Guild Wars 2

GhOst.4019:

Bukkebruse.2810:

Justin ODell.9517:

GhOst.4019:

What is the grace period? 30s? (Fair number)

Yes, it is 30 seconds.

Don’t you think that’s a bit too low? Keep in mind that when someone dcs during a game they have to load two maps (hotm and the PvP map)
I’m playing on a semitoaster and I’m fairly sure I can usually manage to get back into the game within ~20 seconds, but I know there are people with far worse machines than me. I’d recommend that you increase the period to a minute rather than 30 seconds.

Technically you are loaded into the map after about 1-2 seconds. Just because it doesn’t look like you are there yet, you still are. You machine is loading the graphics still but your character is back in the match faster than it looks like. Only takes maybe 10-15 seconds to get back in. The ~20 seconds is just a visual thing for you.

This is correct, and was part of the reasoning behind the 30s grace period.

GhOst.4019:

What is the grace period? 30s? (Fair number)

Yes, it is 30 seconds.

Anti.9156:

what happens if someone gets a DC and is not comming back, but my team still wins the game?

If you still win, and the losing team doesn’t have deserters, you will receive credit for the win.

If you DC and do not reconnect within the grace period, yes.

This is to discourage groups from leaving the match early to invalidate a win for the other team.

Lux.7169:

Do we need to actually play a game to let the desertion losses take effect before the ladder season ends to receive credit/be eligible for the test-season rewards?

I planned on not playing this week but I’ll go home and hop on if I need to.

No, you can risk your position and not play to preserve your w/l ratio.

These desertions, while not shown, were already included in your ladder points.

Vapour.7348:

This is erroneous because I did not disconnect at all in the last hour and yet the system posted that I had 12 losses. The reality is I had 3 wins and 1 loss. How did it come to 12 losses? Prior to this, my win/loss ratio was closer to 60%.

I don’t understand why the system is so punitive, even to those who play fair and honest.

The 11 losses unexplained losses are from the 11 desertions you had over the course of the ladder. They are not desertions from the last hour.

REGGADO.1237:

Well I had +3win-21lose in https://leaderboards.guildwars2.com/ too, but didn’t made screenshot. That makes 25games per hour. So my games last 2minutes, lessmore than que window pops.

It rolled your previous 31 desertions into it, but this is a recent change in what we display, so that’s why it had the sudden jump.

Correction: 21 desertions, 10 of those were before the test ladder season started.

Morwath.9817:

Whenever you get DC you get -3 points.

Up to -3 points and as little as -1 points. It depends on your odds of victory.

Since you don’t know your odds (yet) you might as well act as if it’s -3.

The leaderboard now rolls desertions into the losses column.

Try to look at it from the eyes of a beginner guys.

The NPCs award points, and give you a buff. Who wouldn’t want that?

It takes a higher level of understanding of the both the map and conquest mechanics to get that securing mid is important.

How would you guys suggest we go about teaching this to players?

Wonderly.1324:

I kind of see where Solstice is coming from. Dishonor is necessary if there is at least even one pug member, but it can be argued that the system should always be disabled for full teams of five because it is not necessary. After all, if you rage quit on your friends (or even if you constantly d/c on them) that’ll act as it’s own dishonor system (no one will want to q with you anymore).

I see it too, but desertions on the other team will hurt as well if we decide to ignore the results when the losing team has desertions. I could easily see some group of players purposely leaving to deny the other team a win.

If this problem is being caused by another person still being queued as TPMN.1483 suggests (and that sounds plausible), then it will be fixed on Tuesday.

Daishi.6027:

I recall a thread regarding specific class bugs, and a dev confirmed they were either complete or unable to make it in, for the upcoming feature patch at the time.

Content bugs are in a whole other realm, and must be bundled with a game update.

Laserbolt.6731:

“I sympathize with the feelings people have, but feelings aren’t dictators of reality.”

No, but blowouts are. What about solo people mixed up against grouped players. That confounds the matcher when there are not all that many people in the queue.

I completely agree. We still have places we can improve things, but this isn’t related to the MMR formula, but more with matchmaking itself.

