funnels for 50% win rate ~

I love the idea but after three years and 6,000 matches I can see that this does not work with Guild Wars 2 but the reasons why are not so obvious. Why do some people seem to get good matches and others seem to get poor matches? Why do some people rise above 50% and some stay under? Why do some people get bad lose streaks while others don’t experience it so often? Well I’ve figured out precisely why. Certain classes or build archetypes are just plainly better in conquest concerning the statistics of solo que joins and if they will be an asset on a team or not. Take a D/D Cele Elementalist as example who is joining as a solo que. It does not matter what comp he is with, he will be an asset to the team’s comp. So whether he is joining in on the Team A with the high MMRs or being forced in to Team B with the lower MMRs, he will still be a potent asset and create functionality within the team. This raises his statistical chances of winning matches. In contrast to the D/D Cele Elementalist, let’s say someone is joining as a D/P Thief “which is also a good meta build”. The problem here is that he becomes a liability if there is already a D/P Thief on that team or worse, there were already two D/Ps on that team, which pretty much means that team is going to lose because they cannot hold points. By joining solo ques as a D/P Thief, he has a larger margin of error for landing in bad team setups than the D/D Cele Elementalist does. It doesn’t matter if he is landing in Team A, three D/P Thieves is just a bad setup and should in all likelihood, lose to Team B. You could take every class and every archetype and put them in a list from top to bottom of which classes have the best probabilities of being able to land in a random team comp and still be valuable. At the top of the list, would be the D/D Cele Elementalist. At the bottom of the list would be all the special job builds that are only valuable when placed in actual premade comps to run specific jobs. They’re no good for solo ques if you are a player who is running for win rate in the long run. Might I add that this problem could be easily fixed with further game balancing!

Furthermore you have the problem of solo ques who get qued against premades. Let’s say that Team A consists of 5 stronger players with higher MMRs but they are all solo ques. Team B consists of intermediates but they are a 5 man guild premade communicating on TS3 with a well formed comp. Is this balanced? Mathematically identified by the server yes but the server does not take in to account voice chat systems or people who play alternative accounts for secret MMR smashing or simply that a higher MMR player is trying to learn some new spec at the time so no realistically it is not balanced and this is because there are factors that are not factored in with the MMR system.

In a nutshell: The MMR system aims for 50% win rate for everyone considering the factors that the algorithm has been programmed to work with but there are a significant amount of factors that exist in actual play that heavily tip the scales of which team will win, that the MMR algorithm is not taking in to account and this is exactly what the problem is.

I know that it is impossible to create a perfect MMR algorithm, no one is asking for a perfect system but there are indeed obvious changes that should be made to help clean up the problems which have been widely identified by the community.

I have no idea how the new league system is going to work but if I was to program it myself, here is what I would like to see:

• League is 5 man premade only. Not joined by duo que or trio que or even quad que, no it is 5 man teams. This shuffles all the players who want to play with guild voice chats in to one place.
• Unranked/Ranked just becomes “solo que”. No duo ques, no trio ques or even quad ques, no it needs to be all solo ques. This area of the game would be designed for more casual play that does no require the use of voice chat for success.

I think it is important to point out that merging solo que and team que is when all of the main problems began with people complaining about ridiculously bad match making. This is because merging the world of voice chat teaming with the world of more casual silent solo ques has created large factors in win rate success and overall team MMR that the current MMR algorithm does not take in to account and this is exactly what the problem is. If you simply put a hard wall between voice chat play and non-voice chat play, it would solve most of the problems, aside from general class imbalance issues.

I’d also like to add that creating a difference between unranked and ranked play was a good idea for multiple reasons but merging solo/team play has driven the more casual players out of spvp. This is not a good thing, for all of the reasons mentioned in this thread: https://forum-en.gw2archive.eu/forum/game/pvp/Two-different-types-of-competitive-gamers/first#post5103961

I know you try not to take in to account voice chat within the MMR algorithm but it is real and it creates enormous differences in player performance when used. You guys need to separate voice chat users from the players who just want to listen to mp3s and casual run solo. If you did so, the game wouldn’t feel so kitten demanding concerning the requirement to be on a TS3 every time you log in.

This thread is old but was recently cited by another PvPer on a thread about a losing streak. I can’t disagree with the OP’s comment above.

Class compositions can make a difference but that’s not the meat of the issue.

A scenario in terms of how MMR should work
Lets say we have a 4 man team that are going through matchmaking, picking up a pug to fill that +1 slot. They finally find Player A and they win. They go through several more games with that player and they win a couple games but lose every so often.

