Greedy Goblin

Thursday, January 12, 2017

Another League of Legends update

Riot announced a Warwick update. Deploying that update (it's currently on test server) will end my project anyway, as the new Warwick has skill moves that means constantly increasing skill with games skew any result.

I'm understanding the matchmaker even better. There are three kinds of teams, determined before the actual games starts:
  • Easy wins: games where the matchmaker clearly wants us to win.
  • Exploited defeats: games where the matchmaker clearly wants us to lose, but I still win more than half, thanks to understanding the weakness of the "Easy win" team on the other side.
  • Fair games: the matchmaker can't care less who wins.
Below you can see how these teams have very different matchups and how their internal win chance can be determined by the distribution of teammates. I mean if a team has lots of Type C members, exploit mode is my only chance and having some Type B predicts that I win, while Type A signs a likely loss. I try anyway as a few wins against the odds rises the winrate over 50% in these cases. I'm also considering dodging these games. Currently I only dodge if I don't get jungle role or someone picks/bans Warwick.

The most important is to see how even is the type distribution of the fair games and how uneven it is for unfair games. Anyway, the solution is clearly working:

This Monday or the next, depending on how fast I can climb I post the final results with the explanation of the Types which will clearly prove that Riot is acting maliciously.


maxim said...

Being able to win more than half games the matchmaker wants you to lose makes me doubt the motives you ascribe to the matchmaker.
Either the matchmaker is not malicious, or you actually now have enough skill at the game to beat it, thus skewing the results by your own admission.

Gevlon said...

@Maxim: If I was just "skilled" I should win much more "fair" games. Again, I don't blame anyone not believing yet, since the proof of the maliciousness lies in the types and the way of the exploit mode play.

Anonymous said...

'Anyway, the solution is clearly working:'

Erm. No. You were already in s2 sometime last season.

Anonymous said...

Player D is someone who starts talking in the champion within the first 10 seconds I bet.

Anonymous said...

Are these results statistically significant? What is the error on each of them?

At a glance, they dont look wildly unexpected.

Gevlon said...

I was in S2 by dodging 75% of the games. Now I don't dodge and win every type of games.

Nope, Type D is different. All types are objectively defined after looking at their history

It's not wildly unexpected that some games have practically only A and D players and these games won at extremely high rate?!
It's not wildly unexpected that some games have most C players and can only be won if you play in a completely stupid manner?

dobablo said...

How likely is it to get that sort of team selections from your player pool by random selection? At a glance your total population seems to be approximately 2/5 bad, 2/5 Ok, 1/5th good which falls in line with your exploit/fair/easy team outcomes. If that is the case you have not proved rigging but you have created a system that identifies if you have a strong/weak team and lets you tailor your tactics to accordingly.

If the team selection was rigged you would see that either:
- the A/B/C/D ratio of the pool of player you were matched with differs significantly from the general silver population.
- Your team make-up was highly unlikely to have occurred through random selection of from the player pool.

Gevlon said...

@Dobablo: type D is not good. Type C is not bad.
Even if they were, it doesn't explain how I win with lots of C teammates despite I'm not at all carrying those games. Finally, if I'd just classify random selections, my winrate should be good with good games, average with average games and bad with bad games. However my winrate is the worst with average games.

Soon all will be revealed!

Anonymous said...

Have you considered the possibility that you as a player are atm roughly ~S3-S1 level WW jungle player no matter what?

That is how I would at first glance interpret this. The number of games seems too low to make any other assumptions.

Gevlon said...

Yep, the extreme differences in teams are all because of my skill. Jesus!

Anonymous said...

What I meant that with that low number of games I'm sure you can come up with at least 5 very different models which support the data.

That is, unless you are willing to increase the number of games by a lot, the most likely explanation is just that you are around s3-s1 and rest is random variance.

Gevlon said...

Yep, 60% of the teammates are Type C in games that I've always lost until I created a specific playstyle against such games is random variance.

Are you a Riot dev? Jumpy a bit that your little cashgrab gets exposed?

Anonymous said...

You have created yourself a very strong 'this is most likely happen' mindset. It must for sure affect also the outcome of the games.

Players who have played long have seen long win and losing streaks, say ~20 wins in a row or similar number of losses in a row. In these games the players tend to play in a similar way. How does your theory view this phenomenon?

