General DiscussionMany low skill players say that MMR is just a number but when they pl...
Many low skill players say that MMR is just a number but when they play with higher skill players they can't even keep up . What is your approach to this topic ? in General Discussion
There's a similar ELO system in AOE, except it takes into account the amount of players (variable) and skill difference (also variable)
People accept that ELO (like mmr) is just a number, it's only the amount of wins and losses depending on the situation, however there is a correlation between win/loss and skill (obviously)
However people don't judge their skill by their number, unless there's a big gap, which is something most people in dota are obsessed with (people in 2k flame for a difference of not even 200 mmr)
In AOE though everyone starts at the same ELO, which personally I think would be better in dota since the calibration is pretty garbage
I think a big difference between AOE and Dota is that smurfing is an issue in Dota 2, if everyone started at the same mmr I think we'd see a huge issue with smurfing. You would have to set this bar very low or else it would totally ruin games for players who are meant to be around that mmr (imagine if everyone started at 3k, suddenly every sub 3k just starts another account whenever they drop below 2k and ruins matchmaking). Then with everyone starting at 1k if I ever felt like shitting on some kids I'd just make a new account and enjoy my easy 1k games.
@CutnPaste fucks sake the whole point of what im saying is that u cant compare between regions accurately, which is in and of itself a flaw. Mmr shud (ideally) be a rating that can compare all dota players but it cant rn.
They don't vs each other because they are on different regions how the fuck can you expect to compare them? If there is something to be learnt from 100 years of sports experience there is no way to compare top teams/players in anything unless they actually play against each other. This is the reason we have TI finals instead of just giving the prize money to the highest 5 mmr players in the world.
Im trying to be nice and all, but it seems like u guys are all so fucking devolved into dumb shit fuck boys who can only stare at a rating in an online action RTS game as proof of their complete superiority. Fuck u dumb shit blue stars i sincerely hope u all have 2 beautiful children who die of cancer.
That was a valid point, but the thing is that if they all play at the same level, and since the mmr system is the same for all of them then they should all be similar mmr. Either way im fucking done posting, this site is full of so much elitist bull shit cancer.
Just read the words on the dota 2 leaderboards and your 'problem' with mmr answered itself, mmr isn't comparable between regions and by thinking for more than 5 seconds you can immediately figure out why (they don't play against each other). One day we will live in a utopia where all Dota 2 players are segregated into a giant dome playing on the same ping with the same server and we can finally compare Sumail's mmr with Miracle's but until then we just have to accept mmr as it is.
If theyre similar skill (which we know because international tournaments do exist), then the elo system shud award them the same points over time regardless if they play each other, because they amongst themselves, and since theyre similar skill they shud gain similar mmr.
There's an unhealthy obsession with MMR in dota because it places you into a certain category where you don't play anyone else but people your own level
In AOE people don't care about rating because they can play anyone they want, if they play someone much higher than them then they barely lose any ELO, so you can play anyone for fun
Not sure how well it would work in dota because people are obsessed with their rank and would probably try to abuse the system, personally I think it's the community's fault
(also aoe is a super old game barely anyone plays lul)
If the chess community were somehow split into groups of people that physically could never play each other I would 100% expect the ratings to diverge over time. Of course since chess players regularly play international tournaments and the Elo system is used for all these games they are able to keep everyone on the same rating metric.
An interesting thing related to this is when I played SC2 there was a website called aligulac that basically attempted to be the same as the chess Elo system but for SC2 and only using tournament games. Quite quickly a flaw developed where due to EU players mainly playing each other in random online cups except for rare international tournaments and the same with KR players who overall played less often, KR players ended up being massively underrated compared to EU players, and in the rare times they ended up playing each other you'd see the KR players smash the EU players and get massive ratings gain. Not sure if it's similar now since I dont really follow the scene but just looking at the list they have a Polish player in the top 5 which seems questionable if you want to know the top 5 players in the world.
MMR means how good you are at winnig pubs. It matters something. But it can have huge inaccuracy when someone don't play very often. MMR is huge grind. But there can be diff of 1.5k mmr and this lower mmr player trash other on middle very hard.
I remember two 6k retard arguing on bulldog's stream. And he watched how they play 1v1 on SF. I would say they were trash if they were even 3k.
Another thing. As i said it's about you ability to winning pubs. If person is great drafter, tempo feeler, he will be not so great at climbing as a hero spammer. But he can shine when he will go to battlecup or 5v5 mm.
