Yunavers

Gold is average? I’m in my gold promos currently, waiting to get home to finish them, and I feel kinda bad now. I have over 700 games in Silver and I feel like shit now :/

Edit : I came home, lost my final promo game, then won 4 GAMES IN A ROW AND FINALLY AFTER OVER 700 GAMES, 4 LOST PROMOTIONAL SERIES, I AM FINALLY IN GOLD. It’s been a long 5 months... but I finally made it

UN : Cold Peach Tea / NA

I'm not sure they're right and I'm also not sure they're wrong. I'd be wary of where Reddit collects their data.

cavecricket49

You're operating off very dated information. Gold has become the top of the bell curve.

What's your source on this?

loyal_achades

You can't insert new players into the ranked system at an MMR different from the average without inflating or deflating the overall MMR of the system, because MMR gains and losses across teams are 0-sum. The only way to insert new players into ranked with a lower MMR would be to artificially add inflationary elements elsewhere in the ranked system, which is significantly more complex than just letting players enter in with the average MMR and a high K-value so they quickly correct to where they should be.

Hey there, I'm the game designer working on this space right now, and this explanation is pretty accurate! There's a couple other issues with entering people in away from the middle:

- If your median starts at value A, and you start to add new players in at value B, where B

- Not every new player is below average. Some are well above average, some are well below, but we're still operating with limited information in all those cases. Starting people too low means that above average new players create disruptions in the system for a longer period of time (because we're matching them and updating MMRs based on their current position, which is very different from the "true" position where they will settle).

This isn't to say we're happy with the new player experience to ranked right now - the way we do things now works out over time, but doesn't feel good for everyone. Our goal is to hopefully keep the individual matches feeling fair for both the new player and the experienced teammates they eventually get. We're working on some plans for next year that should help make the early ranked experience better, but it's unfortunately not as simple as placing people lower to start.

Surrendernuts

Why not just have new accounts play 10 draft games before they can play ranked. From those 10 draft games you get a rough estimation about their skill level, and from that skill level you play your first ranked game?

Example First draft game is gold 5, after 10 draft games you become silver 3, then after 10 ranked games you are placed bronze 1

It's not a bad idea, but draft games aren't played with the same competitive intent as ranked games, so we can't guarantee that how you do in those 10 games is representative of how you'd do in Ranked. I wouldn't trust Normal data as the only way of seeding someone, but it could be part of the solution along with other info.

elrd333

Can you explain how profesionnal player are given an account for international event. Don't say start at D1 mmr and inflate mmr?

Event accounts are seeded at a higher than usual MMR, but they do a lot less to disrupt the MMR balance because:

• They're only active for a smaller period of time
• They're seeded much closer to a reasonable approximation of their skill, since we know the pro player's performance already
• The number of pro players coming in is fairly small.

These accounts probably cause a slight overall deflation (as we don't seed them as high as some of these players are on their home server, so they tend to gain MMR), but it's considerably less than if we let those players enter the system via smurf accounts without higher seeding.

Greatius

I got question for you then. Why do I end placements in silver or even bronze, with 9/1 placements, where last few games were wins in high gold/low plat elo. Last few wins in placements give +40ish lp. After placements 1st win skips you 2 divs. Next 3 wins give up to 38 lp, and after you pass 100 lp you just go to next div without promos.

So why are next 10-20 games after placements more influential than 10 placement games themselves. 1/9 or 9/1 no difference at all, it is all about next 20ish games after placements.

While I'd disagree that provisional games are less impactful than post-placement games, you do make a good point that since you can't skip divisions during provisionals, you may get bigger single game jumps, depending on where and when those wins happen.

Some atypical things do happen when you go 9/1 or 10/0 in provisionals, though - I think this is more a product of you having an excellent win streak and the system cautiously catching you up than anything else.

LooseCandidate

I've had an even better experience, i just got level 30 after playing die-hard arams and normal games for 8 days straight, I jump into ranked play two games and i get flamed for being boosted and having "bought" my account, the guy swore he would get me banned, and guess what? 1 Game later, my account is suspended for suspicious activity, L M F A O.

I just made a support ticket to see why they locked my account and what was so suspect about it, except for the fact i was clapping up some kids as TF ;).

Sorry to hear it - support ticket is going to be your best bet. Hope it gets resolved for you!

Etheri

Is there any possibility of moving away from centered distribution (I.e. normal distribution) and towards a skewed, bottom heavy distribution?

I imagine guaranteeing stability, no rank inflation, ... is much more difficult. But it may be more representative and allow you to adress some of these issues.

We try to model our skill estimation to the results we get, not towards a particular shape of a distribution we have in mind before starting. So aiming for a bottom-heavy distribution wouldn't necessarily be representative of the data we see in game.

neberhax

Wouldn't it be possible to split up mmr for new players? Just hear me out.

The initial mmr's would be on one side the average mmr a new account has after like 20 games, and on the other side the actual average mmr of the ranked playerbase. Would it not be possible to have matchmaking (and visible rank) based on the 1st mmr, but have the 2nd there to keep the ranked system in balance?

Eventually both mmr's should meet in the middle since the 2nd one barely has any gains and significant losses.

Does this make any sense?

Yup, get where you're coming from on this, and it's a cool idea! But it's operating on the assumption that new players fall between those two points. We don't necessarily know that to be true (and in the specific case of saying they should be between average and somewhere below average, we actually KNOW that isn't true - many new accounts are above average skill).

