Why the system sucks

hmmmm i probably shouldnt have been so flippant in my dismissal of this thread. my apologies.

first ill answer primarchXIII

- the reason you have that is because someone on thier team has a lower mmr, which in turn drags down the average mmr so they balance that by matching a 300-400 high mmr player in with that team to make the average mmr closer to each other for both teams. also you think 300-400 is bad? on my ranked games its normal for there to be a player that is thousands ahead of the nearest one. my most was a 3.9k mmr average on both team but they had a 4.9k mmr player on there side... yea... "balanced"

ok my "more thinking thinkful thinky thoughts" on mmr.

on the system:

its broken but its the best system that there possibly could be. imagine if we had a system like proposed above? ^ hmmm scary stuff or my personal hell would be if mmr was based upon k/d/a only... oh god the horror. the problem of "balance" arises when people play out of their minds and still lose due to them being placed with someone who feeds or plays horribly. it can be hard to take when you do so well but you still lose,it hits you harder then anything. i had a game on QoP where i went 32-1-19 and our jugg went BoT-Sange-halbred-Dagon. at high 3 k mmr. imagine how salty i was. hint... the red sea had nothing on me.

i feel honestly that a recalibration could be viable but i would say make it "opt-in" and only available once a year.


On Smurfs:

a smurf is exactly like a normal account. within 30ish games you will be back to your actual skill level. normal account unranked high or very high, ranked is normal though. now im sure everyone already knows my story but i will reiterate it.

at lvl 14ish ranked was released. did my 10 games over one weekend got a 1.7k mmr. didnt do ranked for 6 months went back and won 80% of ranked games to get to 2.2k ish. went away and just played then after i saw my unranked games were being placed in high skill i was like hmm ok ill go back to ranked. won 3/4 games and arrived at 2.8 honestly it was taking its toll, having to deal with people that didnt take ranked seriously, or just fed, was annoying and draining and noticing that i won 100ish games just t grind 700-800 mmr was soul crushing.so i did the next best thing. i created a smurf.

got to lvl 13 did my ten over a week or so. got calibrated at 3552 mmr. it took me about a while but instead of having to deal with hundreds of soul crushing games i could play relatively happy and have fun winning with out the ranked stress. 10 cali games later. i raised my mmr by close to a thousand. now it currently sits at 3780k something and when i do decide to play on my smurf i win . seriously. ive lost like 5 maybe 6 games in total on my smurf in ranked.

thats all a smurf does. it reduces the time one requires to receive an accurate skill level.

if you are bemoaning your mmr and saying im so better then my actual lvl. make a smurf. if after 50+ games ur still at that level then there is nothing you can say except. this is my skill level (normal, high, very high, **** off you pro)




TL;DR :

Smurfs give you your actual skill level quicker and more accurately, if your ****, your ****, if your good, your good.

The system is broken but its the best one we could possibly have and until someone comes up with a better one.... live with it. or make a smurf and still be **** or good ;)

Dota isnt a game.... Doto is lyfe

As Sando said once, you win 10% of your games because the enemy team is terrible. You lose 10% of your games because YOUR team is terrible. And then there is the 80%, the games where you can positively or negatively affect the outcome of the game. Let's face it, the reason the mmr system is the way it is because, like it says on the loading screen "don't make more enemies than the 5 that you are already facing" ;) Valve was trying to create a simplistic ranking system that rewards you for winning, and doesn't reward you when you lose. BUT THAT DOESNT WORK IN TEAM GAMES. Sure, something like that works in a 1v1 such as Starcraft II, where it is your skill versus your opponents skill. But the heart of team games is teamwork. And when you are playing with other people who you do/don't know, there will be advantages and disadvantages. You can't put a machinacal aspect on a human response. Humanity varies, therefore the system will vary as well. Valves thinking was this: "ok, we have a bunch of players. Let's rank them all, then sort them out so each team has an equal average mmr. Surely that will fix things." But the problem is, you can't reduce a human player to 4 numbers. For instance, I look around in my 2k bracket, and I can point out many, many, innumerable mistakes my teammates are making. But however perfectly I play, I can't play for 4 other people. My matchmaking rating is a reflection of engages that I played with other people. It isn't a reflection of my own skill. If anything, it's a small reflection of others skill. The system sucks because it creates a cycle. Sometimes you'll hear a lucky success story where someone "breaks through" and climbs the ladder after thousands of games. But the vast majority of people are stuck where they are because of other people. Ex: I consistently play with people way higher up than me and preform well, and my party mmr kind of reflects that. However, my solo mmr reflects the quality of the people who I lay with who I don't know. So I would go so far as to say that my mmr, is not actually MY mmr. It's my solo queuing teammates.

Holy gaben, 3780k!!??