MMR is tool like GPD in economics. They can show you something, there are corealations like (higher mmr -> more good players, higher GPD -> richer coutries) But there are things like Equatorial Guinea having higher GPD per capita than Finalnd or UK.
So it can tell you something about player. But it doesn't have to. It's not a fucking Starcraft when you play just for yourself.
Team dynamics and support vs core shit make stuff difficult to rate very accurately. Btw mmr works the same no matter who ur comparing, it shudnt matter if i compare a 6k to a 6.1k or if i compare a 3k to a 3.1k, the mmr difference is the same so theoretically the higher one shud have a 64% chance of beating the lower one.
Why are you comparing 1v1, it's a team game not 1v1 mid matchup? If it was 6.1k average vs 6.0k average I could believe a 64% winrate, if the winrate wasn't 64% the players would change mmrs until they stabilise, this is literally how the system functions. 6.1k vs 6.0k average doesn't sound like a lot until you realise thats a total mmr difference of 500 points, it's the difference between having a 6k flat player and a 6.5k player on your team.
but people are constantly changing mmr by climbing or dropping. 100 mmr difference being a big difference would matter if people were aready at their correct mmr, but they arent.
A 6100 does not have a 64% win chance over a 6000 player. It is a 5v5 game so that is only a 20 average mmr difference were they to replace each other in a game which barely affects winrate. A 6500 equates to a 64% winrate difference compared to a 6k flat assuming 6k average vs 6.1k average which I think most people would find reasonable.
I meant team avg my bad i stated it that way the first time i said it. The way u say a 6k avg team vs a 6.1k avg team is like a 6k vs a 6.5k solo doesnt make sense to me. Each player has a different mmr and u cant say cuz each individual player is on avg 100 mmr greater. U wud have to compare ever single mmr difference and see if the avg is 100, which it shudnt be. The way ur adding up mmrs is just weird.
100 mmr is barely anything and it barely affects winrate, the system is working as intended? I don't understand what your complaint is? If I was replaced by a player 100 points higher than me the average mmr of my team would increase by 20 which barely affects the outcome of the match, I dont know the actual numbers but I think you end up with a +24 or +23 game instead of +25. It is only when you get 250+ points that mmr difference starts to actually matter and this is reflected in the winrates.
people have unlucky days where they lose maybe 8 games in a row, but at the end of the day its not like they are drastically worse players like the ranking system would say. MMR may indicate skill but its margin of error is fucking huge. like 1500 points, but i cant get a real stat for that so idk.
Its clearly not 1.5k but a 100 point avg difference isnt gonna suddenly make ur team 14% more/less likely to win the game. They need to either adjust how much mmr u get after a match or the way the equation states winrate differences so high for a practical skill difference so small.
100 average mmr across a whole team is 500 points across one player, that is quite a significant difference. If after I found a match I got a player 500 points above me to sit down at the computer and play in my stead I'd expect him to win more often than not, ofc he's not going to win always due to variance but 64% seems pretty reasonable and you could almost argue it is too low. In my case its literally a 6.9k almost 7k player replacing me, having played with high 6ks the skill difference is noticeable. If it didn't work like this how the fuck did he climb to that mmr in the first place?
There are plenty of explanations for that, smurf accounts, actually tryharding in ranked vs just randoming and picking shit, mastering some gimmick playstyle, training in party ranked/unranked/lobby games, actually practicing vs just mindlessly playing ranked, the list goes on. I'm 100% certain that nobody who regularly plays ranked just wakes up, sits down at their computer, changes nothing about their game and gains 1500 mmr. Since ranked came out I've gained 2500 points over a few years and I'm know that I've become a much better dota player during the process, it's not as if I just randomly started gaining points.
Seriously though i think playing in 4k has given me a bad view of the community. Its way too common for people to be drastically different in skill while having the same mmr value. perhaps it is different in other brackets, but in 4k you can never assume someones skill based on their number.
I think you are making the mistake of assuming people's skill based off one game. MMR isn't accurate if you just look at one game in a vacuum, you have to play quite a few games to get an accurate mmr. In pubs where you see one guy once and then never again and he shits the bed as SF or whatever you think 'wtf is this garbage how is the same mmr as me' but you never see his previous SF game where he fucking carried his team. I'd consider myself a really consistent player but I still have bad games, I had this terrible SF game where I got dumpstered mid and my whole team called me a bought account.
