General DiscussionA Supplementary Experiment to Swiftending's

http://dotabuff.com/matches/515841197

Here is an example. This game was a 5 in every other category but took 23 minutes to win.

Pros rarely call gg before 20 min... and even if they did call gg it might take 5 more minutes for the game to end. So if the game is over, ancient dead, by 25 minutes maybe there was no chance to win.

If a carry on a losing team can still manage 400 gpm they have some chance to recover... still they probably need treads, bkb, and at least 1 major dmg item to fight. That's maybe 12k gold farmed including some deaths, so it would take 30 minutes at 400 gpm. Ending the game at 25 min means there is no way the carry was ever ready to fight. 35 min there would be at least one or two teamfights where the losing teams carry was able to fight.

Ok here is an example game with the new time numbers. This is definitely a game either team could win.

http://dotabuff.com/matches/532520869
Both teams had feeders. Our side had excellent vision all game from OM. Their side had a NP who knew how to win and timed his pushes well... although I did kill him with a bmail once because he was so heavily reliant on right click he immediately got a bkb to counter. This game should come out near a 3.

3 Kill score
3 Tower score
3 Gold score
3 XP score
3 Time score

Composite score 3... of course it would have also been a 3 with the old time scoring too. This game both teams really did have a chance to win. While we seem to have a lot of teamwork from assists, that 5 man play to avoid wisp-ursa NP pickoffs really let NP take towers for free. Despite constant wards on roshan ursa still got ageis 3 times. Sometimes I don't why anyone even bothers trying to prevent an ursa from taking it. Even a tiny gap in map position and he can solo rosh before you get there.

Next game was impossible to win. I did could have played better, however we lost because the game was extremely unfair not because of my mistakes.

http://dotabuff.com/matches/532794315
1 Kill Score
1 Tower Score
1 Gold Score
1 XP Score
3 Time Score

Composite score 1.2 NP afk jungled forever and used it to feed later while our lanes collapsed. Sand King was an autoattacker but at least cast his spells on occasion. Most fights my team saw the enemy and started trying to walk away directly down the lane until they died without casting any spells. I had to play troll as a support since no one would do it.

Next game was very easy to win. The opposing team made one attempt at teamplay ganking void top with 4 heroes and taking a T1 tower. After that they completely collapsed getting out played at nearly every turn.

http://dotabuff.com/matches/532877393
5 Kill Score
5 Tower Score
5 Gold Score
5 XP Score
3 Time Score

Composite Score 4.8. The only factor that was not a blowout was time and that's because void had to farm up before we really had a tower killing hero. Razor fed a bit and went tank.

So here we have 3 games... one very close and two blowouts either way. All got a "3" time score. It may just not be a very good indicator of how close the game was. Or we could think of it as a moderating factor. Only the most extreme wins will rack up +20 kills, +8k tower dmg, +1k gpm, +1k xpm, AND still do it all in under 25 minutes. Those large margins take a while to build up. So maybe we can just consider the TIME factor to allow the score to extend to differentiate between regular blowouts and the most extreme cases.

http://dotabuff.com/matches/532732516

1 Kill Score
1 Tower Score
1 Gold Score
1 XP Score
2-3 Time Score?

By your definition with this Composite Score of practically 1 the game was a total blowout in favour of the enemy team.

However I was in a five stack against a five stack and I'm perfectly clear of the reasons why I lost that game. My team was incapable of playing a split push tactic, additionally we had one particularly weak player who last picked Bounty Hunter as a lane support against a duo lane of PoTM + Ogre and fed first blood to them 2 minutes in, giving the opponent an easy lead.

The enemy five-stack had a lower MMR and as a result I lost 31 points for this loss.

From a matchmaking point of view my team was favoured to win. However, I made mistaken assumptions about how my team would play and ended up playing in a style that wasn't suited to the team.

http://dotabuff.com/matches/532863467

With the same stack I give up on playing TB and take Ember mid, snowballing off early kills thanks to rotation and controlled the momentum of the game, getting a much easier game.

Zenoth's example: is it a massive failure of his team? or of the MMR scores that matched them? MMR is supposed to accurately show who is going to win games. That is the overall main objective of having an MMR. Now here we have clearest case for how it is supposed to work, a 5 v 5 game. And because you lost 31 points we have absolute certainty that your teams MMR was much higher on average. To lose 31 points your team had to have an average MMR advantage of at least 200 points. And yet you get totally crushed.

Do you think that normally with these two teams re-matched your team is better and very likely to win? Or does it make more sense to say that your team was overated and the other team underated?

My next game was another one-sided game. I am not surprised that I lost a brood game since I am quite bad at broodmother. But my team fell apart almost immediately despite me killing DK for first blood top, then driving him out of the lane, then killing lycan solo and taking the T1 tower. While I was doing that my team fed like crazy. In this case there is a good explanation. Pudge and Slark were together and admitted that they were drunk.

http://dotabuff.com/matches/532954496
1 Kill Score
1 Tower Score
1 Gold Score
1 XP Score
2 Time Score

Composite Score 1.1. Drunk Vangaurd Slark and drunk pudge ftl. How is Valve supposed to account for that in matching?

What I felt is that if we re-matched and I adjusted my expectations of my team it would have been easy to win. I'm too used to solo queuing where mostly I can expect a more consistent level of play from my teammates.

But that is just my personal impression, it's not like I'll get to find out.

I have a question to zenoth and relentless

http://dotabuff.com/matches/531044554

Played this game yesterday solo queuing, and only get +10 for win. Why only ten? i mean, since i played ranked i never got under 22 points, thats the first time. Are there any specific reason for that or its just a glitch?

off topic but why did the 2 games i play on SEA today have amazing players on my team and all the games where i played on US Servers, i got a bunch of retards

got a bunch of score 1-2 games today :(

i'm removing the 1 game/day condition since nobody actually does it like that anyways, but i'm maintaining the "don't flame/stay calm" while not instalocking a lane. haven't run into a loss streak yet so it should be fine

First 10 games done, let's see what the scores are...I'm gonna edit this post in about 10 minutes.

http://www.docdroid.net/9et2/first-10-games.txt.html

I've only had 1 "3" game so far. None of the other games felt particularly balanced.

