114 replies to this topic

[Egomane]

mogs01gt, on 19 January 2015 - 08:08 AM, said:

That link says nothing about MM and ELO. It's referencing 4x3 MM.

Karl implemetented a complete rewrite of the MM. I'm to lazy to look for more current ones. I only posted the link to show that the other one is outdated by a lot.

If you want to know more about the MM, you can always ask Karl directly. I have linked the thread where he answers community questions, several times now in various Elo and MM discussions, but I'll do it once more.
http://mwomercs.com/...courtesy-phone/

[SgtKinCaiD]

Egomane, on 19 January 2015 - 07:55 AM, said:

Not going to be rude, but that post is from before the rewrite by Karl last year. The current matchmaker was released some time after this post:
http://mwomercs.com/...chmaker-update/

Not going to be rude again but if you remember correctly there was a first attempt to rewrite the MM in order to include the 3/3/3/3 and ELO buckets back in Feb/March 2014 but PGI had to roll it back due to atrocious waiting time. Then TheB3RG (yep after TheB33F comes TheB3RG !) worked on it to bring the 3/3/3/3 on again and some more stuff like IS vs Clan matchmaking. We still got the same old MM back from 2013 especially for those kind of corerules.

Anyway, when rewritting the MM, PGI ditch the ELO buckets although it was a real improvement not the 4*3. Sure the MM needed something to prevent silly drop like 12 heavies/assaults, but it was not the root of the problem.

It's unacceptable for a MM which is supposed to create fair match to have such disparity between teammates. It's getting old to have a match where several of your teammates die in the first 3 minutes due to bad positionning / bad builds, in short a lack of basic knowledge of the game. It's getting old to lose a match and see a table score where you outdamaged and outscored more than a full lance of your teammates ... It's getting old to win a match where half of the opposing team was chasing a single locust. etc. My teammate can count on me, I don't see why the opposite shouldn't be true.

Honestly did you read the number crunching posts of Kiiyor about (C) trial mech ? It's ridiculous that trial mechs are still getting pulled in regular match.

And the supposedly low player count is a false excuse : you still find a match under 3 minutes whatever the queue you're playing in.

But the MM is like the ECM/Ghost Heat : PGI knew it could work better but it's good enough for now because they have more important thing to do. But if the MM is still in this state when MWO goes to Steam, I predict a huge disappointment for PGI.

[Screech]

Last I heard from Russ regarding the Elo on solo queue was that the spread was mostly within a range of 100. When did this change? Would like to read about it.

[mogs01gt]

Egomane, on 19 January 2015 - 08:23 AM, said:

Karl implemetented a complete rewrite of the MM. I'm to lazy to look for more current ones. I only posted the link to show that the other one is outdated by a lot.

If you want to know more about the MM, you can always ask Karl directly. I have linked the thread where he answers community questions, several times now in various Elo and MM discussions, but I'll do it once more.
http://mwomercs.com/...courtesy-phone/

I've read through ever linky on in that thread. None of them refutes that MM could group people with extreme ELO differences. These links suggests that is an accurate statement.
http://mwomercs.com/...14#entry3294614
http://mwomercs.com/...79#entry3322579

Edited by mogs01gt, 19 January 2015 - 09:11 AM.

[Theodore42]

Rorvik, on 18 January 2015 - 10:45 PM, said:

If that is true, then it CLEARLY shows that the MM is NOT working.

I disagree with the assessment that if there are equally skilled players on both sides that it would naturally lead to steam rolls. If you have 24 roughly equally skilled players on both sides, then losing one or two Mechs should not be the end of the match. I have been in matches where teams (either mine or the other team) have managed to come back from almost three Mech deficits, and those matches usually ended up being close, tough, satisfying matches. But they are the rare exception rather than the rule.

Actually, in every team sport or esport, one mistake by one player often results in the opposing team coming out ahead. In fact, if the other team can't successfully exploit the mistake, that is considered a mistake in and of itself, as you pointed out in your post. The consequences of mistakes gets higher the more skilled your opponent is.

Also, I've come back from matches down 3 mechs too. Were they high Elo matches? I have no way to know. All we know is that someone who should know, Kar Berg maybe, was said to have said that high Elo matches often result in a stomp.

To reiterate the argument from my earlier post: if you have a high Elo and die fast then you aren't going to affect the outcome of the match like your Elo expects you to. So the match immediately becomes unbalanced. Also, skilled players are keen to exploiting mistakes by the other team (or at least sensing when they have the advantage).