Ofc in one game I performed below my mmr, and if you were in that game you might say 'mmr is broken my 6.4k SF can't even win lane vs ember' but what about the games where I go apeshit and solo carry? You never see those because you have a sample of only 1 game which is never enough to determine if someones skill matches up to their mmr.
Thing is one player on ur team is 500 mmr less than one from the other team. But u also have someone who is like 400 mmr higher than someone on the enemy team. If u compare both the weakest links, then the difference is still 100, u cant just look at 1 side.
Choose who wins more often than not. Ofc its going to be team A. They won't win always but 64% sounds pretty fucking reasonable to me. I'd imagine Valve have actual mathematicians and people who know what they are doing making this system with all the matchmaking data they could ever dream of as well, not uninformed uneducated forum posters with no data speculating about mmr based on what they think is 'fair'. 'Only' a 100 average mmr difference but the outcome seems clear to anyone who actually realises that 100 average mmr difference is 500 points across the team. If team A didn't win 64% of the time vs team B their mmr would drop which means the system is working as intended, if the 6.5k can't carry them he doesn't deserve his mmr.
Or u cud formulate it like this
Team A: 5.8, 5.9, 6.0, 6.1, 6.2
Average: 6
Team B: 5.9, 6.0, 6.1, 6.2, 6.3
Average: 6.1
And voila what i said came true. Their total mmr is 500 higher but its spread among multiple players so the differences r less. I mean the average shud be the same, but dota is complex, and i dont think that team B has a 64% chance of winning over team A. But idk man, im pretty sure valve doesnt employ mathematicians, they can just use Elo's equation themselves. And not all players have the same impact every game, and then there is the problem that people's mmr fluctuates constantly, maybe they dropped in mmr but they suddenly start playing like their original mmr. Over time it shud stabilize but people change faster than the system. Whatever. The overall principle is sound. I guess i cant come up with a better one so whatever.
At least im not some 5k+ elitist shit head who walks around like he is some sort of all knowing god who should be worshipped by all the lesser mmr peasants, and just calls people ns scrubs. Thx for having a legit debate even tho u let ur dumb fuck autistic blue star self out a little bit.
I am one of the 'uninformed uneducated forum posters', that's why I don't speculate on whether the estimated winrates Valve's mmr system gives are unfair.
I can speculate how ever much i want i never said that oh ho i must be a genius and shud run valves system haha so broken im definetly the most qualified haHaa
Daddy in your example you provided is exactly the same as the 6 6 6 6 6.5 case. If you eliminate the ppl with the same mmr you are left with one guy 5.8 and one guy 6.3. 500 mmr difference. But I kinda get what u mean that the team performance might not seem so different
@ Best lifestealer
Vs how many 6k Players did u even play? Like 1 every 10 games? When i play on Australia Servers i usually have 200 ping and there are mmr gabs of like 1000+ mmr even in high 4k. And still i have like 70% winrate...
I have been getting very good games recently, Valve has done something to the matchmaking and it's legit impossible to find games outside of peak times but when you search at good hours I have gotten really high average games. Last 5 ranked games I've played are all 5.8k+ average.
the mmr given for the unbalanced matches is based on long term statistical data, and the number is made in such a way that the longrun expectation of mmr change is 0.
lets say if a 100 mmr difference means a +20/-30 game for the higher rated team (i dnt remember the exact numbers), it means that the waker one wins 40% of matches of the kind. 30x4=20x6. 500 mmr difference is +10/-40 afair, so the weaker team probably has smthng around 20% estimated winrate.
Будь ласка, увійдіть до системи для написання коментарів.
Where's the global leaderboard?
The MMR of each division is on a different scale, and comparing MMRs across divisions is not currently meaningful.
Can you read??? Or do sub 5ks not even know that much...
There's a similar ELO system in AOE, except it takes into account the amount of players (variable) and skill difference (also variable)
People accept that ELO (like mmr) is just a number, it's only the amount of wins and losses depending on the situation, however there is a correlation between win/loss and skill (obviously)
However people don't judge their skill by their number, unless there's a big gap, which is something most people in dota are obsessed with (people in 2k flame for a difference of not even 200 mmr)
In AOE though everyone starts at the same ELO, which personally I think would be better in dota since the calibration is pretty garbage
I think a big difference between AOE and Dota is that smurfing is an issue in Dota 2, if everyone started at the same mmr I think we'd see a huge issue with smurfing. You would have to set this bar very low or else it would totally ruin games for players who are meant to be around that mmr (imagine if everyone started at 3k, suddenly every sub 3k just starts another account whenever they drop below 2k and ruins matchmaking). Then with everyone starting at 1k if I ever felt like shitting on some kids I'd just make a new account and enjoy my easy 1k games.
