Greedy Goblin

Wednesday, October 26, 2016

League of Legends matchmaking is fundamentally wrong

40 games ago I started to queue dodge in League of Legends. You can see my last 40 games, latest first (press "show more" at the bottom)!
The latest 20 games show normal 50% winrate. The previous 20 games show 90%!!! winrate. No it's not me being great, my personal statistics aren't good at all:

I was oscillating around 1150 MMR for weeks. Now I have 1350 and became the limiting factor for my team. On Friday you'll see why my play was completely wrong and I deserved to be at 1150. But how did I reach the 90% winrate? By understanding how the matchmaker works and playing it for my advantage.

On the surface, League of Legends uses ELO-rating, a hidden MMR raising when you win, sinking when you lose. Over long term it reaches the point when you win 50% of the games, and then it reflects your "true" ranking. So far so good. Let's imagine Joe who played 1000 games, won 500, lost 500. This should mean that his ranking is known and in the next game he has 50% chance to win. The problem comes if we look at his detailed champion data. Joe loves Katarina mid. He plays this whenever he can and he is pretty good at it. Out of the 1000 games he played 600 Katarina mid and won 400, that's 66.7% winrate. When he got another lane or Katarina was banned or picked by the other team, he pulls some other champ and sucks with it. He played 400 games without Katarina and won only 100, that's 25% winrate. The thing is that "Joe's MMR" is totally meaningless, as you never play with "Joe". You either play with the awesome Joe-Katarina or the horrible Joe-not-Katarina. Joe might also be a good runner, but it doesn't make him good swimmer or good football player. Drafting Joe in football after his Marathon results would be ridiculous and Riot does just that: drafts him as Leona support just because of his Katarina mid performance.

So my plan was to check my teammates on or lolking before the match started and clos the champion selection client if I found them playing with champions they are bad with. It's crucial to understand that I did not exclude bad players, sure both teams have lower ranked players, MMR takes care of that. I was looking for a bad champion in the stable of the player. If I saw Joe locking in Katarina, I kept playing. If I saw him with something else, I quit.

Well, that plan didn't last long. To my honest surprise - most players play a lot with champions they don't know, despite having a good champion for their role. Let me show an example:
We see 161 ranked games over 58 champions. While I would gladly team with this player playing Kha'zix or Lux, but in the rest of the cases (80% of the cases!!!) he'll play with something he probably doesn't even know what the buttons do. This isn't a rare example, I see this all the time. I finally got to the point that "10+ games and 40%+ winrate with that champion" = "good". Over the nearly two hundred champion selections I've been in since I've started this evaluation, I have seen one game where all four teammates were "good" by the above terribly low bar. Having three "good" teammates is Godsent. When we had two "good" ones, I took the game. And I'm taking about 30% of them. What kind of games I reject, if I accept a team where half of the people had ten games with their chosen champion?
Mostly the enemy has worse good:bad ratio than us, so I chainwon, until I reached the point where I became the limiting factor. From there, I simply counted as +1 bad player and as I couldn't exclude myself from teams, my "exclude baddies" plan couldn't work. More about that on Friday. Please understand that this doesn't contradict the point of the post: if someone would have evaluated both teams, he could predict the outcome of the games perfectly before they started, which counts for a totally bad matchmaking. Tomorrow we'll see why the matchmakers is worse than just randomly assigning players to teams without MMR.

I can't play for a week for RL reasons, but after that I practice with new playstyle and up I go to Platinum combining playing well and being elitist in champion selection. "Elitist" meaning to demand that 3 teammates have 10+ games with their champ. Tomorrow you'll see why the matchmaker is responsible for this atrocious behavior of bringing zero-experience champions to rated games.

What will Riot do about my results? They will either increase the penalty for queue-dodging or make it impossible to learn past performance of teammates. Of course, neither is a fix, the outcome of the battle is still luck-determined, you are just forced to play 20 mins before the inevitable or can't foresee the outcome before 5 mins into the game. Just like you can't predict the winner of the coinflip championship, but you can't do anything about being the winner. That's how Ghostcrawler reacted when I chain-won Wintergrasp battles in WoW by the same method: he made it impossible to choose my teammates.