Rorvik, on 18 January 2015 - 10:45 PM, said:

It would make sense if there were two commanders going head-to-head controlling 12 Mechs each, but that's not the way it works. It's more like a Chess game where the pieces are all independent of one another and are incapable of coordinating due to no voic chat and no ability to send meaningful messages to their team.

Players actions are far from random, especially on teams with a higher Elo. I'm all for voip but you can still be coordinated.

Theodore42, on 18 January 2015 - 01:50 PM, said:

Also, my lowest win ratio on a mech is 1.08, so MM definitely doesn't make losing matches to force even ratios.

Rorvik, on 18 January 2015 - 10:45 PM, said:

How does that follow? How is it even relevant to the point?

In the OP:

AjerWerklWerkl, on 17 January 2015 - 04:51 PM, said:

I see a lot of posts saying that 1.0 W/L means "MM sucks" or whatnot...

But if its job is supposed to find a game that's at your level, it seems fair that you'd win about 50% and lose about 50%.

Then people wrote things like:

El Bandito, on 17 January 2015 - 05:06 PM, said:

While the MM do try its best to make sure your WLR remains around 1.0, often by giving you dunderheads for teammates, with enough skill one can increase their WLR above 1.0 in pugs.

Zergling, on 17 January 2015 - 05:54 PM, said:

My winrate is consistently around 50%, or 1 win per loss.

It really does feel like the game is actively trying to keep my winrate at that level; when it goes above, I start getting numerous terrible teams, resulting in long losing streaks, and when my winrate is below, I start getting good teams.

And I replied with:

Theodore42, on 18 January 2015 - 01:50 PM, said:

Also, my lowest win ratio on a mech is 1.08, so MM definitely doesn't make losing matches to force even ratios.

It isn't even an argument, it is just a fact that happens to refute the erroneous conclusion that MM makes broken matches to keep player's w/l ratio at 1:1. Can you follow that?

Edit: although El Bandito did say that you could get your pug ratio over 1:1 it seemed like he was implying it was just trying to wreck you. So maybe that isn't really a fair example to quote. But obviously MM is going to try to wreck you until you get wrecked to 1:1 or you break the scale.

Edited by Theodore42, 19 January 2015 - 11:38 AM.

[SgtKinCaiD]

Egomane, on 19 January 2015 - 08:23 AM, said:

Karl implemetented a complete rewrite of the MM. I'm to lazy to look for more current ones. I only posted the link to show that the other one is outdated by a lot.

If you want to know more about the MM, you can always ask Karl directly. I have linked the thread where he answers community questions, several times now in various Elo and MM discussions, but I'll do it once more.
http://mwomercs.com/...courtesy-phone/

I asked him back in July, see for yourself :

Karl Berg, on 05 July 2014 - 11:40 AM, said:

SgtKinCaiD said:

@Karl

Thanks for your time and response, but i got another question :

So, from what i understand from your response, the ELO bucket are not re-introduced with the 4*3. Now i would like to know if the new MM will still use the rule of balancing a high ELO and a low ELO player against 2 medium ELO players ?
Matthew Craig post for reference : http://mwomercs.com/...ost__p__2633230

Bonus question : did you tune the ELO scoring formula for the new MM so that it's not based on Win/Loss only ?

The matchmaker doesn't specifically set out with this goal, no. It starts by seeding with a relatively old request, and uses the calculated Elo of that request to iteratively search the matchmaking queue for similarly skilled team members and opponents. Over time, the window of requests that it examines widens. That said, if one team ends up significantly ahead of the other team in terms of Elo, we have a tunable blending formula that can bias the search in an attempt to re-balance the teams.

From what i understand :

• mixing high and low ELO players is not intended but still happens (i.e. there is no mechanism to prevent it),
  
• if the team ELO sump up are to far apart, the MM can throw in a high or low ELO player in order to compensate the gape between both team.

[Egomane]

If and can are the Keywords here. Meaning it is not the norm. You quoted it yourself:

Quote

The matchmaker doesn't specifically set out with this goal, no.

Some make it out, as if the matchmaker puts them intentionally into matches with worse players all the time. At least they sound and argue like that is the case. This will only happen if the matchmaker has no other choice, which is an exception build into the matchmaker. Exceptions like this are a good thing or you would wait much much longer for a match, but I doubt they happen very often. Maybe you could ask him about some statistics, how often such a thing happens and how far apart those teams have become, when it happens. We don't know the values that are set there and have no idea how to set them into the correct perspective.