In all seriousness though, maybe valve should raise the level required to calibrate ranked mmr to maybe 20-something, or at least encourage people not to calibrate until they feel they are good enough.

lol 3.78k :P

It shouldn't completely reset every six months. Every so often, there should be a period where you play at your current MMR, and then it recals based on that. Done and done. Maybe every 3 months and then the recal time is like 1-2 weeks? That way if you are in scrub tier and you really are better, you will get a new more accurate MMR. If you are in high tier when you're not actually that good, you'll get put down to an accurate MMR.

i guess i used the term "hard reset" wrong here. From what i remember from my calibration matches, those games werent unbalanced at all thanks to the hidden rating system. Unfortunately - altho it definitely is better this way - valve keeps the algorithm for this hidden ranking top-secret, which makes it very hard to tell how it reflects on, the previously mentioned, player improvement.

But as you correctly said, there are lots of people who dont bother at all with ranked after their initial calibration, altho their hidden rating still keeps changing. With a seasoned MMR system those players would at least get some sort of "feedback" on their improvements. Maybe it would even be a way to make smurfing obsolete.

The thing I still don't get is this:

If you actually have 1000 games or so and create a smurf for a fresh ranking, you will be playing with people that honestly are playing ranked games for their first time. This should be enough for you to win the 10 games, since we all know there are some heroes that can carry 1v5 better than others, specially hard carrys (the enemy won't even know what ganking means, so you probably will have space to farm, take objectives, etc). Since you've won all those games, you will be put at a high mmr braket. You then proceed to play against people that are in that braket too. According to Smuggles, if you're bad, then you will start losing and come back to your "true" mmr; if you're good, you will win and maintain the high mmr you earned in the first calibration. BUT everybody has already said that 50% is the regular average wins/losses for anyone, in any braket (unless, of course, you're reaaaaally bad or reaaaally good). Now, this should mean you play 50 games after you recalibration against noobs and you WILL maintain that level you've earned with your smurf, unless, again, if you're really bad (but if you are, you shouldn't have won even those 10 games in the first place). So what Smuggels tells me about how you always find your "true mmr" can't be true.

This means you will find the exact same noobs up in high mmr - except that they are even more convinced they are always right (an Anti-Mage I played against refusing to do a Monkey King Bar against Phantom Assassin because "he had thousands of games with the hero so he knew that wasn't a good item for him" comes to mind - and that person was probably not lying, he had many games with the hero, but he didn't learn how to be flexible - so for instance he still activated his Manta Style, even after seeing it always killed him against Earthshaker, and didn't head to everyone's advices).

The noobs you encountered in your second calibration, however, since they are honestly noobs and lost most of the games against smurfs, will be stuck at a very low mmr. And that's just because they have indeed less experience in dota - they are bad, but they are bad because of the fact that they just started to play.

Please correct me if I'm wrong.

I have the impression that mmr indeed means ****, since it is experience + common sense take makes good players, and I find these across all mmrs (and judging by a lot of high mmr I've seen lately, these are equally absent from those among high mmr). In my experience, you're closer not to someone with your mmr, but with your experience (time spent playing), plus of course other factors, such as the ability to spend time out of the game thinking builds and strategies and watching competitive games. People with high mmr are, indeed, people that have played a lot - whether with or without a smurf, since in reality smurfs have equally spent a lot of time in the game in their real account.

So wouldn't it be more honest to just say that experience equals quality and this is what Dota tries to do when attributing MMR, but people try to rig the system and still end up in the same place they were but more convinced that they are good?

The example of Sando proves my point, I think: he spent a lot of time actually playing ranked in his first account, and in his own words did an effort to achieve where he is now, and once more that translates into: time playing + time thinking about the game = quality. Since he has those factors, he also has a high mmr - which probably is indeed accurate, not because it is a direct indication of quality, but because it results from his experience.

What do you think?

This being said, I still think smurfs are worse than they are benefitial, since you don't have to be that good to create a smurf and beat people that play ranked for the first time (just pick a broken hero), neither to maintain that mmr later, as I explained because of the 50% chance thing. Not only you don't have to be that good, you will probably be less than that, since to achieve your objective you had to play a 1v5 style, which will affect your gameplay later. Most of the people that I know that are very good technically and that are convinced they have a competitive mmr are actually bad team players and thus nowhere near competitive players. And this makes sense even psychologically speaking, since people that like to take short cuts to things they think they deserve are generally more unkind and more impatient - which means less teamplay and less ability to learn from others.

Sorry for the long post too. I don't seem to be able to ever explain myself shortly x)

thanks to ChiChi's latest post i have come to understand, that there are some huge misconceptions about player evaluations.

Firstly, there IS a HIDDEN-RANKING-SYSTEM which assesses individual player skill. This rating is updated after every single game you play (regardless of type), but is not visible to ANYBODY - not even yourself. The algorithm, on which this system ranks players, is kept top-secret and buried deep in valves databases, to avoid exploitation (see swixcaps post).

Since nobody appears to know about his, many are skeptical, if such a system even exists in the first place. But it simply HAS to.