What could Riot do to make the battles skill-determined instead of coinflip? Removing player MMR and introducing individual player-champion MMR. So you'd have a separate Katarina mid MMR, Singed top MMR, Warwick Jungle MMR and so on for every champion/role combination you play. This way the matchmaker could create teams of equal MMR by drafting you into one of your roles. You would queue up listing all the champion/role combinations you want to play and the game will draft you into one of them. For example into a 2000 MMR game as Katarina mid. If you win, you increase your Katarina mid MMR, but your 800 LeBlanc mid rating would stay. Next game you might play LeBlanc against other 800 MMR champions. Players would also pre-select bans, so the matchmaker don't give you opponents with your banned champion. This way the pre-game would be instant, if the matchmaker finds the teams, the game can start instantly.

Will Riot do that? No way: "a champion’s win rate tends to drop during the free-to-play rotation, due to an influx of inexperienced players." So they are fully aware that the guy who plays Syndra first time in his life on a ranked game will suck like hell, but they allow it, because that's what makes the money for them: people buying champions and skins. With my suggestion playing a new champion would throw you back to the bottom of Bronze and you'd have to climb your MMR with that champ, putting a high tax on new champ buying. With my suggestion, non-pro players would have a handful of champions only, bankrupting Riot. So they'll keep up with the bad matchmaking that gives you easy wins half time and impossible losses the other half.


Halycon said...

I don't play LoL, but I play pleanty of other games with this type of system. Just not to the same extreme since character count in almost anything but a MOBA is much smaller.

So here's the thing. Once I hit a certain level of play with a character, I've learned over the years I need to also bring my skill level with other characters up to approaching that of the first to be able to progress any further with it. I need to know those other characters more intimately before I can have an instinctive grasp of how to play around them. That's probably not all people, but it is how I've learned I have to play to get better.

This has also went from picking a character I am very good with and running up the rankings until I hit a glass ceiling and have to drop down to characters I'm very bad with, to a more well rounded playing lots of games with everyone so I progress naturally. Win some lose some, instead of just win win win until I can't win any more, and then needing to lose a lot to be able to get back to winning. I've found it less stressful to do the 60/40 50/50 win ratio in trying to get better than the feast or the famine of either constant wins or loses.

So, there is a reason to play lots of characters badly. At least in my book. A perfectly logical reason for why people do that.

Anonymous said...

The system you are suggesting might work in lower elo (whatever that is). The higher you go, the more meaningful it becomes if you can pick a counter pick to your lane. Ie. lot of players might have a win rate of ~55% with 5 mid (as an example) champions. However their win rate depends a lot on whether they can get a favorable match up. Thus your system would need to be modified so that each champion has a specific win rate % against each champion. And only when those champion pairs are available in the queue, would you be able to start a game. In reality it could mean that queue times would become horrendous as no proper match-ups were found.

I think currently LoL relays on statistics. While in a single game the win rate might depend on the exact champion the players choose, in long term it currently averages itself. Also because win rate depends on the match-up the higher you go -> your method should start losing it's effectiveness the higher you go as you cannot see champion vs champion win rate of players. Whether that point is gold, plat or something else I don't know. Perhaps you shall post the results.

Samus said...

This is really an issue of your current level. Players at your level still haven't tried all the champions yet, and you are just hoping they will be competent at the champion they pick. Players in Gold or higher have played all the champions, and will be at least competent with the champion they pick most of the time. 158 games may be typical in Silver, but I think you would have a very difficult time finding a Gold 4 or higher player who had played fewer than 200 games, most are going to be much more than that.

At that point, they need the freedom to counterpick. Joe may be great at Katarina, but if his opponent Jeff picks his 200 game, 70% winrate Malzahar, Joe's Katarina is going to get wrecked.

There was some discussion about letting players queue for just "new" champions, and the game wouldn't let you pick any champion you had more than 10 games with. This would have a different, obviously much lower MMR.

Skeddar said...

Glad you remembered my comment in one of your 'I'll try LoL again' blog posts and started stats-checking and queue dodging (or maybe you didn't remember it and came up with it yourself?).

As you noticed it's a great way to boost your win rate temporarily, until you reach an elo at which you have to work on your own skill.

nightgerbil said...

Best way to learn to beat an enemy class is to go play that class. Its for this reason I wasn't "just" a hunter in wow, I played the other classes in arenas and bgs so I could attack other hunters and learn how they countered me. Same in swtor where I have all the classes at max level and play them all. In league you would see me with 5-10 games of most of the popular champions, just so i was aware of what there abilties were and prefered item build orders. Once you have played it 5-10 times your going to understand guides that you find and read, then know roughly how to counter them on your mains.