I have, a while ago, asked for threads from the developers to explain key game elements to stop this guessing game. So far I have not recieved an answer, but maybe with a little more support, they will listen and make it happen.
http://mwomercs.com/...sms-matchmaker/

[SgtKinCaiD]

He's playing on word : while it's not intended at start ("specifically"), it still happening because on how the MM handles the ELO score of each player.

You have to consider 2 things :
- ELO scores follows a gaussian curve meaning that up to 70%~75% players will have a relatively close ELO score, near the average.
- The main objective of the MM is to create balanced match (either by having the same ELO sum up on both side, or by having the same average of player's ELO score on both team as some mentioned).

Now what happens when the MM is creating a match ? The MM can handle the vast majority of the playerbase given the relatively close ELO score. The problem are those near both end : high / low ELO players. When dealing with these players, the MM has 2 options :
- it manages to find a similar player : good catch ! ... more a rare catch because the high / low players are outnumbered by a great margin given the gaussian repartition, and for bonus : throw in a specific tonnage requirement and a low waiting time ...
- it can't find such a player : the research scope widens (as mentioned by TheB3RG) until the MM finds an average player. But it's got another problem now because one of the team is at a disadvantage and the MM hates that : it will try to compensate by the finding the polar opposite of the first player ("a tunable blending formula that can bias the search").

While not intended, high / low ELO players are abnormalities for the MM and it uses them to cancel each other out.

Edited by SgtKinCaiD, 19 January 2015 - 04:31 PM.

[Egomane]

We are in agreement on that and, correct me if I am wrong, I never stated otherwise.

But abnormalities, exceptions or whatever you want to call it are not the norm. By their very definition. Let me post a few examples of how I understood Karls quoted explanation (and some of his other posts, I'll not mention and I am to lazy to look for).

Example A* (This should be how it is normaly, from all that I understand. I hope in 85 to 90 % off all matches this is true)
Player John is initiating a search for a match.
- John has an Elo score of 1550 (let's assume that this is the perfect average).
- His request is creating a new match instance.
Now the matchmaker will, for the next two minutes at least, put players of an Elo of +/- 50 within the range of John, into the match.
- The MM searches for other players, looking for an Elo between a 1500 and 1600 score.
- The MM finds them and attributes them to the teams.
- The Elo values of the teams are 1525 and 1555 at the end and the value is acceptable for the MM
The match starts!

Example B*
Same situation as Example A but...
- The MM searches for other players, looking for an Elo between a 1500 and 1600 score, but can't find enough of them in the first two minutes.
- The MM broadens the search spectrum. It now includes players as low as 1400 and up to 1600
- The MM attributes those players to the teams.
- The MM found some players at the higher end, but was able to distribute them equally.
- The Elo values of the teams are 1605 and 1630 at the end and the value is acceptable for the MM
The match starts!
You can witness this, every time the search time goes above the 2 minute mark, but it is not a fixed indicator, as it might hit your earlier, when you are drawn into the creation process, that is already running for a while.

Example C*
Again the same situation as Example A but...
- The MM searches for other players, looking for an Elo between a 1500 and 1600 score.
- For some reason, all the players the MM found, that are below the Elo of John, get placed into one team and all those above, are placed in the other.
- This would create a match with a to big Elo spread and the MM uses one of the exception rules and pulls two players with a higher Elo score into the group with the lower average.
The Match starts!
This is the situation Karl explained above and an exception.