We'll see if this distribution changes. However, I'd like to note that intermediate values will be recorded as x.5

For instance, if the GPM difference is +750, halfway between 1000 and 500, I will note that as 4.5 instead of 5. I round to the nearest 0.5, pretty much.

If you don't want to read through all the notes/summaries...

First 10 Games Tally -

5: 2 games
4: 3 games
3: 1 game
2: 2 games
1: 2 games

i've just confirmed that SEA server is not only higher skill than us servers, but you're also less likely to get autists on your team

i mean zenoth said so himself: "i only get about 10% of games that feel like forced losses"

well in the sample of 10 i provided, i got 2 games that were rated "1" without any shred of doubt, and a lot of those "2's" almost felt like forced losses as well. every game on SEA was a 4 or 5.

Get better.

Finished the next 10 (basically massed games)

http://www.docdroid.net/9fb4/second-10-games.txt.html

Zenoth what's the record cs/minute you've gotten on terroblade? Mine's 13.

http://dotabuff.com/matches/534184797

team was feeding most of the game too

also i've only gained 50 mmr in 20 games, so i need to step it up a bit. don't want to be stuck in the retard lite tier forever (4.5-5k)

i also threw one of those games, resulting in a perfect composite "1" score. i'm sure if i didn't throw it would've been a "2" but it doesn't matter. by the way, 4 out of 20 games so far are "3's" (2.5-3.5), so my hypothesis is actually holding up pretty well. unfortunately one of those 3's was in a game against Winter (yes, the SEA pro) whose teammate disconnected and they were forced to play 4v5 against my team for over 30 minutes. i think if their teammate didn't d/c we would've lost and the game would've registered a "4."

There are quite a few 5's and 1's showing up in every 10 game samples and I think this is enough evidence that "edge cases" in Valve's MM are not rare. They happen quite frequently. Relentless was correct. In a few of those games I tilted the games to 5's (and 1's) but it's unlikely the results would have changed if I played slightly worse or better.

http://dotabuff.com/matches/524893710

11.548 cs/min.

Radiance builds will get much more cs but I don't like the buildup and find it rather unnecessary for the most part personally. Less effective for slow sieging especially. Not that it is without merits though.

People on SEA server try harder and are more skilled on average. US servers tend to have at least 2 people drunk or high, making it much harder to do well.

trying harder and being more skilled is 2 things
i dont think there are more than 10 notable competitive players playing on sea servers
almost every good player in that region is going to play on perfect world servers

Uh, I didn't say they are more skilled at the top, but I'm saying that they're more skilled on average.

which is totally meaningless because if you're better you have higher mmr so what relevance does average skill have

Uh no, the average mmr clearly dictates how much the max can be. Imagine if your average player was as skilled as say Dendi. Then, the maximum MMR is not going to be that much higher than the average cuz people won't be that much better than average. I'm saying the MMR in the SEA/China region is significantly lower than MMR in the EU/USA because the average skill is higher so it's much harder to get a very high MMR.

I have about ~27 games in I think, net of 3 wins (4677 mmr).

Side note: Might instalock TB for some ez wins since I can't stand being stuck here anymore.

What do you think guys? 10 games (maybe even more) of TB in solo RMM? I have a 67% win rate with him right now in RMM and a decent K/D.

Relentless said to play the heroes you're good at in RMM to improve rating, so I'm gonna go on a TB picking spree. Let's see if the distribution changes.

Earlier in the thread, someone mentioned a question along the lines of "How come Juice never gets throwers/shitcans on his smurf account?" Well if you're raping really hard or whatever, the throwers/shitcans never surface.

Since Valve switched to an ELO type system for ranked matches... probably for everything but anti-smurf and calibration now. When you do not gain/lose 25 points it means that the average MMR of the team was not equal.

We can get some idea of how unequal it is from the team matching games where the scores of both teams are visible. When you beat a team more than (approximately) 400 MMR below you get zero points. When you play a team of equal MMR you get 25 points win or lose. If you beat a team more than 400 MMR above you then you gain 50 points.

So for a +10 win I would estimate that your team had an average MMR maybe 240 higher than the other team. Individuals could have been 1000 or even 2000 points apart, but the average is that your team was about 240 MMR higher.

My third 10 games are in.

http://www.docdroid.net/9gpo/third-10-games.txt.html

Some interesting things I've noted:

1) I've won a bunch of games where I didn't play my best at all - I fucked up a lot and we still won, solidly.
2) Some the losses that are "2's" could be swung to 3's (and hence, actually winnable) IF you pick intelligently. There was this one game where we had a Zeus and Furion pick right off the bat, and I picked Slark. Opponent picked Antimage and we struggled to kill him all game long. If I had picked smartly (not saying my allies don't share any of the blame, but I certainly wasn't helpless either) then the game would've been a 3 and the chances of winning would be something like 50% instead of 30%.
3) By picking a hero you are strong with, you can swing games that are 3's to 4's. Similar concept to above.

There were no "3" games this time since I started playing heroes that I am actually better with (terrorblade). I did lose 1 game because the opponent's Ember got too fed, but that's OK. Winning 2/3 games with TB is good enough for me.

So far the distribution of 1's every 10 games hasn't changed...at all. On average, in every set of 10 games (for me), there are 2 "1" games, i.e. games you are extremely unlikely to win. Relentless estimated that every 1/5 games in ranked is a "forced loss" (once again, assuming you are playing at a level comparable to your current MMR) and this is quite accurate. On the bright side, everyone gets "forced" (ez) wins.

I think what I'm going to do now (and this should have a positive impact on the rest of the ~70 games) is to start expecting these "1" games and not to let them get to me.