@CutnPaste fucks sake the whole point of what im saying is that u cant compare between regions accurately, which is in and of itself a flaw. Mmr shud (ideally) be a rating that can compare all dota players but it cant rn.
They don't vs each other because they are on different regions how the fuck can you expect to compare them? If there is something to be learnt from 100 years of sports experience there is no way to compare top teams/players in anything unless they actually play against each other. This is the reason we have TI finals instead of just giving the prize money to the highest 5 mmr players in the world.
Im trying to be nice and all, but it seems like u guys are all so fucking devolved into dumb shit fuck boys who can only stare at a rating in an online action RTS game as proof of their complete superiority. Fuck u dumb shit blue stars i sincerely hope u all have 2 beautiful children who die of cancer.
isnt rtz 9k? Also if "but the equation says that a 3.1k average team has a 64% chance of beating a 3k team" is true thats retarded.
That was a valid point, but the thing is that if they all play at the same level, and since the mmr system is the same for all of them then they should all be similar mmr. Either way im fucking done posting, this site is full of so much elitist bull shit cancer.
Just read the words on the dota 2 leaderboards and your 'problem' with mmr answered itself, mmr isn't comparable between regions and by thinking for more than 5 seconds you can immediately figure out why (they don't play against each other). One day we will live in a utopia where all Dota 2 players are segregated into a giant dome playing on the same ping with the same server and we can finally compare Sumail's mmr with Miracle's but until then we just have to accept mmr as it is.
https://en.m.wikipedia.org/wiki/Elo_rating_system
It is wikipedia but u guys can check the sources if u rlly believe its inaccurate. Just read the intro.
If theyre similar skill (which we know because international tournaments do exist), then the elo system shud award them the same points over time regardless if they play each other, because they amongst themselves, and since theyre similar skill they shud gain similar mmr.
There's an unhealthy obsession with MMR in dota because it places you into a certain category where you don't play anyone else but people your own level
In AOE people don't care about rating because they can play anyone they want, if they play someone much higher than them then they barely lose any ELO, so you can play anyone for fun
Not sure how well it would work in dota because people are obsessed with their rank and would probably try to abuse the system, personally I think it's the community's fault
(also aoe is a super old game barely anyone plays lul)
If the chess community were somehow split into groups of people that physically could never play each other I would 100% expect the ratings to diverge over time. Of course since chess players regularly play international tournaments and the Elo system is used for all these games they are able to keep everyone on the same rating metric.
An interesting thing related to this is when I played SC2 there was a website called aligulac that basically attempted to be the same as the chess Elo system but for SC2 and only using tournament games. Quite quickly a flaw developed where due to EU players mainly playing each other in random online cups except for rare international tournaments and the same with KR players who overall played less often, KR players ended up being massively underrated compared to EU players, and in the rare times they ended up playing each other you'd see the KR players smash the EU players and get massive ratings gain. Not sure if it's similar now since I dont really follow the scene but just looking at the list they have a Polish player in the top 5 which seems questionable if you want to know the top 5 players in the world.
MMR means how good you are at winnig pubs. It matters something. But it can have huge inaccuracy when someone don't play very often. MMR is huge grind. But there can be diff of 1.5k mmr and this lower mmr player trash other on middle very hard.
I remember two 6k retard arguing on bulldog's stream. And he watched how they play 1v1 on SF. I would say they were trash if they were even 3k.
Another thing. As i said it's about you ability to winning pubs. If person is great drafter, tempo feeler, he will be not so great at climbing as a hero spammer. But he can shine when he will go to battlecup or 5v5 mm.
MMR is tool like GPD in economics. They can show you something, there are corealations like (higher mmr -> more good players, higher GPD -> richer coutries) But there are things like Equatorial Guinea having higher GPD per capita than Finalnd or UK.
So it can tell you something about player. But it doesn't have to. It's not a fucking Starcraft when you play just for yourself.
Team dynamics and support vs core shit make stuff difficult to rate very accurately. Btw mmr works the same no matter who ur comparing, it shudnt matter if i compare a 6k to a 6.1k or if i compare a 3k to a 3.1k, the mmr difference is the same so theoretically the higher one shud have a 64% chance of beating the lower one.