Example D*
Now we change the situation. Player Fred is searching for a match. Fred has an Elo score of 1900 and with this is already part of the higher Elo spectrum.
Now the matchmaker will, for the next two minutes at least, put players of an Elo of +/- 50 within the range of Fred, into the match.
- Fred is creating a new match instance
- The MM searches for other players, looking for an Elo between a 1850 and 1950 score.
- The MM has trouble finding enough players within this range and after two minutes broadens the search radius.
- The MM does find some in the range between 1800 and 2000 and puts them into the match, but not enough to fill all slots.
- The MM opens up some more and finds a few more player between 1700 and 2100 Elo, but there are still 3 more slots open.
- Player Jack comes online and initiates a search for a match. He has an Elo of exactly 2100. The MM catches him and puts him into the match, but his release valves are already looking at an even bigger radius, so he pulls in another one, with an Elo score of 2200.
- We are still one player short but one side has reached a pretty high average and is far better rated for success then the other. The MM opens up all release valves and looks for a last player to be placed into the match to even out the score.
- The MM finds a player at 1100 Elo. A new player in his second match and piloting a trial. The MM doesn't and can't know about this. All it sees is the Elo value and it fits to balance the match out.
- The new player is drafted into the match and we get an instance where elite players meet trial mech driving newbie.
The Match starts!
As we have Players of all playing levels within the match, the new player gets frustrated, because he has to face elite ones, who can and should kill him almost immediatly. The Elite players may get frustrated because the feel like they have to carry newbs, because they haven't met most of them yet or didn't play often enough against each other to get a feel for each other.

*all numbers are pulled out of thin air

Each of these situations can happen. But everything except Example A should not be the norm. For those players who are at either end of the Elo curve it may become the norm, but by the very definition of Elo, they are an exception themself. There is nothing the matchmaker can do for them. Even if we change it from Elo to something else. They are at their end of the curve for a reason.

I very much doubt, that all those complaining, are at either end. This would make for one very strange Elo curve in MWO.

Once again: I agree that there are problem players. But by the sheer amount of complaint threads I do not believe that all those complaints are justified and that most of them come from a misconception of how the MM works. There is no intention by the MM to make higher skilled players having to carry lower skilled ones. This is always a low percentage exception (at least I hope so) and might even be the other way around, that a lower skilled player is pulled into a match to hinder better players. There might even be times of the day, where for some players the probability of an exceptional matchup is much higher, then at the prime times. We would need numbers only PGI can provide, to proof that though.

Feeling that my conclusion and assumptions are wrong, doesn't make them wrong. It doesn't make me right either, I am the first to admit that, but feelings are not a good thing to base an argumentation of the MM about. We should better work with what we know. And from that I came to the above conclusions and hope they are correct.

[Joseph Mallan]

Alistair Winter, on 17 January 2015 - 04:59 PM, said:

It's fair to say that Elo doesn't work that well for this game.

It's also fair to say there's not necessarily any correlation between a 1.0 WLR and enjoyable gameplay (e.g. being teamed up with and against players at your own level)

I'm grinding Commandos right now. I am at 5 wins(Carried all the way), 13 losses (in 19 matches... ) and 0 kills with 17 deaths, My Elo needs to drop drastically for Lights

[SgtKinCaiD]

@Egomane , i agree with most of what you said. I will just add some precisions.

Egomane, on 20 January 2015 - 04:17 AM, said:

Each of these situations can happen. But everything except Example A should not be the norm. For those players who are at either end of the Elo curve it may become the norm, but by the very definition of Elo, they are an exception themself. There is nothing the matchmaker can do for them. ... They are at their end of the curve for a reason.

That's why i asked several time to reintroduce the ELO buckets : divide the player population in 3/4 buckets (same number of players in each bucket), use the player ELO score to determine the bucket he's in, and have him play only with and against players of the same bucket, (exceptionnaly the one directly underneath/above if one or two players are missing). The match will be better balance because every player in both team will have roughly the same skill. The real problem is that PGI has to rewrite the MM nearly entirely for that.

Egomane, on 20 January 2015 - 04:17 AM, said:

Even if we change it from Elo to something else.

Right now the problem is not directly the ELO score, it's how it's calculated : it is only based on the W/L ratio which is not a good indicator of the actual skill of a player, especially teamplay. It should take more parameters into account : for example the player match score compared the average score of his team, etc.

Egomane, on 20 January 2015 - 04:17 AM, said:

There is no intention by the MM to make higher skilled players having to carry lower skilled ones. This is always a low percentage exception (at least I hope so) and might even be the other way around, that a lower skilled player is pulled into a match to hinder better players.

In the end, it's the same thing and the common complain in nearly every thread about the MM : carrying hard teammates who are derping around / just there to hinder their team.

Plus you have to consider the fact past a certain point, a low ELO player won't decrease anymore because basicaly he's caried by his team every time. So the MM is not throwing one low ELO player to compensate for one high ELO player, but several. And given the nature of the game and the snowball effect, you're at a huge disadvantage when you lack the firepower/tanking ability of even one teammate.

Edited by SgtKinCaiD, 20 January 2015 - 06:04 AM.