Of some minor interest to this topic I have found, experimenting over the last few days, that small MMR ranges are possible IF you expand other options. Essentially I found all those other people who are never in my AP ranked games.

When you select servers you see that average que times for each server range from 1 to 2 minutes. But if I que for AP ranked in my region it will take at least 3 min and sometimes as long as 7 min. This remains the same whatever server I choose. Yesterday and today I've been selecting lots of servers and all modes and I often get instant que and ques less than 30 seconds. What this means is the vast majority of dota players are not using AP mode ranked.

So because few people are using AP ranked, canceling the que is required to maintain a small MMR range for your game. There may be an average of 350 players entering the que (globally) in a 30 second window. But they are not playing AP ranked. I've found that there must be huge number of player using All Random and Random Draft modes since those are very likely to generate que times of a few seconds. Even though my que times for All Random might be 3 seconds I will still end up as blue/pink suggesting that everyone in the game had a nearly instant que time.

If you cancel que and reset every 2 minutes... we would expect an average 1400 players to click "find match" during that 2 min. Maybe 10% to 25% would be in your selected server region (140 to 350). But it could be only 10% select AP ranked (14 to 25). These few are unlikely to contain 10 players near your MMR. So you can either wait for the MMR range to expand until people 2,000 MMR below you are added to your team... or you can cancel the que until you get lucky and enough people in the que are within 500 MMR.

Of course we won't know how large the MMR ranges really are compared to that search range unless Valve reveals a lot more information. Having a small MMR range does not assure that the game will be close. But it does make it very likely that everyone in the game are at least capable of competing, of putting in their fair share of the effort to win if they pick correctly and attempt to play well and cooperate. Maybe they won't care that much. Maybe they will pick crap and still get steamrolled. But at a large MMR range some players will be incapable of doing anything but feeding.

http://dotabuff.com/matches/536392563

I watched this game from every players perspective. Two of the players were totally out of their league on the hero they were randomly assigned. Both TA and ES had absolutely no idea how to play their hero. They both had to spend over a full minute of game time READING the description of their skills trying to decide what to do. But the ES laned with me and I dominated the lane by myself so he started ok. While TA's lane got smashed and he never got off the ground. Both of these players had no chance of playing at the level of this game... no chance of doing anything more than accidentally hitting some small % of their skills. When a player is way below the average MMR of a game (on the hero they got), they can't contribute in a difficult situation. This TA showed every indication that he wanted to win the game and was paying attention to the map, actively trying to fight, watching the mini-map, thinking about how to win... but was completely incapable of using his hero. Judging by his mini-map awareness he is probably a very experienced pub dota player who just never plays TA.

Have you considered that 1s or 5s are not because you are extremely unlikely/likely to win, but because the playstyle of the winning team is through a snowball fashion?

My last 10 wins were practically all either 4/5s. Winning the game usually guarantees a significant additional gold lead through the destruction of the structures in the enemies' base, and one's team is only likely to win if said team has the advantage in kills, GPM and XPM. Tower damage is necessarily in favour as well due to having to push.

Take my TB, for example. Every game I win I average 700-800 GPM and 10k+ Tower Damage. This easily swings every game I win into a 4/5 by this rating system. However, I don't feel that it was because my team was very likely to win, but rather that the playstyle I used to win results in the rating being very skewed to one direction. Most of the games were quite close up to some point where the enemy team lost a bad fight and got punished by a set of rax or got split pushed without being able to respond appropriately.

Anyway, @Relentless - in SEA Ranked people only seem to queue AP ranked. I get 3-5 minutes of queue times at my rating for AP only, and once spent 20 minutes when I checked every other mode. I added one of the players afterwards and he was 2000 MMR below me. Perhaps it's regional differences, but practically all the high rated players I know only play AP Rankled.

http://dotabuff.com/matches/537188776

Terrorblade easily outcarries a six-slotted AM who eventually sold his Bfury for MKB in order to even touch my illusions. 3 Butterflies build, 72.5% evasion, 70% damage reduction from armour on top of 2547 hp, and they literally have to spam ship/torrent/chain frost/concoction to even damage my illusions when I send them in to backdoor, only for the next illusion to come 16 seconds later.

That's considering I was shut down by a trilane of Alch/Lich/Visage and had no cs for the first 5 or so minutes, lacking in exp and gold. TB's catchup farm, however, is so fast that it's ridiculous - even AM with a significant gold lead and all the space to farm couldn't farm as fast as me.

This game my team was behind in networth practically all the way until I fountain dived them at the end.

First close win for me in quite awhile.

Many of the 1's and 5's are due to picks, but also due to people playing improper roles. If your orange/brown player instalocks mid/carry, then the chances of getting a 1 suddenly skyrockets. That's not something Matchmaking can fix (I suppose you can blame Valve for making MMR values hidden) so I don't get too worked up about it.

Top 3 reasons for 1 games:

- shittiest player on ur team instalocks carry/mid
- supports doing nothing (yes, i have people who pull for 8 minutes or just sit behind me and auto-deny while offlane gets level 6 more quickly than me, the safelane carry, and then wonder why lanes aren't winning)
- bad picks overall, lack of counters to a specific snowball or lategame hero

those are the 3 main problems i've encountered

the game you just played doesn't surprise me tbh. i've supported offensive trilanes before and then wonder why the carry of that trilane is less farmed 30 minutes later compared to their carry. it is likely you are just more skilled than they are (and possibly your team did a good job creating space after lane phase)

First 1 game of games 31-40: http://dotabuff.com/matches/537239677

Look at those picks. I don't even know how to accomodate (I last picked by the way) them. Tinker/Furion is useless in teamfights until late game, we have 1 grand stun total (bane), and the lanes are messed up. Veno pulls all the time and gets solo killed by cw twice, something I have no control over. I made some mistakes later on due to the stress getting to me (cuz I knew these players are all considerably lower skill than I am, and I'm probably going to dodge them in the future) and the unlikeliness of being able to carry, but I'm not angry. Games like these are inevitable.