Why are you comparing 1v1, it's a team game not 1v1 mid matchup? If it was 6.1k average vs 6.0k average I could believe a 64% winrate, if the winrate wasn't 64% the players would change mmrs until they stabilise, this is literally how the system functions. 6.1k vs 6.0k average doesn't sound like a lot until you realise thats a total mmr difference of 500 points, it's the difference between having a 6k flat player and a 6.5k player on your team.
but people are constantly changing mmr by climbing or dropping. 100 mmr difference being a big difference would matter if people were aready at their correct mmr, but they arent.
A 6100 does not have a 64% win chance over a 6000 player. It is a 5v5 game so that is only a 20 average mmr difference were they to replace each other in a game which barely affects winrate. A 6500 equates to a 64% winrate difference compared to a 6k flat assuming 6k average vs 6.1k average which I think most people would find reasonable.
M m r just a numba guys DendiFace
I spam magnus to 8k DendiFace
I meant team avg my bad i stated it that way the first time i said it. The way u say a 6k avg team vs a 6.1k avg team is like a 6k vs a 6.5k solo doesnt make sense to me. Each player has a different mmr and u cant say cuz each individual player is on avg 100 mmr greater. U wud have to compare ever single mmr difference and see if the avg is 100, which it shudnt be. The way ur adding up mmrs is just weird.
Sorry replace my 100 mmr difference with 20 but the point stands
100 mmr is literally 4 games. Like the difference isnt that big man.
I think MMR only matters on 6k mmr and above or on professional level.
100 mmr is barely anything and it barely affects winrate, the system is working as intended? I don't understand what your complaint is? If I was replaced by a player 100 points higher than me the average mmr of my team would increase by 20 which barely affects the outcome of the match, I dont know the actual numbers but I think you end up with a +24 or +23 game instead of +25. It is only when you get 250+ points that mmr difference starts to actually matter and this is reflected in the winrates.
people have unlucky days where they lose maybe 8 games in a row, but at the end of the day its not like they are drastically worse players like the ranking system would say. MMR may indicate skill but its margin of error is fucking huge. like 1500 points, but i cant get a real stat for that so idk.
If you think the margin for error is 1500 points you are seriously deluded, I'd smash any 4k player even in roles I don't play at all like mid.
I mean 100 average mmr. Given 5 players on a daily basis their mmr can fluctuate wildly, which can swing winrates wildly too.
Its clearly not 1.5k but a 100 point avg difference isnt gonna suddenly make ur team 14% more/less likely to win the game. They need to either adjust how much mmr u get after a match or the way the equation states winrate differences so high for a practical skill difference so small.
100 average mmr across a whole team is 500 points across one player, that is quite a significant difference. If after I found a match I got a player 500 points above me to sit down at the computer and play in my stead I'd expect him to win more often than not, ofc he's not going to win always due to variance but 64% seems pretty reasonable and you could almost argue it is too low. In my case its literally a 6.9k almost 7k player replacing me, having played with high 6ks the skill difference is noticeable. If it didn't work like this how the fuck did he climb to that mmr in the first place?
how the fuck do you think people climb 1.5k in short times then. its not common but it happens.
There are plenty of explanations for that, smurf accounts, actually tryharding in ranked vs just randoming and picking shit, mastering some gimmick playstyle, training in party ranked/unranked/lobby games, actually practicing vs just mindlessly playing ranked, the list goes on. I'm 100% certain that nobody who regularly plays ranked just wakes up, sits down at their computer, changes nothing about their game and gains 1500 mmr. Since ranked came out I've gained 2500 points over a few years and I'm know that I've become a much better dota player during the process, it's not as if I just randomly started gaining points.
did you start listening to yung lean?
Seriously though i think playing in 4k has given me a bad view of the community. Its way too common for people to be drastically different in skill while having the same mmr value. perhaps it is different in other brackets, but in 4k you can never assume someones skill based on their number.
I think you are making the mistake of assuming people's skill based off one game. MMR isn't accurate if you just look at one game in a vacuum, you have to play quite a few games to get an accurate mmr. In pubs where you see one guy once and then never again and he shits the bed as SF or whatever you think 'wtf is this garbage how is the same mmr as me' but you never see his previous SF game where he fucking carried his team. I'd consider myself a really consistent player but I still have bad games, I had this terrible SF game where I got dumpstered mid and my whole team called me a bought account.
https://www.dotabuff.com/matches/2904068583
Ofc in one game I performed below my mmr, and if you were in that game you might say 'mmr is broken my 6.4k SF can't even win lane vs ember' but what about the games where I go apeshit and solo carry? You never see those because you have a sample of only 1 game which is never enough to determine if someones skill matches up to their mmr.