[Thorqemada]

All nice and well arguements - the Players POV is that he finds himself in teams where the Assaults boldly advance into the oposite direction of the enemy, the Locust breaks its Legs running around and theTimberwolf snipers 1000m away Targets with SPLs and brawls with LRMs and the R Key is missing on their Keyboards anyway.

Edited by Thorqemada, 20 January 2015 - 06:27 AM.

[SgtKinCaiD]

Don't forget the teammate asking how to save the game ... I was breathless after i read that in team chat.

[Rorvik]

Theodore42, on 19 January 2015 - 11:31 AM, said:

Actually, in every team sport or esport, one mistake by one player often results in the opposing team coming out ahead. In fact, if the other team can't successfully exploit the mistake, that is considered a mistake in and of itself, as you pointed out in your post. The consequences of mistakes gets higher the more skilled your opponent is.

I disagree; that the consequences of mistakes are lower the higher and more skilled the players in a match are, because with higher skill there is, or should be, a better ability to adapt to the handicap of losing teammates.

Think back to all those games you played where the final score was 12-11 or 10-12. How did the score get like that? Most of the time, when one team loses a player, the other team loses one soon afterward. It's extraordinarly rare a team to come back from 1-5. So, clearly, when you have the aforementioned close games, losing one or two teammates isn't the be-all and end-all you make it out to be.

However, I do understand the point you are getting at here. If the team has both high and low skilled players, then if you lose the low skilled player it's unlikely to make a difference as his contribution would have been negligble to begin with. While this is certainly true, it is only so in matches where there are large differences in player skill...which are only going to happen if the player population is really low, or the MM algorithm is faulty.

Theodore42, on 19 January 2015 - 11:31 AM, said:

Also, I've come back from matches down 3 mechs too. Were they high Elo matches? I have no way to know. All we know is that someone who should know, Kar Berg maybe, was said to have said that high Elo matches often result in a stomp.

Which I maintain means the MM is NOT working. Refer to my example above as to why.

Theodore42, on 19 January 2015 - 11:31 AM, said:

Players actions are far from random, especially on teams with a higher Elo. I'm all for voip but you can still be coordinated.

I never said player actions were random, just that they are incapable of coordinating due to the lack of a meaningful communication system (i.e. that ****** chat does not count).

Theodore42, on 19 January 2015 - 11:31 AM, said:

It isn't even an argument, it is just a fact that happens to refute the erroneous conclusion that MM makes broken matches to keep player's w/l ratio at 1:1. Can you follow that?

You seem to be misunderstanding the argument here.

The OP was asking whether having a 1:1 W/L proves the MM is working. It's "working", in the sense that you have a 1:1 W/L ratio, but HOW you get that ratio is the real issue many of us are arguing over.

Ideally, the MM will set you up with other players where there is a 50/50 chance of winning, meaning on a long enough timeline you'll have a 50/50 W/L, because you are playing with and against players at your skill level. Again, refer to my example above for how this should ideally play out.

But several people are saying that that isn't actually what's happening, that, in fact, the MM is "forcing" wins and losses by putting you in stacked teams if you end up with too many wins or losses, so as to balance out your W/L. So, practically speaking, you do NOT have a 50/50 chance of winning every match, only some matches, hence the steamrolls that seem to follow winning / losing streaks.

The fact you have a 1.08 W/L on your least winningest Mech is a completely irrelevant fact within this context. It in no way proves that the MM isn't forcing wins and losses via stacked teams.

[Deathlike]

SgtKinCaiD, on 20 January 2015 - 06:01 AM, said:

That's why i asked several time to reintroduce the ELO buckets : divide the player population in 3/4 buckets (same number of players in each bucket), use the player ELO score to determine the bucket he's in, and have him play only with and against players of the same bucket, (exceptionnaly the one directly underneath/above if one or two players are missing). The match will be better balance because every player in both team will have roughly the same skill. The real problem is that PGI has to rewrite the MM nearly entirely for that.

I'm convinced that our population is strained at different times of the day, which make the bucket difficult to construct... especially at the very high and very low Elo levels.

[Thorqemada]

From the Patchnotes:
- Default game save folder changed to C:\Users\[username]\Saved Games directory

Edited by Thorqemada, 20 January 2015 - 03:57 PM.

[SgtKinCaiD]

Deathlike, on 20 January 2015 - 11:38 AM, said:

I'm convinced that our population is strained at different times of the day, which make the bucket difficult to construct... especially at the very high and very low Elo levels.