Khisanth.2948:

Not because ANet decided on a 3 month update schedule?

Edit: … or was that 5. Either way, really long time especially for smaller stuff.

The new backend servers can usually be updated out of band with the rest of the game, but we generally try to do in sync with the rest of the game to minimize the load on QA.

The old code was a pain, and couldn’t not be updated so frequently.

Neither have/had any barring on balance changes though, those have there own separate concerns.

Exedore.6320:

If a bunch of people are around the same MMR, but some should be higher and others lower, the outcome over many iterations still ends up looking correct because it averages to 50% ratios.

I’m not 100% sure I follow what you’re trying to say here. Maybe if we made completely random teams and MMR was ignored it would average out to 50% over many iterations. That scenario doesn’t match our system though, we place by MMR and that MMR adjust over time based on outcome. This isn’t random sampling.

Exedore.6320:

Although the accuracy of predictions with wider MMR ranges seems to imply that MMRs are correct, it doesn’t. As long as you have some people higher or lower towards the directions they should be, it helps offset the people with incorrect MMRs when you lump them into teams. In theory, that should eventually settle, but if you keep injecting new players, it stays chaotic.

This seems to be going in the other direction, and only would be true if players stayed in the same team for every iteration. Our system, even though not a random sample, often generates matches where the same person may be the highest MMR or the lowest MMR in any given match.

Exedore.6320:

I have to agree with the feeling that MMR calculation isn’t correct or there are many outliers. While I see many opponents and teammates that I would say are roughly around my skill level, I often that one really bad player on either team that is nowhere close. Problem is, that one outlier has a huge impact on how even the match-up feels.

I sympathize with the feelings people have, but feelings aren’t dictators of reality. We humans have many cognitive biases that cloud our judgement that we must keep in check when aiming for objective analysis.

Looking into it. Who were the other players?

Laserbolt.6731:

Is it possible the Match Prediction Algorithm/MMRs can only accurately predict at the extremes? And in the vast middle ground it often fails to predict unbalanced matches?

No, our mean rating diff + std deviation * 2 is still less than the starting devation for players. So I wouldn’t call that extreme.

As for not predicting in the middle ground… that is sort of implied in the process. If two players have the exact same MMR then it should be a 50/50 odds of winning for either player. So it is expected that we have a lower prediction accuracy the more even the match is.

Laserbolt.6731:

How did your simulations go against past games?

Very well, and its doing just as well with the games since the ladder started.

Mindtrick.5190:

Lol holidays are for college kids, most adults worked. People who need breaks from work don’t really love what they do. “When you want to succeed as bad as you want to breath then you will be succesful”.

My employer is very generous when it comes to paid holiday leave. I’ll also remind you that it’s a Sunday and I’m here looking at data and answering questions.

Csele.1647:

The MMR system doesnt seem to work at all in GW2 .. compared to other games that i play , where you only play with people relatively similar to your skill level.

I’ve been looking at the data, and by my estimate it disagrees with what your saying. Not only are wide differences in MMR rare, we more accurately predict the outcome the wider they become.

This doesn’t mean we can’t find ways to keep improving it more.

Csele.1647:

In gw2 you play with all ranges of skill people it is almost never balanced.. games end up 500 vs 100-200 most of the time.. and with 2 thiefs or 3 rangers.. i personally dont see any diference at all before and after the implement of the MMR.. everything that was suppose to be fixed with it.. wasnt , the pvp balance of teams is as broken as it always have been.

A lot of factors are in play when you look at the difference between scores. Out side of skill you also have issues like morale, mechanics that encourage snowballing, lack of comeback mechanics, team composition, personality conflicts, human error, etc..

Csele.1647:

for example i never get “bronze league” people in my team, in starcraft because iam 4 leagues above them.. , leagues " or ranks" mean something there . same as counter strike , and many other competitive games .

I’ll take a look at your account to see if anything stands out.

Daishi.6027:

It’s clear on a regular basis players are not pleased with the method or gaps between updates and fixes.

