Greedy Goblin

Monday, January 16, 2017

League of Legends prison

Over the last weeks I had 57.5% winrate in League of Legends. Now check out my last 40 games:

25 losses, 15 wins. What's the chance of this happening by bad luck?
And it's not just that I've "lost". I saw more AFK-ers, DC-ers in the last weekend than I saw in the last month. I got a botter whose bot bugged and just followed me around in the jungle. I got a totally new player, I wonder what he does in Mid-Silver (and yes, he played exactly like my mum would if I'd give him an account). I had extreme collection of toxic guys. And - on the opposing team - I saw absolutely oppressive good play. I've learned more about jungle tricks from a few Kha, Shacko and Lee in the last weekend than I learned in months.

It was clear that the matchmaker wanted me to lose and not just the usual way. While I identified that about 40% of the games are meant to be lost, it's not because of me, but because of giving a free win for the buyers on the other side. That's why the exploit works: I know that they are boosted baddies and figured out how to break that. By the way I barely saw such games, I found 11 out of 40 and won only 2, because they weren't "normal exploits" (I declare the game one way or another based on champion history of teammates which is far from perfect, but good enough). Most games were "fair" and I lost them by a large margin.

Why? Because last Friday I analyzed the data and figured out a "shortcut to gold". I classify the games into "easy win", "fair", and "exploit", but I also realized how the outcome of the games is predictable within classes. For example a fair game with more teammates with high winrate is more likely won than one with baddies (go figure!). Exploit games also have a - pretty surprising - prediction method. This is much less predictive than the main classification (free wins have 80%+ winrate, fair games have 55% and exploited games played without exploit have 25%). But still, by the "goods - bads" prediction I could split the fair games into halves, one with 65% winrate and one with 45%. So why should I waste time? Let's dodge the lower half!

That's when and why everything went crazy! The game really hates queue dodgers since they both cheat (it gives advantage even if the matchmaker is fair) and because it wastes 5-10 minutes of 9 players. Punishing queue dodgers is a fair idea. So is punishing toxic players, botters and quitters. But instead of banning their accounts (they just make one new free) or giving out some in-game punishment (hey, they can just drop their account and make one new free), they put them to a "prison zone" and give them losses, by giving them each other as teammates and way better enemies.

It's actually genius. The dodgers don't know that they've lost because of being punished, they think that whatever dodge formula they made up is not working. The toxic people don't realize that they've lost because they are punished, they are given the famous loading screen quote "players who criticize teammates after a mistake lose 25% more games"! Botters believe that their bots are badly written and quitters - well, they quit. The really new players who play ranked on an account they haven't used for years (or just bought on ebay) will lose anyway, so they can lose in the prison, serving punishment to the inmates instead of annoying honest teammates.

While I admit that this is a genius plan, it's also a perfect example of the matchmaker being totally rigged. I haven't dodged on Sunday and I wonder how many lost games or real days I have to serve in the prison before I can return to evaluating the "normal" rigging again, the one which is aimed to help the paying players.

Update: I played one more late night and lost to another AFK-er. But after midnight we won! So from the sample of one I dare to say that my day of imprisonment is over. One can hope.


Eaten by a Grue said...

Heroes of the Storm has something this, but they declare it openly. If you disconnect from a game, it makes you play a game with a team full of people who also recently disconnected. The queue takes longer, because there is a smaller pool of such players, so that is the main pain of it. The game itself does not seem unfair, because the opponents are also disconnecters.

With regard to your findings, it's an interesting hypothesis, but you are in a science field, so you should be familiar with statistics. Bad runs are normal. I had a run of 20 losses in Heroes of the Storm, and thought nothing of it. You seem to be placing way too much weight in results from very small sample sizes.

Zyrus said...

Ah, the Prisoner's Island.

The funny thing is that if Riot came out and said "Yes, we do have a prisoner's island for dodgers, leavers, bots, toxic people, etc" I don't think they'd particularly be looked down upon.

Look at Dota 2, they have prisoner's island.

Interestingly enough, Riot denies they have such.

Gevlon said...

@Grue: have you checked the chance of this run? Also why did I see objectively harmful play (AFK, DC, toxic, bot) more. If I had bad runs, I'd just lose.