Example:
Lets say you are a player who has played 500 games but havent played a single solo- or team RANKED game yet. In your profile, both these stats show the value "to be determined". And yet, if you play a normal-matchmaking game (be it team or solo), both teams are on about the same, average(!) skilllevel. Even if there are 1 or 2 OBVIOUS smurfs in this game, they are still getting queued with you. This is exactly because the system determines, from their own respective hidden MMRs, that your skilllevels are close to each other.

Another thing, that such a system, IN FACT, exists:

The first match, of your calibration series, that you play in ranked, queues you with people, based on those hidden MMRs. Otherwise these matches would be insanely unbalanced, which they arent. Also, there are people all over the internet, that claim (and actually have proof), that they lost ALL of their 10 calibration matches and still got placed in high-mmr brackets....there also are people, that say, that the exact opposite has happened to them.
Even personally, i have made similar experience. I have won 6 / lost 4 of my calibration matches and still got placed way higher than some of my friends with better stats. On the other hand i also got placed way lower than some, that did worse.

This means, that no matter what gamemode you play, you will always get queued with people of similar skilllevel. Even when playing with friends....altho in this case it means, that your (hidden)MMRs will be brought to an average value, that is used to find proper opponents.

From what I've heard, your invisible MMR is calculated exactly the same way as your visible MMR - it's all about wins, you just don't see the changes.

Also, AFAIK, calibration games also work partially from your invisible MMR - it gives you games higher and lower than that mark in a bigger range than you're normally used to. These games have substantially more impact than your standard games in terms of altering your ranked MMR.

ChiChi:

To answer your points, you're partially right, but not entirely.

In your Anti-Mage example, that guy is very experienced with the hero but has a closed mind. If he can win 6/10 games with his build and never changes it...he'll gradually go up MMR. If he was more flexible he might win 8/10 games and move up faster. Or he'll hit an MMR when being inflexible no longer works and will stay there at 5/10.

Experience is important - but only if you learn the right lessons from it. I have a friend who started playing DOTA at exactly the same time as me (literally within a week, we were playing BOT games together). He's got a lot more hours logged than me, and nearly double the number of games. His MMR? About 2.5k.

Why is this? Basically he has fundamental flaws in his playstyle, he's inflexible and weak on game theory. He's played a lot, but never really learned from his mistakes.

Practice is only truly beneficial if you're actually practicing doing something right. If you're practicing doing it wrong then all you're really doing is becoming a little less bad, and you'll always be limited by it. I regularly see people with 2x / 3x more games than me at my MMR, they're not improving, so they stay there.

Part of the problem is people learn by doing most of the time - if their pubstomp Dagon build gets them lots of wins and moves them up...they'll keep doing it because "it works"...and if stops working...it must be because their team sucks?

I know as a support player I've absorbed some "pub support" habits. Stuff that helps you win disorganised games, but may be actively bad in competitive ones. You get used to playing at your level and what works there...moving up means adjusting, or at least being aware of your habits.

Swixcap:

I kinda got to go back to his points here - I've been around for a while and often get asked for feedback on games, and "why isn't my MMR higher?".

In about 80-90% of cases, you look at one game where they played well enough to beat up the scrubs around them despite their own mistakes. Then you look at the 3 games in their DOTABUFF around it that they didn't send you - they played badly, made basic mistakes, fed, fell out with their team or whatever. All of those games count - the one where you owned is weighed just the same as the one where you got into an argument over who's mid and blew it.

I'd suggest to anyone concerned over their MMR to post up games here and ask for feedback - we can all improve and the best and most certain way of improving your rating is simply to be better and win more games. That means breaking out of your patterns, analysing your own play critically and working hard to make genuine improvements in all areas.

At the end of the day MMR is not an exact science, I'd say it gives a reasonable way of matchmaking players within about 500 or so points. Even then, people have good and bad days, picks that work or don't work, team chemistry, whether people are playing their best roles/heroes/etc.

I knew of this "hidden ranking system", of course, since if you don't play ranked you are still queued with people of your skill level, more or less. This makes sense, and it is the only way for the game to be fun at all.

But again you didn't answer my question: what if you don't have a hidden MMR, because you didn't play one single game with that account?! Your first game in the smurf should be a piece of cake. Then you get your hidden mmr higher, and are matched consequently, until the 10 games of calibration are finished. This would mean the ranked matches would get incresingly difficult, but actually you're playing with people that are also looking for their first mmr in ranked - and they can be people with very few experience that got lucky and thus won their ranked games but that are still bad, so you're supposedly playing against a 3k but in fact he's bad and you win easily, getting your own mmr higher).

That hidden ranking system functions as the "experience system" I wrote about: it makes you play with players that have roughfully the same experience, or what they consider to be the same level (for instance they have more experience but are not that good, so they lose more than the 50% average, and if you are level 50 you end up playing with a level 100 that is as good as you, or vice-versa).




Exactly! That proves my point. The inconsistencies you and your friends felt when calibrating prove that mmr is not accurate in the sense it does not equals skill level, so it's dumb to be playing "the game of mmr" improvement and the "search for the lost mmr", convinced that having a high mmr places someone in a high skill level per se (when it's the other way around).