So I wouldn't say that player you gave as an example was bad. I see him as being more well rounded then someone with just 40 games played with one champion. I'd also like to see you play more champions yourself, espec if your trying to reach plat/diamond.

Anonymous said...

there is a way easier choice: just as you cannot use the free-this-week champs in ranked, you should need X-mastery points on a champ before you unlock him for ranked.
let's say ... mastery lvl 4 and above = clear for ranked.

Anonymous said...

Not good enough!

We need MMR per Position-Matchup, including Junglers and Lanemate and possible swaps which resets as soon as any of them gets buffed or nerfed.

The real reason that will never happen is that putting a Diamond Syndra in Bronze for playing LB might have, lets say, a questionable outcome.

Also you just removed draft mode from the game, screwed over queue times for high ranked games beyond repair and removed any semblence of real competition from ranked.

You got my vote, i want to see reddit explode.

Ekareya said...

just my 2 cents, but a heroes you never played MMR shouldn't start at rock bottom. that way, you can create a better of both world matchmaking.

Anonymous said...

A more socially acceptable method would use your overall MMR to decide in which player pool you are grouped (as now), but balance teams around champion MMR.
For example, if you are Silver 3 you will always play against approximately Silver 3 players. If three players on the other team have never played their champion it would be balanced by similarly poor ranked champion MMOs on your side.
It would help encourage players to try new champions. Instead of new champions being a handicap that results in a loss, they become a handicap that will be mirrored on the other team.

Gevlon said...

@Anon: counter-picking only work when you actually know how to play that champion. I'm pretty sure that Average Joe picking the strongest hard counter will lose against Average Jane if he had 3 games with that champion and has to read the tooltip above the buttons.

@Samus: there are 120 champions. We can agree that 10 games are needed to be at least somewhat competent (not good just above "I wonder what happens if I press Q"). That's 1200 games. A game with queue and selection takes an hour. With 20 hours a week, it would be over a year to get there. I doubt if most Gold players are there.

@Skeddar: everyone and his mother offered queue-dodging, that's a pretty straightforward thing. The information in this post isn't about that obvious thing. It's about the surprising fact that most players pick champions they have no clue about and instead of winrates, I screened for match counts.

@Nightgerbil: trying out things is fine. But that's what normal games are for: faster queue, faster matchmaking and no consequences. This is what I assumed people do. The surprising result is that they pull their newly bought champion and see what happens on a ranked game.

@Anonymous: demanding masteries would help a bit, but wouldn't solve the core problem: people inherently play on different MMRs with different champions. No matter how many games I'd play, I wouldn't be as good Assassin as Jungler.

@Last anon: who said that the same queues should be for high ranks as low ranks.
I don't understand why there is a solo queue for high ranks at the first place. I'd end the ladder with Gold for "plebs" and from there only 5 man teams could compete as teamplay and synergies start to matter.

Anonymous said...

@@Nightgerbil Normals and ranked have seperate mmr, no insight can be gained from playing against ppl 3 tiers below you.

@@Last anon The same? We are talking here 10-20min vs 1h+ aka unplayable. Are you implying that the highest MMR achievable would be the end of current gold? Bring Magnus Carlson to your local tournament because he and the players there are considerd "equal strength"? Or do you mean that soloq gets disabled as soon as your mmr get's to high, which gives incentive to level another smurf because my teammates are not online?

And as far as i am concerned leaving champ select should be 1h play restriction and -25 points.

Gevlon said...

@Dobablo: how do you plan to achieve that? Players are selected for their MMR to the champion selection screen. Then they start picking champions. 3 on my team picks something they have zero experience with. Then you force 3 opponents to do the same?

Or would you have a separate "Mastery 5+" and "new champion" queues?

Anonymous said...

I still don't get it, why you all assume, that someone with low ranked games played is inherently inexperienced, bad player.
There are different modes besides ranked, where you can hone your skills with specific champ. Not to mention, having a smurf account is unrestricted and quite common.

I, for example, have very few games played this season, still I was able to end in Platinum because of performance from previous season. One should keep in mind, that stats can be misleading.