Thing is one player on ur team is 500 mmr less than one from the other team. But u also have someone who is like 400 mmr higher than someone on the enemy team. If u compare both the weakest links, then the difference is still 100, u cant just look at 1 side.
I dont think you understand my point.
TEAM A: 6.5, 6, 6, 6, 6
AVERAGE: 6.1
TEAM B: 6, 6, 6, 6, 6,
AVERAGE: 6.0
Choose who wins more often than not. Ofc its going to be team A. They won't win always but 64% sounds pretty fucking reasonable to me. I'd imagine Valve have actual mathematicians and people who know what they are doing making this system with all the matchmaking data they could ever dream of as well, not uninformed uneducated forum posters with no data speculating about mmr based on what they think is 'fair'. 'Only' a 100 average mmr difference but the outcome seems clear to anyone who actually realises that 100 average mmr difference is 500 points across the team. If team A didn't win 64% of the time vs team B their mmr would drop which means the system is working as intended, if the 6.5k can't carry them he doesn't deserve his mmr.
Or u cud formulate it like this
Team A: 5.8, 5.9, 6.0, 6.1, 6.2
Average: 6
Team B: 5.9, 6.0, 6.1, 6.2, 6.3
Average: 6.1
And voila what i said came true. Their total mmr is 500 higher but its spread among multiple players so the differences r less. I mean the average shud be the same, but dota is complex, and i dont think that team B has a 64% chance of winning over team A. But idk man, im pretty sure valve doesnt employ mathematicians, they can just use Elo's equation themselves. And not all players have the same impact every game, and then there is the problem that people's mmr fluctuates constantly, maybe they dropped in mmr but they suddenly start playing like their original mmr. Over time it shud stabilize but people change faster than the system. Whatever. The overall principle is sound. I guess i cant come up with a better one so whatever.
At least im not some 5k+ elitist shit head who walks around like he is some sort of all knowing god who should be worshipped by all the lesser mmr peasants, and just calls people ns scrubs. Thx for having a legit debate even tho u let ur dumb fuck autistic blue star self out a little bit.
whats happening here?
there is absolutely no chance mmr has a 1500 margin of error, THINK
there is less than zero chance of me being able to perform at a 7.2k level, i already lose lane in most even matchups against 5k/6ks
there is also zero chance of me going on a losing streak in 4.5k avg games unless i am playing while seriously impaired due to fatigue or something
Ye 1500 is massive.
I am one of the 'uninformed uneducated forum posters', that's why I don't speculate on whether the estimated winrates Valve's mmr system gives are unfair.
I can speculate how ever much i want i never said that oh ho i must be a genius and shud run valves system haha so broken im definetly the most qualified haHaa
who are you mocking right now actually
Holy crap. This escalated quickly.
-mmr is just a number
*8k mmr professional millionare*
the perfect quote for a 2k incompetent kid to use
Daddy in your example you provided is exactly the same as the 6 6 6 6 6.5 case. If you eliminate the ppl with the same mmr you are left with one guy 5.8 and one guy 6.3. 500 mmr difference. But I kinda get what u mean that the team performance might not seem so different
@ Best lifestealer
Vs how many 6k Players did u even play? Like 1 every 10 games? When i play on Australia Servers i usually have 200 ping and there are mmr gabs of like 1000+ mmr even in high 4k. And still i have like 70% winrate...
I have been getting very good games recently, Valve has done something to the matchmaking and it's legit impossible to find games outside of peak times but when you search at good hours I have gotten really high average games. Last 5 ranked games I've played are all 5.8k+ average.
https://www.dotabuff.com/matches/2963280450
Highest average Aus game I've ever played, 7 6ks, mid bear was also 7k on 5.9k smurf.
the mmr given for the unbalanced matches is based on long term statistical data, and the number is made in such a way that the longrun expectation of mmr change is 0.
lets say if a 100 mmr difference means a +20/-30 game for the higher rated team (i dnt remember the exact numbers), it means that the waker one wins 40% of matches of the kind. 30x4=20x6. 500 mmr difference is +10/-40 afair, so the weaker team probably has smthng around 20% estimated winrate.