It’s only been this long because of the holidays in between updates.

mistsim.2748:

hey Justin, anything in the works to prevent 3 rangers on one team vs. none on the other, or 2 engies + 2 necros vs. none of either on the other team type stuff from happening? i.e. better prof distribution. this is by far the greatest flaw of the matchmaker atm, because not ever prof can fill every role well. for example, a team with 2 thieves and 2 power rangers have no point presence, usually results in loss. stuff like that. and will the matchmaker ever be intelligent enough to look at the spec as well?

There already is something in place, it’s just being overwhelmed. We also had to lower the enforcement of duplicate professions because it was overshadowing party size scoring.

By spec do you mean build? We considered adding amulet as a matchmaking variable, but since we want people to be able to change builds in that pre-game window we decided against it.

ArrDee.2573:

Justin ODell.9517:

Yes. I have fixed this bug and the fix will be LIVE on Tuesday.

Is there a reason that for a glaring bug you say has already been fixed, you wait until patch day to release it!

Wahoo! Bye frands!

Protocol. The bug isn’t critical as it doesn’t prevent play. It’s more annoying than anything else.

Solstice.1097:

Apologies if this is already announced, but can we confirm the bug preventing us from actually leaving queue will be fixed prior to any penalty for denying queue pop?

Yes. I have fixed this bug and the fix will be LIVE on Tuesday.

Urejt.5648:

yea, arenanet is trying to make system, while blizzard is doing it.
1. Blizzard does not allow 4 premades vs 2 soloquers.
2. Current gw2 dezert system puts 1 min ban on ranked, while blizzard give 20 min ban becouse player is forced to play full unranked game.

1. Their wording suggests they do the same thing we do. That is, they will try to avoid matching premades and soloers, but will if certain conditions are not present.
2. You actually get a 16 minute ban (when dishonor is enabled) for deserting the first time. It escalates from there. You only get a small ban if you decline a queue.

Cynz.9437:

race condition o.o?

Basically, it’s a problem caused by two things that are trying to happen at the same time, but the that one wins (happens first) isn’t the one we expected to win.

http://en.wikipedia.org/wiki/Race_condition

I looked up the time period from the screenshot and found the match.

It doesn’t look like you were ever added to a second match, so this may be a race condition bug when presenting the data.

Cynz.9437:

Ever had an idea that people maybe don’t want to play particular classes nor will be good at it no matter what…? Especially if their mmr is actually relative high due to how well they play their main class?

Yes, we are aware of this. This is why the server supports blending in a profession specific MMR. It’s currently disabled because we want to give those profession specific MMRs time to settle. We’ll likely enable them after this test season.

It’s true some people just don’t like a certain profession, or their play style doesn’t fit well with it. It’s also true that some people try a new profession and find they like it more and have an easier time with it.

Rahktaahl.3082:

So you’re telling me that it is a complex problem where blame cannot be placed with certainty; but you are willing to give dishonor those who disconnect.

Regardless of who is to blame, the disconnection ruins the games for others. Also, we cannot tell the difference between a rage quit and legit disconnect. Because of these two things we feel it is important to minimize the impact to other players.

I’m sorry if this happens to you frequently, but looking at the data it’s still fairly rare. We may add a grace period for disconnects, but you will still get dishonor if it’s beyond that time limit.

Hi rwolf, thanks for sharing.

What you said prompted me to look into your record. You’re currently running 35w/32l for the ranked arena, and 131w/141l in unranked, both of which are healthy w/l ratios.

Was it thief in particular that you were having difficulty with? Maybe someone here can share some tips for others having trouble with the profession.

Zenith.6403:

That fits the theory of the ranking system. I’m surprised that the ranking differences of individuals in the same match are so large. It’s a factor of queue times and player volume, but also noise caused by secondary goals like preventing multiple classes on same party.

It’s pretty rare, and you pretty much nailed the reasons why it happens when it does happen.

Zenith.6403:

Some people are like that, insecure about their own performance. I’ve seen dungeon groups on LFG tool have description like “5k+ AP points only” or players asking others to ping their gear as if a few points of damage made any difference in the grand scheme of things.