Congrats on 5.3k again, it's been a while lol.

Zenoth your experience makes sense but you drew the wrong conclusion from it. You had to wait for a 2k MMR difference, a huge search range, to get a game in the other modes. But that is not because no one is playing the mode... its because no one near 5.3k MMR is playing the mode. Nearly ALL players in SEA, just like in other regions are casual players far far below your MMR. And they are playing non-ranked "fun" modes.

I am currently 2k MMR below you. I will get a near instant que on SEA for All random or Random draft. AND I will be blue/pink in the game. But you have to wait for the search to go all the way out and then wait more for the very unlikely and rare event of a 5.3k MMR player trying those modes to balance against you on the other team. This is because of Valves "goal #4".

http://dotabuff.com/topics/2014-01-01-and-matchmaking-keeps-sucking-even-worse?page=4 <=== for any who missed the discussion of matchmaking "goal #4".
---------------------------------------------------------------------------------

Blowout wins happen in dota. Pro teams who can beat each other will sometimes trade games where each one stomps the other... but as you say MANY of these pub games are 1s or 5s. In fact about 40% of them are. 20% you stomp. 20% you get stomped. Yet Valve is apparently trying to set up "3" type games.

40% of games between teams of near equal abilities should not be blowout wins. The reason this is happening is matchmaking failures.

when people pick high farm heroes (furion/tinker) that aren't super useful in teamfights at least until they get some major items and end up dying all the time to ember + naix/clock + naix bomb, it's pretty hard to not get a "1" game. i agree valve "can't control what heroes people want to pick" so that's not something i am blaming MM on. it is the players' faults for picking like shit.

what i'm trying to convince people (and the results strongly suggest this already, if you haven't noticed) is that these games have systematic patterns. they are not random. if they were, i might get 9 "5" games and 1 "2" game in the first set of 10 games, and in the next set of 10 games I might get all "1's." But that clearly isn't happening, and I don't think it will. That will only happen if someone is severely underrated/overrated.

Next 10 in (40/100 completed)

http://www.docdroid.net/9hvn/fourth-10-games.txt.html

Number of "1" games still remains unchanged. (the only factor that seems constant in every set of 10 games)

Honestly, I think that blowout games are normal and an expected consequence between balanced teams.

Don't forget that most pros call gg fairly early. If the game was allowed to go on until the ancient was actually killed, the stats would likely be extremely skewed as well.

All those "forced wins" or "forced losses" are really games that could have gone either way. I have stomped and gotten stomped by the same stacks many times. The very nature of my playstyle and picks ensures most of my games are either 1 or 5 because I play a farm oriented greedy style with a fair bit of aggression. Either I snowball alot or I get neutered early.

Back on my losing streaks I had like 2 pages of "1" games, and on a good day almost all my games are 5.

My point that matchmaking is incapable of forcing wins or losses still stands. This would be a feature difficult to implement if it was even possible, when you consider that all MM looks at is simply the MMR alone.

Ok Zenoth you contend that blowout games are normal and expected between balanced teams? I will definatively prove to you and everyone that pro teams who are remotely close to the same caliber have only a tiny fraction of their games result in the dramatic blowouts much more regularly seen in pub games. That is composite balance scores lower than 1.5 or higher than 4.5 are rare for truly fair games.

Let's consider the first 40 TI3 games as a sample. We can then judge just how far apart teams are in dota 2 skill to end up with various balance scores.
And before you say "games don't all go to completion"... I'm leaving the time score as it is. So calling gg to end early will give a bigger blowout time score. That's the purpose of the time score anyway. If you won fast, it was a bigger win.

Based on their placement in the Main Event I am assigning the following ranks. Teams eliminated in the same round receive the lowest rank possible for that round (#1) Alliance (#2) Navi (#3) Orange (#4) Tongfu (#6) DK (#6) IG (#8)Fnatic (#8) Liquid (#12) Dignitas (#12) LGD.INT(#12) Zenith(#12) LGD.CN (#16) MouseSports (#16) Rattlesnake (#16) MUFC (#16) VirtusPro

All scores from the perspective of the higher ranked team or Dire in the case of "balanced" games. This will show the predictive power of the ranking or lack thereof.

http://dotabuff.com/matches/262276878 DK (#6) vs Dignitas (#12)
4 Kill Score
3 Tower Score
4 Gold Score
4 XP Score
5 Time Score
Composite Score 4.0

http://dotabuff.com/matches/262279207 Fnatic (#8) vs Navi (#2) UPSET
2 Kill Score
2 Tower Score
3 Gold Score
2 XP Score
3 Time Score
Composite Score 2.4

http://dotabuff.com/matches/262281327 MUFC (#16) vs LGD.CN (#12)
4 Kill Score
4 Tower Score
4 Gold Score
4 XP Score
4 Time Score
Composite Score 4.0

http://dotabuff.com/matches/262482014 Zenith (#12) vs Fnatic (#8) UPSET
3 Kill Score
1 Tower Score
2 Gold Score
3 XP Score
3 Time Score
Composite Score 2.4

http://dotabuff.com/matches/262475417 DK (#6) vs LGD.CN (#12) UPSET
2 Kill Score
3 Tower Score
3 Gold Score
3 XP Score
3 Time Score
Composite Score 2.8

http://dotabuff.com/matches/262467064 Navi (#2) vs Mous (#16)
4 Kill Score
4 Tower Score
4 Gold Score
3 XP Score
4 Time Score
Composite Score 3.8

http://dotabuff.com/matches/262457078 Zenith (#12) vs Fnatic (#8)
4 Kill Score
4 Tower Score
5 Gold Score
4 XP Score
4 Time Score
Composite Score 4.2

http://dotabuff.com/matches/262445112 LGD.CN (#12 vs DK (#6)
4 Kill Score
5 Tower Score
4 Gold Score
4 XP Score
4 Time Score
Composite Score 4.2