The same 4 man team drops Player A and picks up Player B. They’re winning a bit more and lose only occationally… BUT, the enemy’s team score is a bit higher when they won. As a matter of fact, when they lose a game, they lost with a lower end score than when they lost with Player A.
But thanks to the 50% win/lose and the RNG effect, the 4 man team won a bit more games than they did with Player A, so they stick with Player B a bit longer… unknowing that Player B is actually worse than player A because of the points they’re finishing with.

That’s typically the problem with the current MMR. Matchmaking isn’t giving enough credit to Player A because MMR is mainly based on the Glicko system – a system based on wins and losses. Player A is actually reducing the score gap but Glicko doesn’t look at that.

Glicko doesn’t calculate the enemy’s team’s points when you win a match.
Glicko doesn’t calculate how many points you fall short of 500 when you lose a match.
It doesn’t average out all of these game’s end score deviations to further calculate MMR
These Win end score deviation and Loss end score deviations (including the enemy team) should govern a player’s True MMR.

The “Odds of winning a match” is govern by MMR that isn’t well calculated.
To top it off, the rank points given is based on a 100 – 200 – 300 – 400 point scaling rather than a more detailed 100 – 105 – 110 – 115 rank point scaling. This should make for a more defining point system.

That’s the issue with MMR

@Saiyan:

Everything has its good things and its bad things. One thing it has been mentioned by Developers on the forums is that MMR is not equal to matchmaking. Matchmaking takes MMR mainly, but it’s not the unique thing it takes into account. It accounts for your roster for example, and maybe other things (although there may be not any other factor, we don’t know).

I have two problems with your suggestion then:

- First of all, not every build can impact the match the same way. Of course, if you are better than the enemy, you will make your team score more. So I’ll leave this as “not really a problem” in the end.

- More importantly, the existence of Forest and Legacy side objectives makes this certainly unreliable. It takes no effort to rush Lord or focus on Svanir/Chieftain, which can give you an advantage on points when a game is lost (i.e. you all know that it’s going to be a 100-500 for example, because you can’t win any fight).

The later problem is what would really concern me if matchmaking took into account your end team score, which can be inflated in some maps, giving you a “higher skill level” that you haven’t.

Or maybe I’m just overthinking it and I’m wrong! It can happen :P

@Trevor Boyer:

Hm, I understand your point, but you are suggesting several things I don’t (personally) like:

- You need to have a team as well as good voice communication (a problem in EU as we have multiple languages) to participate in Leagues, the only incentive and rewards we will have when HoT launches.

- Removal of Unranked/Ranked Arena for a Solo Arena. Which is acceptable, there’s no problem. But where are the queues for those people who just want to play it casually, with a couple of friends? They need to do hot join?

As for the rest of the post, I’d need to think more in order to give my opinión on it.

@ Poliator

I see exactly what you mean about casual duo ques. I often team up with one other good buddy and run duos earlier in the day before the rest of the guild begins logging in. Considering what you said in your response, it made me think of this:

Suggestion – Advanced LFG System for team ques
Say you are running a duo que. The advanced LFG system would allow you to see a detailed roster of other players who are currently queing. It would show solos, other duos, trios, anything and everything that your party has room for. The system would show rank, total matches played on those player’s accounts, total matches played on the class they are currently on and of course a small message per party listed "looking for bunker and peeler, here is our ts3 address, ect.. ". The new team que option would actually make you select your teammates before you can que. This would would be an excellent idea for three reasons:

• Allows players to choose their que wait times. If you want to pick up the first 3 solo quers you see, you can get in a match quickly. If you want to wait to pick up a good 3 man que to your duo and discuss a TS3 to enter, you’ll have a stronger team but a longer que time. This would eliminate all reasons and basis for kittening concerning wait times as it would now be your choice.
• Allows players to choice their own team comp and experience level of players joined. No more kittening about being placed in dysfunctional teams as it will now be your choice, not the MMR algorithm’s choice.
• After you have built a team, the current MMR algorithm would be satisfactory for match-making 5 man premades vs other 5 man premades. This is how team pvp in conquest should have worked from day 1.

Then for those who don’t want to worry about team comps, voice chats and organization in general, they can have the old solo que back. Only solos go in and only solos go against each other. The old solo algorithm could be tweaked slightly and it would be satisfactory.