@Zyrus: that's the problem: Riot claims that its matchmaker is fair, when it's obviously not.

HotSplayer said...

Don't they have a "Low Priority" Queue?

I had assumed that this would work the same as HotS and you would get Q'd with other people in the low priority queue, hence the drastic jump in AFK, DC, toxic, bot players.

Anonymous said...

I experience that "prison" during weekends for the last 4 years. I call it "weekend matchmaking nonsense", due to it usually occurring from friday evening to monday morning. Rather than time period, it is usually a set of games, works on both normal and ranked, I find myself getting about 13 roflstomps in a row, after which the game resumes normally.

Usually I get my account chat restricted or banned during this period, because, despite me fully knowing what's happening, the team refuses to ff at 20, or at any other period - matchmaking is incredibly good at finding the most stubborn feeder idiots. At minute 30 they start teaching you that you live your life in a wrong manner, from the height of their 1/13/4 score. And then they mass-report you, and you get a ban.

Anonymous said...


They call it "Low Priority" Queue, but that's not anything like that, it's just you're not allowed to queue in 5-20 minutes after you pressed play every time you press play, which they decided to call "Low Priority" for some rito-level retarded reason I cannot perceive.

Gevlon said...

@Anon: I've never had chat removal, probably because I don't chat much. I suggest to mute all at start. I've also haven't seen rating drop during weekends before. So I'm fully convinced that this is because of my queue dodges.

Hanura H'arasch said...

"Heroes of the Storm has something this, but they declare it openly. If you disconnect from a game, it makes you play a game with a team full of people who also recently disconnected."

It should also be noted that these "leaver games", as they are called, must be played in unranked mode. So they will NOT influence your ranking in any way.

Punishing people by (effectively) reducing their MMR for reasons other than actual poor in-game performance is a terrible idea, if you want a fair matchmaking system. It just leads to unbalanced game after unbalanced game. And that's weirdly enough exactly as I remember LoL ...

Anonymous said...

@Gevlon: chat removal is not given because you chat or not, it's usually the first ban warning regardless of actual violations. It's given even if you simply afk too much, so muting all at start doesn't seem to help.

It may be due to qdodges, ofc.

maxim said...

Sorry, i do not see any real data here. Just you jumping to conclusions about both how stuff works and motivations behind it.
Weren't you supposed to show the formula behind your player categories? Also, if you are right, then there is no way for you to prove the exploit by riding a counterexploit tactic to high gold.

Anonymous said...

If a algorithm works, but does some cliche calculations, it might not be intentional rigging. I try to give alternative explanation. Lets guess you get 50% winrate no matter what. If you play out every game, you will stay roughly on same spot. If you dodge, you lose some LP points every time you dodge. On the long run, you will lose ranks. That means you will get lower lever opponents and teammates.

Lets add complexity, imagine winning-losing streaks improve elo more. What will happen if you pick only teams who have good chance to victory. Dodging wont influence elo, and you will win often many games in row and hardly get any losing streaks. Your elo will rise alot, but your rank will stay roughly same, thanks to lot of dodging. So you will be matched against high elo and low rank players. There are 2 ways to get those, either you dodge alot, or you play against traded account players who are on losing streak. Botters, afkers are good example of the account trading side. It might not be intentional for that algorithm to make "prison island", but its just a quirk on extrem case scenario.

Btw, is LP gain changed recently? It was 22-24 per win and 14-16 per defeat. Now it seems both are around 18-20 per victory/defeat. If i would rig something, i would use LP Gain, its way simpler and effective.

Gevlon said...

@Maxim: 60% winrate goes to 30% in a second when it should go up to 70% by dodging bad games is a clear proof. This event is not connected to the main rigging, it's a side issue to punish dodgers and other problematic players, hence no player groups. I wish I didn't dodge because I wasted lot of time. I can clearly climb back (I believe I'm out of prison, I'm winning again more than losing and games stopped being retarded). I'll simply reach gold without dodging.

@Anon: my teammates got much worse, I mean I routinely get bronze players now.

maxim said...

Ah, i see, the point was that you were losing winrate as you dodged and regained it as you stopped dodging.
Fair enough, i can see a system that would do that. In fact, i'd probably do that myself.