http://dotabuff.com/matches/262438280 MUFC (#16) vs Dignitas (#12)
4 Kill Score
5 Tower Score
5 Gold Score
4 XP Score
4 Time Score
Composite Score 4.4

http://dotabuff.com/matches/262427613 IG (#8) vs Rattlesnake (#16)
3 Kill Score
3 Tower Score
4 Gold Score
3 XP Score
5 Time Score
Composite Score 3.6

http://dotabuff.com/matches/262426855 LGD.INT (#12) vs Alliance (#1)
4 Kill Score
3 Tower Score
4 Gold Score
4 XP Score
3 Time Score
Composite Score 3.8

http://dotabuff.com/matches/262411148 Dignitas (#12) vs MUFC (#16)
4 Kill Score
4 Tower Score
4 Gold Score
4 XP Score
5 Time Score
Composite Score 4.2

http://dotabuff.com/matches/262392411 VirtusPros (#16) vs Liquid (#8)
3 Kill Score
4 Tower Score
3 Gold Score
3 XP Score
3 Time Score
Composite Score 3.2

http://dotabuff.com/matches/262385766 Alliance (#1) vs LGD.INT (#12)
4 Kill Score
4 Tower Score
4 Gold Score
4 XP Score
3 Time Score
Composite Score 3.8

http://dotabuff.com/matches/262374910 IG (#6) vs Rattlesnake (#16)
4 Kill Score
4 Tower Score
4 Gold Score
4 XP Score
4 Time Score
Composite Score 4.0

http://dotabuff.com/matches/262367754 Tongfu (#4) vs Orange (#3) UPSET
2 Kill Score
2 Tower Score
2 Gold Score
2 XP Score
2 Time Score
Composite Score 2.0

http://dotabuff.com/matches/262357611 Liquid (#8) vs VirtusPro (#16) UPSET
2 Kill Score
1 Tower Score
1 Gold Score
2 XP Score
2 Time Score
Composite Score 1.6

http://dotabuff.com/matches/262341896 Orange (#3) vs Tongfu (#4) UPSET
2 Kill Score
2 Tower Score
1 Gold Score
2 XP Score
1 Time Score
Composite Score 1.6

http://dotabuff.com/matches/262324734 Navi (#2) vs Fnatic (#8) UPSET
1 Kill Score
2 Tower Score
2 Gold Score
2 XP Score
3 Time Score
Composite Score 2.0

http://dotabuff.com/matches/262322091 Zenith (#12) vs Mous (#16)
3 Kill Score
4 Tower Score
4 Gold Score
4 XP Score
3 Time Score
Composite Score 3.6

http://dotabuff.com/matches/262312446 LGD.CN (#12) vs MUFC (#16)
4 Kill Score
3 Tower Score
4 Gold Score
4 XP Score
4 Time Score
Composite Score 3.8

http://dotabuff.com/matches/262303778 Dignitas (#12) vs DK (#6)
3 Kill Score
3 Tower Score
4 Gold Score
3 XP Score
5 Time Score
Composite Score 3.6

http://dotabuff.com/matches/262283296 Mous (#16) vs Zenith (#12) UPSET
3 Kill Score
1 Tower Score
2 Gold Score
3 XP Score
3 Time Score
Composite Score 2.6

http://dotabuff.com/matches/262613112 IG (#6) vs LGD.INT (#12)
4 Kill Score
3 Tower Score
4 Gold Score
4 XP Score
5 Time Score
Composite Score 4.0

http://dotabuff.com/matches/262604402 VirtusPro (#16) vs Orange (#3)
4 Kill Score
4 Tower Score
4 Gold Score
3 XP Score
3 Time Score
Composite Score 3.6

http://dotabuff.com/matches/262603145 Fnatic (#8) vs Mous (#16)
4 Kill Score
5 Tower Score
4 Gold Score
3 XP Score
4 Time Score
Composite Score 4.0

http://dotabuff.com/matches/262596112 Liquid (#8) vs Tongfu (#4) UPSET
2 Kill Score
2 Tower Score
3 Gold Score
2 XP Score
3 Time Score
Composite Score 2.4

http://dotabuff.com/matches/262584768 Navi (#2) vs Zenith (#12)
4 Kill Score
5 Tower Score
4 Gold Score
4 XP Score
4 Time Score
Composite Score 4.2

http://dotabuff.com/matches/262580105 MUFC (#16) vs DK (#6)
5 Kill Score
3 Tower Score
5 Gold Score
4 XP Score
4 Time Score
Composite Score 4.2

http://dotabuff.com/matches/262559167 Navi (#2) vs Zenith (#12)
4 Kill Score
5 Tower Score
4 Gold Score
4 XP Score
3 Time Score
Composite Score 4.0

http://dotabuff.com/matches/262546824 IG (#6) vs Alliance (#1)
3 Kill Score
3 Tower Score
3 Gold Score
3 XP Score
3 Time Score
Composite Score 3.0

http://dotabuff.com/matches/262535265 Dignitas (#12) vs LGD.CN (#12)
2 Kill Score
3 Tower Score
2 Gold Score
2 XP Score
2 Time Score
Composite Score 2.2

http://dotabuff.com/matches/262531941 LGD.INT (#12) vs Rattlesnake (#16) UPSET
2 Kill Score
3 Tower Score
3 Gold Score
2 XP Score
3 Time Score
Composite Score 2.6

http://dotabuff.com/matches/262520933 Orange (#3) vs Liquid (#8)
3 Kill Score
4 Tower Score
3 Gold Score
3 XP Score
3 Time Score
Composite Score 3.2

http://dotabuff.com/matches/262520546 IG (#6) vs Alliance (#1)
3 Kill Score
5 Tower Score
4 Gold Score
3 XP Score
3 Time Score
Composite Score 3.6

http://dotabuff.com/matches/262511470 LGD.INT (#12) vs Rattlesnake (#16)
4 Kill Score
3 Tower Score
3 Gold Score
3 XP Score
4 Time Score
Composite Score 3.4

http://dotabuff.com/matches/262509769 Tongfu (#4) vs VirtusPro (#16)
5 Kill Score
3 Tower Score
3 Gold Score
4 XP Score
4 Time Score
Composite Score 3.8