Rating points don’t tell other players anything. If the concern is “is this guy a good player?” after seeing that he has rating of 1500, what can be said about his value? The matchmaking already balances teams based on ratings of individuals. The logical choices:
1) 1500 is an accurate representation of the player’s ability.
2) 1500 is underrated ranking, because the player has been on a streak of losses and is likely to win this match.
3) 1500 is overrated ranking, because the player has been on a winning streak and is now anxious about losing this match.

So seeing individual rating doesn’t help predict the match, if the match is fair.

That being said, it could very well be just simple personal command like /rating or /rating_volatility on chat for those curious individuals who seek it to improve their own game.

I have to disagree with this. The whole point of MMR is to gauge the skill of players. It does this by reacting to the result of the games they play.

If this is accurate, and I believe it is, it would have a very good chance to predict the outcome. For example, right now we have stats that show when the MMR difference between the average of each team is approaches or surpasses 200, we correctly predict the winner 95% of the time. That difference of 200 is less than 10% of the spread between the top player and bottom players.

Zenith.6403:

What are those problems of showing MMR?

Promoting harassment, elitism, and a toxic environment where artificial numbers are put over playing well together.

I understand the desire. I believe there may be some way of showing standing without the issue above, but I haven’t found it yet. Hopefully some one will soon.

Rangelost.4857:

It could be interesting to display the MMR, even without numeric values, in a way that players can understand it.

While I love your mockup, I’m not sure how that would avoid the problems associated with showing MMR.

Zenith.6403:

- I would really like to see this kind of information available on a website. Would it be possible to add it to API for third party developers?

While the official policy is still to not show MMR, we do have on our backlog providing a public API for PvP data.

Morwath.9817:

I wonder, how it’s possible that in last 7 games, when I’ve done 4-3 and lost 10 lb points without disconnects, char switching etc. Even one game was 3v5 (3 in my team, lol)…

It shouldn’t be. The max points from 3 losses would be -9. I’ll take a look.

Update:
I found nothing wrong with your account. Could have it been just an issue with the leaderboard site only updating once an hour?

NeXeD.3042:

actually it was with a power necro on my main account. this acct. i can pretty much only post once a day on this thing.

At least you made it count.

heartless.6803:

I have a question on matchmaking, it doesn’t seem to do anything for me. I’m not that good in structured pvp so I usually end up at the bottom almost every time. Sometimes I win because the rest of the team is good, but rarely am I actually much help.

Why isn’t match making system placing me against teams more of my own skill level? Do teams like that even exist?

It looks like you just recently started playing, (I know duh, you said that but I had to verify), which means you had a default rating and wide deviation. It takes some time for the system to place you, but it looks like it’s nearly done now so you should start seeing more stable games.

heartless.6803:

The more I’ve done structured pvp the more I found out that only the really good players actually do it. Which means for newer players learning pvp, it’s entirely a waste of time. There is no reliable methods in game for learning how to get better. Is this something that is actually a concern to you guys?

Yes, it is one of our concerns. We have been actively discussion this issue for some time internally. It is one of our priorities.

Solstice.1097:

I just swapped toons while I was waiting for queue. When i logged back in the queue had popped and the accept/cancel screen was up with about 12 seconds left. i accepted.

since the class i played as was different than the class i queued as, does this mean it is

A) i get penalty even though i accepted the queue as the class i was on
B) i unintentionally discovered a possible exploit of the (future) profession balance system
or C) somehow the queue popped after i was logged in as that character but still in loading screen which was probably about 8 seconds long and that’s why i saw 12 seconds left on the accept/cancel screen

??

A) No, you only get a penalty if you fail to enter the instance or leave it.

B) There is a bug, which will likely be fixed on the 13th, that prevents players from really leaving the queue. This includes the automatic queue leaving that happens when you log out or change maps.

C) Possibly.

TL;DR
You can change professions due to a bug, as long as you do it before confirming a match.

Dixo.3876:

Why publish win and loss rates then. People near the top of the leaderboards get harassed all the time; especially ppl with lower win rates.