Samus said...


You forget, the game is 7 years old now. Basically every player Gold or higher will have played for several years. I would guess nearly all of them have more than 2000+ total games by now. They also only need a game or two to try a new champion and discover they don't like it, which they did long ago. Even most professional players have champion pools of under 10 champions that they play nearly all of their games on.

I am not saying you won't still benefit from dodging, even when you are Gold. But I think by then you will find nearly every teammate will be on a champion that fit your "good" definition, and you will need to change it to be more useful.

Andi Y said...

the problem you face in silver elo, is that many also use it for their second and third account.
so i also try out new champs on my alt-accounts.

on some champs i have 70% winrates. and on others, it is worse.

also in silver are many unexperienced players and trolls. call it what you want. they just play for the LuL.

if they would MAIN a champ and you are not total brainless, you will raise to gold. automatically. the issue is just, the people keep switching champs. they saw youtube someone owning with lee sin, they pick it. they are living in a dream.

still it is interesting strategy to dodge.

99smite said...

A very good insight. I do not currently play LoL but similar habits can be seen in other games where players can choose different heroes or classes. One always encounters players with the same "player exp", but with a character he know not how to handle. This sometimes sucks, but is unavoidable.
In "Paladins" for example, ranked matches only will unlock after players have reached lv4 with at least 10 out of 18 heroes...

Riot could change a lot in their matchmaking mechanics, but they chose a rather simple mechanic, which "works" in most cases. Riot could also protect their sales if they unlock certain ranks only with a certain unlocked hero pool so that people would only advance if they have unlocked a certain pool of champions...

maxim said...

The practice of "smurfing" is pretty common on highly competitive ladders. Smurfing is when you abandon your old account and start a new one, complete with fresh MMR.

Normally, smurfing is used by highly skilled players to get accounts with unrealistic win ratios. However, the reason i'm bringing it up is that restarting from 0 MMR is not actually all that high of a tax IMO.

Hanura H'arasch said...

It's really strange how elaborate the API of LoL is. I mean just look at it. You can retrieve practically everything from any in real time without needing that players API key or anything.

One should be able to build an application that tells you in real time how likely you are to win before the game even starts, using statistical analysis.

Gevlon said...

@Hanura: the problem is that you don't know the names of the opposing team before the game starts. In the champion selection you just see "Summoner 1", "Summoner 2"... as opponents. So while you can query the data of anyone, you don't know who to query.

Anonymous said...

But you can still see your teammates, so that's 50% of the formula.

LoL matchmaking isn't perfect but it works. Some people manage to climb the ladder, some just don't. Or do you think, that people who manage to get into the TOP, never ever get bad teammates ? Can't blame the system for everything.

Introducing even more artificial barriers to ranked games is just a big NONO. You either end up with long queue times or large MMR differences between the players - both game killers.

Dangphat said...

If they are not in the same match making system as the "pro" players, the "plebs" would not join the ladder as that is one of the "fun" draws to think that they are in the same league but lower in it.

Also if you agree it brings in players to the game and makes them spend more, is your argument purely theoretical? as what company would target your goals over profit for their share holders?

Gevlon said...

@Dangphat: it would be a stepping stone. After you reached Gold 1, you can queue for the "pro" league with current rules.

Anonymous said...

The idea of "has played only 5 times with a hero so sucks" is actually a very M&S silver ELO based viewpoint.

From experience, I can pick up a hero in a MOBA I am used to, that I have never played, and generally understand how it is used. While I won't be the same level of skill as a perfect player, I can outperform many players who main the hero in high Silver and low Gold simply because of knowledge about how the game works, looking at the scaling of the skills and their effects, and judging the enemy.

A large part of that though comes from knowing the other champions and how they work. As someone earlier noted there is a lot to be said for countering, and using some of the recent Dota 2 international games as an example, in some of them the game is over by the end of the picking stage because of counters.

That's not to say at Silver ELO a team with the wrong champions can't win, but that it is an uphill slog, and one player who is of Diamond skill playing a new champion will probably outplay the general people he is playing with/against, despite supposedly only having one or two games on the champion.

All of this is supposing the M&S you love you talk about though Gevlon are not involved. And I think that its not a black and white, but a definitive shade of gray where there are lots of partially M&S who play in low ELO's, not exactly terrible but also not bothering to improve.