------------------------------------------------------

RESULTS: 40 TI3 groupstage games

ZERO games were extreme blowouts (less than 1.6 or more than 4.4)

15 close games (between 2.4 and 3.6)

17 advantage games (2.0,2.2,3.8,4.0)

8 large advantage games (1.6,1.8,4.2,4.4) (also all UPSET games where a surprise strat dominated the game)

---------------------------------------------------
What this shows is that even with team as far apart as Alliance vs Rattlesnake you still do not get scores as lopsided as often show up in pub game scores.

http://dotabuff.com/matches/262650433 Alliance (#1) vs Rattlesnake (#16)
3 Kill Score
3 Tower Score
3 Gold Score
4 XP Score
5 Time Score
Composite Score 3.6

Alliance was during TI3 vastly superior to Rattlesnake. I'm sure bets for this game would be 95% for Alliance. Yet they do not totally stomp on them because Rattlesnake is still at least on the same general skill level.

In contrast this game I just played features my team of 5 solo players totally outclasssing the 3 solos and group of 2 on the opposing team. MMR was equal? If so the rating for some players must be way off the mark.

http://dotabuff.com/matches/537705693
5 Kill Score
4 Tower Score
5 Gold Score
5 XP Score
4 Time Score
Composite Score 4.6 Total domination. Watching the other team's perspetive, they really are trying to win. They are just completely outclassed by the opposing players.

Matching failed. This is a forced win.

Your time factor doesn't offset alot of things.

A game where one team would already be assured of a win extended by 10-15 minutes will result in a much more significant skew. In that 15 minutes, it would not be unreasonable for the team to get 10 more kills, an extra 4k tower damage, 500 GPM and 500 XPM from said kills and towers. This is at the cost of one point for the time factor.

Of course, it is hard to extrapolate for sure. But for example, in the game between Alliance and Rattlesnake you pointed out, if Rattlesnake didn't call GG at 21 minutes Alliance would have most likely gotten 5 in every metric except Time. Alliance did not even get to push much at that point, and Loda's PL was just at the point where his farm would begin to accelerate significantly and come closer to the ~700 GPM that he usually gets on PL when the game goes late. S4's storm would have likely gotten even more pickoffs, pulling the gold and XP lead further ahead. Given 15 minutes more, I am fairly sure it is reasonable to assume everything would have been 5 except for Time, and 15 minutes is a reasonable assumption for how long it would take Alliance to end the game if Rattlesnake didn't call GG, since none of their heroes are particularly adept at sieging the base.

The thing about pro games is that they can usually recognize the point at which it is extremely unlikely to win, and cut their losses and call it a game at that point. Of course the scores won't seem so skewed. But pubs are different. There will be those who try to defend futilely and feed kills, or get fountain dived. The t3 and t4 towers added up also make alot of difference both GPM wise and Tower Damage wise.

On the other hand, look at the game you just posted. Rather than saying matching failed, why not look at the fact that their picks were terrible? Morphling, Luna and Medusa are all heavily farm-dependent cores, with the latter 2 renowned for being especially squishy once the team starts to fall behind. As long as that lineup loses the laning stage or doesn't come out of it dominant, the 3 cores will not be able to find much space to farm while being extremely liable to pick offs, without even considering that your team has much stronger nuking power. Picking is part of winning. No matter how hard they tried, when they made the decision to random or pick badly it contributed to their loss heavily. If I was playing on that lineup I myself would likely struggle to even find any farm. The difference is that I wouldn't have made those kind of picks had I been in that team. But even that is assuming knowledge of the picks before hand. You just need one last player to make some kind of stupid pick, and your team is severely handicapped right from the first horn.

My personal opinion is that picks are a significant part of the game, sometimes even more important than your actual level of play. This is a large part of why I win so much with Terrorblade, because as a hero he is not easy to play against and requires a high level of coordination and teamwork to deal with his split push. This allows me to first pick him confidently and let my teammates pick around it, without having to worry too much about being counter picked. Even when the opponents offensive tri-lane and zone me out from the creep wave for 10 minutes I still can easily get 700 GPM if they don't manage to maintain the pressure mid-game or push and end early.

Put some thought to this: how is it even possible for matchmaking to "force" a win or loss? If I simply give you the MMR of all the players on both teams, can you even predict which team will win? Before the game starts, it is extremely unlikely. Given 100 games you would probably guess right only 50% of the time. The system has to match games based on that information alone. It doesn't have the capacity to "force" wins or losses. That would be giving it far too much credit, ironically enough. A matchmaking system that can force wins or losses would actually be really impressive.

Any advice on dealing with ember spirit as TB? He seems to wreck illusion heroes.

Split push. He needs decent cleave to kill your illusions, and as long as you don't send them in with creeps or clumped up he struggles to deal with them when sleight is down.

A good Ember will make it much harder, though - but TB will force him to stay in base if you can pressure enough lanes.

http://dotabuff.com/matches/533488027

I was the Ember against TB this game and while my loss was due to various other factors (Rubick dc-ing every 2-3 minutes), the biggest challenge I faced as Ember was being unable to push out. Even with Ember and Tinker combined it was not enough to simply kill the illusions as TB can pressure the lanes so hard that the opponents are not able to push out.

Zenoth in no time people are going to start calling you TB picker :P , btw do you always go phase on him ?

Depends on the laning early. Phase gives you alot more killing power in lane, and makes last hitting easier against a contested lane. When playing from behind, though, I prefer to get Treads for HP until I start jungling and switch to agi. The difference is moot, really.