The key fundamental of what I am saying should however, not be misunderstood: 5 solo players that do not want to use voice chats should not have to fight players that are using voice chats and those solo players should have the right to an environment where this is not an issue. Considering that Guild Wars 2 doesn’t even support the voice chat in-game, it would only be fair.

Evan Lesh.3295:

Saiyan.1704:

That’s typically the problem with the current MMR. Matchmaking isn’t giving enough credit to Player A because MMR is mainly based on the Glicko system – a system based on wins and losses. Player A is actually reducing the score gap but Glicko doesn’t look at that.

Glicko does support score-ratios, but in our tests we found it causes MMR to move so glacially slow that you end up having more bad matches because your MMR can’t keep up with your true skill level. This is due to the fact that most losing teams still get around 300 points.

We could have a configurable range to rescale the win ratio to help mitigate this (per game mode, unfortunately). We would have to run simulations first, though, because making any changes that affect glicko can be very dangerous.

For curiosity Evan, you tweak something in the matchmaking formulas for the pvp leagues or it is basically what we have now?

I think after 2 beta seasons you have more data to stabilize the matchmaking better.

perhaps changing your account name may help you?

Evan Lesh.3295:

I’ve just posted something relating to this original post. The bugged behavior worked as such (assuming solo queue only):

Players ranked by MMR: 10, 9, 8, 7, 6, 5, 4, 3, 2, 1
Team1: 10, 4, 3, 2, 1
Team2: 9, 8, 7, 6, 5

Well yeah, that’ll cause some issues… x.X

Evan Lesh.3295:

I’ve just posted something relating to this original post. The bugged behavior worked as such (assuming solo queue only):

Players ranked by MMR: 10, 9, 8, 7, 6, 5, 4, 3, 2, 1
Team1: 10, 4, 3, 2, 1
Team2: 9, 8, 7, 6, 5

Now probably everyone will think they were always the number 10.

Evan Lesh.3295:

I’ve just posted something relating to this original post. The bugged behavior worked as such (assuming solo queue only):

Players ranked by MMR: 10, 9, 8, 7, 6, 5, 4, 3, 2, 1
Team1: 10, 4, 3, 2, 1
Team2: 9, 8, 7, 6, 5

Ouch. Just adding those up you see you’re pitting~

20 vs. 35

If there’s any real spread of skills on the field that has to be AGONIZING for the top player. He’s being asked to make up for FOUR sub-average players against 5 players average or better.

I would have guessed the sorting was more like

10, 8, 5, 3, 1 vs. 9, 7, 6, 4, 2

Team A gets the top player, Team B gets one more ‘point’ overall. Both teams display a spread of skill levels.

Is the fix to this bug going to be immediate or will we have to wait a while?

Evan Lesh.3295:

The way it is intended to work is the next biggest roster gets put onto the smallest team so far, or the next highest MMR roster gets put onto the team with the lowest total MMR.

The pattern I’m describing proposes “Worst player picks first.”

So 1 forms Team A & grabs 10 for team A
2 forms B & grabs 9 for team B
3 joins A & grabs 8 for team A
4 joins B & grabs 7 for team B
5 joins A
6 Joins B

I think if you approach it as tending to the worst first instead of the best, you may get more equitable results.

You also get an array where if best fights best, second best fights second best all the way down to worst vs worst you see
10 vs 9 (favors A)
8 vs 7 (favors A)
5 vs 6 (favors B)
3 vs 4 (favors B)
1 vs 2 (favors B)

Which gives you closer challenges and best player suffers only a tiny handicap. Whereas the current array give you~

10 vs 9 (favors A)
4 vs 8 (hugely favors B)
3 vs 7 (hugely favors B)
2 vs 6 (hugely favors B)
1 vs 5 (hugely favors B)

I don’t see how you can expect anything but a blowout in that environment. Number 9 only needs to tangle up 10 briefly, even losing to the top player at every clash and the rest of his team eats Team A alive.

Evan Lesh.3295:

This was caused by a bug, not intentionally; we didn’t want it to happen this way :P

All good. I’m glad it’s fixed now. I’m pretty sure I’ve been #4 in a few of those match-ups where there’s one guy on my team who is kinda the top buttkicker of the whole match (#10) and is screaming at everybody except me. I know I’m not great, but I don’t think I precisely suck either .

https://forum-en.gw2archive.eu/forum/game/pvp/Updates-to-Matchmaking/first#post5524291

~ Thank you

so i didn’t imagine it!!!! read the dev post about updating mmr =D

This is the thread that started it all.