I am not working for Riot. I do wish that you would be able to remain objective in your forum posts (and comments here) and consider all possible explanations rather than 'this is how it must be since I don't want to consider other alternatives'. Of course you might get more readers with Trump-like simplifications.

It's a question of whether you want to be objective and really educate people or have readers with XXX busted! titles.

Gevlon said...

I was not looking for this. I was looking for a honest error:

I assumed that the player's MMR is meaningless as he plays different champions on different skill. Based on that I started to dodge games where players had bad winrate with their chosen champion. This - despite heavy dodging - provided only limited result. Analyzing the data afterwards shown that the best predictor of a win is high amount of Type D players. This was a total surprise to me. It cannot be explained by anything else than malicious acts on the side of Riot.

Anonymous said...

Did you compare your results to random chance? Your sample size is pretty small and you appear to be looking at your own team. Do you look at the rating on the opposite team at all? If your team had D players and enemy team has D players. Would it be evened out?

In the mean time I will probably try to find your replays to see if they have the exploit type play style.

Slawomir Chmielewski said...

"Never ascribe to malice what could be ascribed to stupidity"
or randomness.

Have you analyzed the population of players and likelihoods of getting these team compositions as a result of random matchmaking?

If you can win some games that you previously lost, it means YOU found an "exploit", not that Riot are bad people. Unless you have some other evidence, of course.

LazinessEvolved said...

I played League for 2 years, low gold. I think Gevlon is onto something here. I have my ideas about what Type D actually represents, too. I won't embarrass myself with a guess here, or steal his thunder. Whatever it is, it's absolutely connected to monetization. I long had the perception that the games felt "wrong" when certain types of players were on the enemy team. Excited to find out what Gevlon is hypothesizing!

Dàchéng said...

"Exploited defeats: games where the matchmaker clearly wants us to lose, but I still win more than half, thanks to understanding the weakness of the "Easy win" team on the other side."

Then what is your evidence that the matchmaker wanted you to lose these games? Is it simply because your theory predicted that the matchmaker wanted you to lose those games? If so, then consider that your theory might be mistaken.

Anonymous said...

"I was not looking for this. I was looking for a honest error:"

And yet just a few days ago you said:

"LoL is a free-to-play game, so rigging is just as obvious as "perpeetum mobile isn't working". Finding out HOW is interesting."

I certainly won't be surprised if you have uncovered convincing evidence that LoL is rigged to favour a certain type of player who is more likely to spend money on the game (though I'd be pretty surprised if you could prove they *actively* rigged matches to favour specific people who *had* spent money, rather than just the generic *type* of player)...but you pretty clearly *are* biased and *looking* for evidence of foul play/rigging from the beginning of your project here.

I honestly think your experiences in EVE have removed a lot of your objectivity as you analyze games.

I'm not saying your math is wrong, and I'll withhold judgement of whether your bias has impacted the quality of your analysis until after we see your final results (a lot of people seem to be preemptively assuming it has here before they even see the results in question - which is ridiculous at this point, and just an overreaction in the other direction) - but whether it has impacted the results or not I really don't think you can honestly say you entered this project entirely objectively and unbiased.

For your next project, if you were serious about it when you mentioned it as a possibility in a previous blog entry, I think many of us would love to see a gaming project where you actually gave up on playing solo and started a group-based project to dominate a multiplayer game thoroughly using rational strategies and logical play styles. I'm sure you would have no shortage of volunteers to join in and help, even though you might have some issues weeding out the Socials and the M&S players...

Barring that, I at least hope for your next project that you are able to find a game for which you can write the sort of money-making/power-gathering guides that made your blog famous - or at the very least that you are able to begin a project that allows you to return to your rational, objective analysis of the game in some way.

In any case, I do look forward to finally seeing your full results regarding LoL, as well as seeing where you go for here. One way or another I'm sure it will be interesting.

Anonymous said...

Hmmm.. A lot of Riot Dev's are Ex-Eve employees...

99smite said...

Hm... I can't care less whether matchmaking is rigged or not as I do not play LoL.
What might have been interesting if gevlon had actually tried to get some LoL in-game friends with whom he could have teamed up.
It would have been interesting to see whether a team with communication and a decent understanding of game mechanics would win much more often than casual teams.

After all, most MMOs are meant to be played by teams. IF these teams coordinate their actions, they might have significant advantages over casual teams with some afk leechers, troll-dumb feeders and other morons...