I only started getting Phase after getting stomped by Meracle with it a couple of times. The active actually scales very well, and Drums active + Phase is really good at getting away from most TP reactions.

I see, i like this hero but i suck too hard to play with him.

welp i just ran across a loss streak. Lots of 1's and 2's.

if there's anything i've confirmed so far, it's that your current MMR has little/no correlation with how well you pick. I had a "5300 MMR" (or so he claimed) picking prophet against a lineup of dazzle/abaddon/qop/night stalker/void (for some perspective, prophet cannot even kill dazzle solo cuz dazzle can just grave + tp, let alone any of their cores) , Demon (yes, former EG player) picking Jugg against the same lineup, and then each blaming the other for failing even though THEY ARE BOTH TO BLAME. The 5300 MMR refused to pick a support (he said over mic "I don't want to support") and then proceeded to afk in jungle mostly and wonder why the team didn't win the game for him.

oh and he also said "i rarely play prophet" about 5 minutes in when Demon was pinging him to help out safe lane, if anyone still doubts his intentions. obviously, being the 5300 MMR player he is, why would he want to support us peasants?

here's the game if anyone's interested.

http://dotabuff.com/matches/538054837

anyways i'm done (not gonna waste more time solo queuing - i have lots of work to catch up on, plus solo is stressful and even if i get a lot better and reach 5500 MMR, I can look forward to those types of games!), so i think i'm gonna stack with my > 5k friends every now and then and get some ez wins. 50 games is less of a sample size than 100 games, but it should suffice.

Games 41-50: http://www.docdroid.net/9jcc/asdf.txt.html

Gonna stop here, I don't think the next 50 games are going to look significantly different from the first 50, and I CBA to determine the rest.

Total Games by Category/Percentage of Total:

5 aka blowout wins: 9 games / 18%
4 aka easy wins: 12 games / 24%
3 aka balanced games: 8 games / 16%
2 aka hard losses: 7 games / 14%
1 aka blowout losses: 14 games / 28%

1 + 2 games account for 42% (VERY CLOSE to 40%)
4 + 5 games account for 42% (VERY CLOSE to 40%)
3 aka the balanced games account for 16% (Not too close to 20%, but within a reasonable margin of error)

I won 5 out of the 8 balanced games, gained about 64 mmr from doing so in 50 games (I won 2 games that were slanted in favor of other team, gaining > 30 mmr for each of both games)

You can deduce from this what you will. I'm usually rated between 4600-4800 so these games should be pretty repesentative of the rest of my games, unless I get significantly better at playing 2-4 heroes and pick them in -ap over and over, which is what people who want to grind MMR typically do. I don't think this distribution should change even if you're in a different MMR bracket (4-4.5k, 3.5-4k, etc.)

To be perfectly honest, the last 10 games didn't have to look like that. I could just have gone on a terrorblade picking spree and won more then half of them. Choosing to play heroes I don't play often at all caused me to lose a bunch of games, but the primary culprit for lost games (at least in the 4.5-5k bracket) is still prima donna symdrome, sprinkled on top of dunning kruger with tiny dick syndrome. Basically everyone wants to play carry/mid cuz they're the special cherries on top of all the nubcakes, and if anyone has the gall to pick their heroes/lanes (gasp), they decide they're "too good" to support and go jungle instead. Unfortunately, Valve MM CANNOT fix player's attitudes, so I agree with Zenoth on this point. Personally if I was in charge of this system, anyone who picks something like "Jungle naix" or "Jungle prophet" or something ridiculous like "Jungle Wraith King" (lmao) after the other team already has a lineup that strongly suggests an aggro trilane and then causes the team to fail should get -2000 MMR since that's essentially the type of player they come off as, but I won't get into any fantasies.

Other than this, I think the experiment confirmed Relentless' points about how most games are not actually balanced. What Valve does is give you a bunch of games that are lopsided, and then in order to increase your MMR, you have to win more than 50% of the balanced games (which constitute about 20% of your games.)

Recall the reasoning behind this: Since you are 20% of the team, in the long run, you will influence about 20% of your games.

This results in a slow increase for those who are playing slightly above their level; whether or not players agree with this is another matter altogether. Does this mean people can climb out of their MMRs? Yes, but you have to wade through a bunch of games that are going to be shitstomps. Assuming I won all 8 balanced games, I would have + 200 MMR, 4600 -> 4800 in 50 games. The typical player who plays dota casually (10-25 games per week) could need anywhere from 2 weeks to 5 weeks to get a rating gain of a measly 200 points.

Then there are those who gain astronomical amounts of MMR within a short time, (I've heard stories of people getting from 3.5k to 5k since ranked MM came out) but they do so by picking the same few heroes over and over. I guess that's OK too, if you don't get bored of it.

tl;dr almost everything relentless said in other threads pertaining to the same topic is accurate.

i also followed waves' advice and played some on SEA/1 game on EU. The players were certainly better than the players I get on US West/East, but the games didn't skew the distribution of the aggregate at all.

There is one glaring problem in what you and Relentless agree upon.

Yes, there is a correlation. But there is no causation.

Answer me this, then. How does Valve give you a bunch of games which are lopsided? Based on the only metric for skill which they use for matchmaking, your rating, Valve has no way of telling who is favoured to win unless one side has a higher rating. In this case they compensate for it by adjusting the points gained or lost. For what you say to be true, before any player even picks the game has to be already in favour of one team despite the average rating on both teams being the same. Yes, the spread is not necessarily good, and there are other potentially bad things that ruin games, but on average both teams have it as bad. How can one team be favoured to win at all?

Necessarily if anyone plays a bunch of games at random some will be stomps one way, some will be stomps the other and some will be in between. This is the product of simple random distribution alone, and not because matchmaking dictates it. If you adjust the metrics to account for early gg calls the same will apply to pro games as well.