I'm not sure about that : even with a queue greater than 50% in the heavy queue, you still find a match under 3 minutes in a TBR.

[Deathlike]

SgtKinCaiD, on 21 January 2015 - 01:11 AM, said:

I'm not sure about that : even with a queue greater than 50% in the heavy queue, you still find a match under 3 minutes in a TBR.

The higher your Elo, the longer the wait.

It has always been that way.

[Divine Retribution]

Not a fan of Elo. Won 62.68% of my 7,270 archived games, won 61.25% of my 2,803 current games, so I don't notice a difference there. That's with a mix of solo and playing with my casual group.

What I do notice is how my win % has stayed about the same in both matchmakers. Before Elo-based matchmaking I could end up with terrible teammates, but at least it happened randomly. Now when I get bad teammates I know that it is by design. It's not any more fair to the bad player now than it used to be, but at least then it was just luck of the draw. Now it's a "you've been selected as fodder, person who can't aim and walk at the same time, good luck". On the other side of the coin I sometimes see it as "I've predicted your defeat and have done everything in my power to make it come true. You can still win, technically, you just have to solo half the enemy team to give your 11 teammates a chance to defeat the other half. Carry, die, or carry until you die".

Other players have written that good players don't want Elo so they can just PUG stomp forever, but again my win % is about the same in both matchmaker implementations. As someone who enjoys playing with a casual group that spans a number of skill levels, I'll add another argument against Elo-based matchmaking: grouping.

Let's say a great player and a terrible player are friends and play together. Over time, the terrible player's Elo scores will be artificially inflated. After a while that player decides to PUG for a bit and is basically beaten senseless in game after game. Not very fun, I'd imagine, and I think a lot of players have experienced this situation. The great player has probably been dragged down a bit, and when pugging will tend to be the best performer. Their Elo scores, which don't represent their skill level, will skew every game PUG game until their Elo scores readjust.

If both PUG long enough, their Elo ratings will stabilize where they belong. But when they team up again the process begins anew. The group matches will seem easier for the terrible player than pugging, but still more difficult than before he or she began playing with the great player. The group matches will seem easier to the great player than pugging. Neither will be playing where they belong.

The effect is that Elo-based matchmaking discourages playing with friends that aren't near your own skill level. Or worse, makes you dependent on playing with your friends because you don't want to go back to getting stomped in PUG games with an artificially inflated Elo for however long it takes to get back to your true Elo levels. I imagine a number of players have quit MWO because they failed to understand why they kept losing in PUG games and became frustrated before their Elo readjusted from playing with a group.

I know PGI has shown some data regarding team Elo spread and other stats about opposing teams in matches. What I don't recall seeing is a stat showing how accurate the matchmaker is when selecting a winner. If the matchmaker decides team A should win, how often is the matchmaker correct in that prediction? If it's a high %, then I will be assured that the current matchmaker holds no value to me.

To me the job of a good matchmaker is to create matches where both sides have as close to 50% chance of winning as possible given the available player pool (and possibly other variables), not attempt to artificially create a 50% win rate by stacking the teams. How this can be achieved without discouraging friends of varying skill levels playing together, I don't know. That's why I would rather just dump Elo, at least in group games.

Deathlike, on 21 January 2015 - 01:39 AM, said:

The higher your Elo, the longer the wait.

It has always been that way.

And the longer the wait, the worse the match is likely to be (win or lose, roll or be rolled). After 5 minutes or so pretty much anything can happen.

Edited by Divine Retribution, 21 January 2015 - 03:52 AM.

[SgtKinCaiD]

@Divine Retribution : I agree with most of what you said but there are some misunderstandings :

• The (now archived) stats were put in place with the first iteration of the ELO based MM. So actually, you don't have stats when the MM was only class based. It's a sad thing because the "new" ELO based MM was pre-seeded with those matchs.
  
• In group, your ELO is multiplied by a certain amount to reflect your enhanced teamplay, etc. the more the heavier is the multiplier.
  
• The actual MM doesn't specifically stack the odds against you : for him, a match between two average teams and a match beetwen an average team against a team composed of high/low ELO players are the same as long as the ELO sum up of both team are equal (or the average ELO score on both team). For the MM, both match have a 50/50 win rate for both team. It's nice on paper but the results on the field are more tendentious.