The OP explained what I have observed in-game around the test ladder season. It’s also my go-to thread after feeling let down by how bad certain matches ended up. I referenced this a couple days ago to a pvper caught in a bad streak not knowing what the heck was going on (most of us have been there)

It was just absolutely epic that Evan announced earlier today that one of the bugs they’ve discovered and are currently working sound identical to what was explained here.

Props to the op OP of this thread.

OMFG – is this ture? Something´s going to change?

But isn´t it already too late?

This will be nice to have in PvP, but we still have premade teams stomping over solo q’ers all the time. How about doing what LoL does and only allow duo and solo queues in ranked?

Pimsley.3681:

This is the thread that started it all.

The OP explained what I have observed in-game around the test ladder season. It’s also my go-to thread after feeling let down by how bad certain matches ended up. I referenced this a couple days ago to a pvper caught in a bad streak not knowing what the heck was going on (most of us have been there)

It was just absolutely epic that Evan announced earlier today that one of the bugs they’ve discovered and are currently working sound identical to what was explained here.

Props to the op OP of this thread.

THANK YOU

For linking this post on an active thread that Evan participated in. That may have led his team to finding the bugs or at least knowing what to look for.

Teutos.8620:

Evan Lesh.3295:

I’ve just posted something relating to this original post. The bugged behavior worked as such (assuming solo queue only):

Players ranked by MMR: 10, 9, 8, 7, 6, 5, 4, 3, 2, 1
Team1: 10, 4, 3, 2, 1
Team2: 9, 8, 7, 6, 5

Now probably everyone will think they were always the number 10.

Too bad its not only the 10 getting screwed but also 4,3,2,1 as well and really even for the winners as it makes for crap matches. Its an awful bug and really get your nose out of a-nets butt this is a horrendous bug. And for like a year we were told stuff like that wasnt happening. They should apologise to us fixing is not enough.

the main Problem i face at the Moment is the following. I did try out several classes now and quite a lot of build and i found out that if you are one of the higher ranked Players in solo que you get in Trouble if you Play Support builds -.-
I did love playing bunker Guardian but that is almost not playable in solo que, what does it matter if i can Support my whole Team but I end up in matches. where they are like, oh nice Guardian is mid he keeps them busy we chase now with 3 man that thief over the whole map -.-
I almost never had hat Problems with my Ranger or Mesmer. their it was like ok one guy defend that Points i nuke him take the Point let that 3 guys chase the thief we got at least the Point.

and if you have a high MMR you can be almost sure that you get almost every match up guys which have no clue that you try to Support them and the first Moment after you did buffed them and healed them up they run away from battle to next Point…. For me it was super annoying to Play Guardian in solo que because i had to Play a build i did not like and was not so good at(but because i did not depend on my Team it was more efficent in the end … )[I love playing Support Guaridan but if i wanted to win in solo que i had to Play Meditation Guardian which i dont like so much]

Exactly insomnia. That feeling of being stuck in a 1v3 since your teammates chase that random thief or w/e across the map.

The annoying thing is that it happens almost every game during certain stretches.

insomniaproject.1348:

..all true..

I loved to play bunker, too. But after these matches of hate, Niflhel for example, i quit playing supporter in an unorganized match. These matches on Niflhel when u start to go mid, seeing 2 players battling the chief. And certainly u die mid pitifully and humiliating after over 6000 games played.

Sometimes i write in the chat on Niflhel map: .. If i see 1 or 2 player fighting mob at beginning of the match, i go instantly afk or log out.

Sometimes it works, sometimes it doesn´t. If it doesn´t work, then it becomes a disgusting match and raging is inevitabel cause this is too much after 6000 matches played.

Dev team please read! as this is the final and best feedback that I can give you before the HoT release.

The #1 biggest problem with the MMR algorithm is that it does not factor in the match type that it is balancing for. Sure it does it’s best job to numerically crunch a Team A and then a Team B. The numbers may look even to the system but that system is not understanding how a conquest match is played and won.

Let’s take the “Glicko-Hell” example of: 6, 5, 5, 4, 4 against that sad 10, 4, 3, 3, 3 assuming 10 is elite MMR range. This is Team A average of 24 vs. Team B average of 23. Numerically this looks balanced but it is far from balanced because of how conquest is actually played and won. Even though the Team B has the elite player, he can only be on one given node at any given point in time. During this match up, the 10 MMR could win every 1v1 and team fight that he engages in but his low MMR team mates will be slaughtered by the intermediate MMR players on the other two nodes and will constantly be back-capped. Of course we all know that conquest is won by holding 2 nodes and because of this, that 10 MMR to balance the low MMR team may look numerically balanced to the MMR algorithm but it is indeed not balanced at all. Not only is it not balanced but it is actually completely lopsided to the point that Team B never had a chance of winning to begin with, due to how three node play actually works.