There are 2 possible explanations I can come up with:

1) The inherently imbalanced nature of -ap. I used an analogy a while back: if you are a 5200 MMR player, currently rated at 5200, you will win approximately 50% of your games. It's like a coin toss experiment with each experiment consisting of 50 tosses - if you do the experiment enough times (let's say 50 tosses for each experiment, 50 times) you will get an average value of 50. The difference between dota matches and coin tosses is that each individual dota match is not 50/50. It could be 75/25 for one game, and then the next game 20/80. The expected value remains the same, but each "toss" aka each "game" does not have the same 50/50 value.

Why? It's simple - some heroes are just awful right now: heroes like medusa/drow are usually free losses in solo queue, nature's prophet/tinker require LOTS of farm yet contribute very little to teamfights until then and easily become liabilities to the team (and let's face it, most people even around 5k don't play them properly), etc. I get a bunch of players who pick shitty carries like Phantom Assassin, go bfury, and wonder why they don't do any damage in teamfights (hint to the retards: critical procs rely on attack speed just as much as attack damage). In fact, half of the dota hero pool is off-limits if you want to win in solo queue, but players still pick them because in the end, dota is a game meant to be played for FUN, to all but a few (the pros).

What happens when you get awful picks? You're gonna get awful games. That's just -ap in a nutshell.

2) Information asymmetry is everywhere. Everyone's MMR suggests something different. Look at the most recent 20 games on my dotabuff record. I played 15 unique heroes in 20 games, and over half of those heroes I rarely play. This strongly suggests that even though my current MMR is 4664, my MMR on those heroes is at least moderately lower.

This is the other ugly side of the problem. If most people will only play x number of heroes in their favorable pool of heroes, and then those heroes (or those roles) are already picked, what can they do?

Valve hides a considerable amount of information. I don't know what the MMR values for my teammates are, so I can only guess. I don't know what heroes they are strong/weak at, so I can only guess and try to accomodate them (if I'm not in a selfish/greedy playstyle mood), and pray they don't feed or suck at last hitting.

In the end, the solution for me is simple: Don't place too much weight on the MMR number. Am I worse than the typical 5500 player? Almost certainly. I can play as well as a 5000 MMR player on a good day, but not consistently so, which is why I am not rated 5000. I am not under the illusion I am "underrated" or whatever, since I have never claimed that. But I also acknowledge the fact that people who grind MMR by picking Invoker/(previously Earth Spirit)/Lycan/etc. just to go up 400-500 points are no better players than they were before, and those who think otherwise are deluding themselves. Being good at dota =/= picking the strongest heroes in -ap pubs.

As for 1), this isn't an issue about matchmaking. Blame it on the game mode, blame it on retards picking bad heroes, blame it on selfish assholes who don't want to support, but the match doesn't go 75/25 until after people make their picks. Yes, it sucks a lot, and it results in many games that feel out of your control no matter how well you play. But to be clear this isn't a matchmaking issue.

2) is just the result of having multiple heroes in the game. Unless matchmaking implemented a system where everyone has a specific rating for their heroes and you pick a hero before you queue, this issue can't possibly be fixed as well. With the exception of Earth Spirit in particular, from my experience if you can do very well on one particular hero many of the skills required to do so will carry over to other heroes, such as awareness, micromanagement, last hitting etc. Most factors that contribute to how well one plays are basic mechanics not unique to that particular hero. Take my games with TA for example. I have won lanes against competitive players, and am sitting on an extremely high win rate for her considering her pub win rate is a terrible 45%. But I acknowledge that a large part of it is because I like to pick TA late after others have made their picks, resulting in an easy match-up for me without specific hero counters on the other team. This, however, is only possible in part because I have relatively good laning mechanics - I have taken many unconventional heroes mid and performed decently with them. This shows 2 things - picking alone plays a significant part in my win rate, and that being good with one hero helps with playing other heroes to some extent.

You're right, the number itself doesn't mean a lot. It is an approximate for the system to make games which are balanced by this number. But for a system to "force" wins or losses - this means that the game is imbalanced before the picks even begin. And like I said, this is impossible for a system to do when it matched by rating alone.

As Relentless has said, there are definitely bad games. Games where the MMR spread is very big from queuing for too long. However, this applies to both teams as well. It simply makes the game more likely to turn into one-sided stomp based on whether the worst player picks a core hero or which team can capitalize on the opponent's worst player more. However, until the players have picked their heroes, the game is still equally likely to go both ways. Even these bad games don't "force" one team to win as he puts it. It simply makes the game more likely to devolve into a one-sided stomp.

Perhaps I didn't communicate my point properly before, but my goal was to analyze the results of the MM system, not to prescribe solutions. When I use the term "forced," I don't literally mean that you are 100% guaranteed to lose/win before the game begins. What does bother me a bit is that games are immediately likely to be lopsided after pick phase is over (and then you have to spend the next ~40 minutes duking it out, even though one team is already favored to win).

It's kind of a tongue-in-cheek/shortcut term for extremely imbalanced game.

I won't dispute that there are things MM cannot fix. In every MM dota 1 system I've played on before: dotalicious, RGC which was notorious for stacks vs solos all the time in their "high skilled" (ROFL) leagues, the same problems surfaced. Dota 2 MM is no different.

I will just conclude that: based on a combination of the factors we have already discussed, -ap games are going to roughly end up with the distribution (40-20-40). It's possible the distribution could change as you move up/down the ladder, but I don't think it would change significantly.

As for your claim that games are almost always balanced before picks, I assume you are correct, but Relentless is the one you should be arguing with. I have never claimed the games are lopsided before picks - just that games being lopsided immediately after picks is extremely bothersome.

I contend that given you are winning half of your games on average, that would be an extremely likely distribution. What I am trying to contest is Relentless's definition of "forced", because he is blaming matchmaking for unfair games, and you seemed to agree with his points. I think it is a pity that someone like him who is at least relatively rational and capable of reasoning is blaming matchmaking for problems that are inherent in the game and the players itself.