Basically… because it was already there. If this is a source of harassment it might be worth looking at replacing or augmenting it with other data.

oneply.9586:

the one weird thing i get is that i usually have to select my choice (both map and accept/cancel) 2-5 times before the system accepts it. its the only place in the entire game that i have run into a problem like this.

I’ve seen this before too. I think it’s because the UI doesn’t register clicks while its being updated (which it does whenever someone else’s click is registered). I’ll ping Evan about it.

Solstice.1097:

Thanks, however what i meant was if you are in a pre-made team of 5 people who are CHOOSING to play with you, then the dishonor system itself isn’t necessary to apply to that scenario. We don’t need dishonor in a scenario where you get to choose your teammates.

You and your friends may decent and honorable people, in which case the system wasn’t designed for you and shouldn’t affect you.

Torin.7129:

Justin you are saying:

-We don’t want people focus on MMR, and we don’t want people to shame and harass people with lower MMR.

Is this a joke? Everyone want to focus on self MMR. We want to know where are we standing at the moment, the leaderboard doesn’t show it.
Don’t we get matches with people with nearly same MMR? Then why do we want to flame them? It has no sense.

No joke. I understand you want to see where you stand, and we looking at that. As for why some people do what they do, I have no clue.

Torin.7129:

I think increased queue time is much better than unbalanced terrible games. It’s simply not fun to play matches vs 4 engis or with 3 bunker guards. Or give us the option to do something before the match is starting. I mean if i can’t change character, i want to change build. But 1 min isn’t enough. We need trait and skilltemplates to change builds.
Or fix this MMR based matchmaking and do not let the system make double professioned matches.

I hear you. We’re looking for this sort of feedback. We’ll likely start a thread post season to discuss with the community potential changes.

Solstice.1097:

also, part of me feels that there should be no such thing as dishonor for a full team of 5 pre-made. if their ally DC’s much they can choose not to queue with that person? the intention of dishonor does not really apply to a pre-made team imo.

This would not be the case. Only the leaver(s) would get dishonor.

Morwath.9817:

For me its a bit longer: You DC, you restart entire client (because it won’t you reconnect), you authorize login, you login, you start loading back, takes sometimes a bit more than just 30 seconds.

It’s certainly possible that some number of people wouldn’t be able to make some limit, but you also have to balance it against the impact it has on your team’s chances of succeeding.

The longer the grace period, the more potential damage to the teams odds of winning.

Errant Venture.9371:

It’s simply not reasonable to expect a team playing down a person for 2 minutes to have any chance of victory.

I tend to agree. I’ve seen people suggest a grace period of 2 minutes, and that seems far too long to me.

I think something like 15-30 seconds is better, keeping in mind that you reconnect to the game before you finish loading and you may persist for a few seconds before the game registers the disconnection.

style.6173:

It would be better if you ran another check AFTER the map selection.

Map selection is so quick, I’m skeptical about how much that would improve things, if at all.

I’ll look at the data and see if I can figure out just how often people confirm but leave during the vote. As well as how many games start with a player missing.

Errant Venture.9371:

So are we only using profession-specific rating? Or only core rating?

Only core.

Errant Venture.9371:

Can you look at my account and tell me if over the last week people i’m paired with have especially low MMR or my MMR has tanked? I’ve had very few matches (less than 25%) where more than 1 other player knew how to rotate, contain, or even hit T to follow pinged target,

It looks like you had a very high rating and a wide deviation… which suggests to me that you may have been hit by one of the early bugs that distorted your rating. Since then it has been stabilizing downward and is now near your previous Solo/Team arena ratings, which is still above average.

You should start seeing more consistent games.

Flumek.9043:

What is “blending” ?

Does it mean all my alts share same MMR with my main?

Or that my main is great, but my alt which i only played a few matches with and could be considered beginerish – but is “upleveled” because im obviously a great player with my main.

The latter.

We keep a core rating per arena, and a profession-specific rating per arena. If blending is enabled, we would combine your core and profession ratings and use that for matchmaking.