A better split concerning those 10 players of: 10, 6, 5, 5, 4, 4, 4, 3, 3, 3 for a conquest match up would look like this:

• Team A – 10, 5, 4, 3, 3 = 25
• Team B – 6, 5, 4, 4, 3 = 22

Even though it places Team A with the elite 10 MMR as a 25 team total vs Team B with a 22, this is actually more balanced concerning three node play. In fact it is as close to perfect as it could be. The 10 on A will surely conquer the 6 on B but the 5, 4, 4, 3 on B is slightly more capable in team fights than the 5, 4, 3, 3 on A. Of course this is disregarding team comps and class counters but for now let’s safely assume for the sake of argument that both sides are playing mirrored team comps. It is indeed an oversimplification but it is the best oversimplification I can write in place of some thesis that you’d rather not read.

Ultimately, yes the numeric MMR player range matters a great deal for the algorithm to attempt to balance numerically but there is that factor of “how conquest is actually played” that needs to be identified and factored in by the MMR algorithm for match making purposes. In a nutshell ~ the system needs to identify that even though Team A – 6, 5, 5, 4, 4 = 24 vs. Team B – 10, 4, 3, 3, 3 = 23 looks balanced numerically, it is not balanced for three node play. The system then should attempt to ping pong players back and forth until it finds two different team splits where the numbers will be the most balanced on two nodes, most of the time. This would be an excellent system for identifying with how three node play actually works.

The 2nd biggest problem is that the MMR algorithm does not identify the difference in the values of match altering weight concerning high and low MMRs. This is actually difficult to explain, but here is my best shot:

• Take a team of 10, 5, 5, 5, 5 vs. 5, 5, 5, 5, 5. The 10 will usually create a win for his team but the 5s are certainly not helpless against him. With good target calling, positioning and correct class counter engagement, the 5, 5, 5, 5, 5 actually has a chance of winning that match. The 10 is certainly an asset but he doesn’t guarantee a victory. Again, that 10 MMR can only ever be on one node at any given time.
• Now take a team of 5, 5, 5, 5, 1 vs. 5, 5, 5, 5, 5. The team with the 1 MMR may as well be in a 4v5 and being in a 4v5 in three node play is almost always an immediate loss unless your four players are significantly better than the five man team. The 1 MMR will die almost immediately in any combat engagement and he won’t understand conquest rotations. I have actually heard many players claim that having 1ish MMRs in their match is actually worse than being in a 4v5 because the 1 MMR feeds the other team kill points. I can’t say I disagree.

So in other words, a high MMR is not as much of an asset as a low MMR is a liability. Throwing the 10 in to the world of 5s is much less disruptive to the outcome of matches than throwing the 1 in to the world of 5s. Most players would certainly agree with me on this and this applies to solo ques as well as team ques. The MMR algorithm concerning match making needs to identify that low MMR players are worth more match altering weight than high MMR players in three node conquest play and it should understand how to balance that weight a little more accurately. This is again, an oversimplification but it is the best I can do without making it a tedious read.

The above two problems with match making are critical errors in the factoring of the MMR algorithm and they are what bring in the real complaints that we see in this forum, concerning the match making. If the above two problems were addressed, wins/losses would begin to look more like 500-400 and a lot less like 500-100. People don’t so much mind losing when it feels like a good match. The amount of Ragers/Leavers/AFKers or AKA: Toxicity, would be greatly culled if we could just eliminate the unreasonable and impractical matches as detailed in this post.

To the dev team:
~ Thank you for reading my text walls.

I said this in some other thread ,leagues are going to be pain in the kitten if they dont differentiate number of games played by classess and imho player with MMR score 10 should never ever have to play with player with score 1<5.

What is happening now will hapen again in new league since evry1 will start from scratch ,
meaning you will have first week of league with teams
A: 10.8.4.2.1
B: 8.8.8.2.2
where players with score 2 or 1 will break or make the game…
Then you will have players elevated or draged down by this and ,very very frustrating week or month of first league.
In general i dont thin leagues will solve anything if matchmaking will still use current system.
PPl should be asigned to leagues acroding to matches they have played with a class and a score they have, not every1 from 0 ,or nothing will change.