There's probably a version of this where you could pick a point below average and a point above average, and execute the same function - but this would still constrain players into that range for their first few games, and do a poor job of handling outliers who are either really good or really bad. And for players who are really good, we'd be dropping them into below-average skill games for much longer, which makes those matches feel unfair to the players who really fit into those games.

candoodle

Why dont you factor in normal game mmr for the few first ranked game

if someone is getting shit on by bronze players in normal why toss them into mid gold mmr in their first ranked game

Normal games may be part of the answer, but results in Normals don't necessarily correlate 100% to results in Ranked. People don't always have the same competitive intent in Normal mode, so a player of average skill who tries hard may be slightly above average in normals, while a player of above average skill who messes around may end up below average. Then when those two players go into Ranked and start both trying hard, their Normal MMR wouldn't actually reflect their relative skill.

It is a good starting place, though!

hellyeah222

Hi, Can you please provide an explanation why there is such a significant divide between hidden MMR and the LP system? And I am not talking about new players here, so pardon for going off the topic of the OP.

If a player is consistently matched with people above his current division/LP, shouldn't he be promoted to the same rank as the people he's playing against? Why employ arbitrary limitations to LP gain maxing out at 35 or so LP per win, and limiting division jumps to 2 divisions (from what I've experienced myself. Not to mention that these jumps only happen if you REALLY shouldn't be in your current LP rank).

This is an issue I've faced in all of my accounts, where the MMR gains outpace the LP gains and I end up playing versus players with higher LP ranks than me. When I play Support on my main, I get to Gold 2 or so, and get matched with Plat players, but I struggle to reach plat myself due to the limited amount of games I have time to play. I know that with an adequate amount of games at a theoretical 51% winrate, I would climb, but I have a really limited amount of time I can spend on the game. Meanwhile, I am constantly playing with and against players in higher ranks without receiving the same rewards. I do not claim that I am Diamond level, but I would make a case that I am a Plat level player, but due to the limited time I have to play, I can almost never reach it.

Really, what is the downside for Riot to show the LP rank more closely to the hidden MMR rank? Here I am assuming that if I get matched vs Plat players, then it means that my MMR rank is plat.

TL;DR: Why not make the LP gains keep up with the MMR gains? Currently they lag behind and the division system screws over people with a limited time to play.

So MMR is meant to measure your skill, and that is all it's meant to care about. This means that we want it to be able to adapt quickly, be fine with overestimating, and move around quickly as we get new info.

LP is meant to be more steady and certain, so it moves slower and is less responsive to small trends. This is why it takes more games for your LP to move upwards and sometimes catch up with your MMR - it's effectively waiting for a larger sample size.

Also, if you are making new accounts (which I'm inferring from your post), it's going to take longer for your LP to catch up, because again, we're starting over from new info each time.

Futomus

One thing I've been wondering though. Wouldn't be feasible to look at the input stream from a player and apply an algorithm to judge the mmr? I don't know if this is part of smurf detection or mmr already, but I'm thinking that a deep learning network would be farily good at guessing the real mmr of a player given the proper input and also quite resistant to trying to gaming it?

So generally I agree, but tuning those kinds of models is not super easy, and can run into issues where once you've tuned it, you may promote strategies that focus on convincing the system you're good rather than actually winning the game. We are experimenting with some options along these lines, though, and they show promise.

Gkkiux

How does the system determine the final rank though? After losing first game in my placements and finishing around 7-3 I was placed in Silver 1, but after another win the system bumped me up to gold 4. Wouldn't it make more more sense to do adjustments like these in placement games? Or is it that MMR and rank need to be far enough apart for this kind of jump to occur at any point?

So provisional games aren't meant to mark the best rank you can achieve, just where you're starting your season. From your particular case, it sounds like you ended right on the threshold of Silver I and Gold IV, so as you won more games, it makes sense that you'd move upwards.

lilivnv

Are winrates not taken into account? I’ve seen so many people with 70-80%+ winrates in mid/low gold with like 40-50 games versus someone with <48% winrate with hundreds of games

Winrates are a very fuzzy metric, because they don't take into account who you won against. A lot of new accounts that eventually reach Diamond+ will start with 70 or 80% winrates, because they're a Diamond skilled player facing off against Gold skilled opposition. But as they move through the ranks and face harder opponents, their winrate will go down. That doesn't mean they've gotten worse - it means we've gotten enough information to give them more fair games.

dlabguy

A while back, you guys changed the API so websites like op.gg can no longer display MMR. Is it possible to get that feature back ? Perhaps in the main client ?

We don't plan on bringing visible MMR back - when MMR is working correctly, it does a lot of things that feel bad when you look at it in isolation. The biggest one is that for half the population, MMR only goes down, because we start people in the middle and then naturally, half the population has to be the bottom half.

As a side note, the API was not sending MMRs out to third party sites, that was a separate estimation done by those places. Part of the reason we asked for it to be removed was that it was often substantially wrong and led to a lot of false expectations.

KaisaPermanente

Why not have placements start lower and have win streaks worth more and loss streaks adjust match mmr more? I think the issue might be more the fact that the first two or three matches of a fresh account starts at gold, which is basically a coin flip of it being a new player, a bronze "smurf", or a high elo smurf.

I also have a question, why can i climb so much faster and easier on an account that has less games than another at the same elo? Example, having 100 games played at gold 3 vs 300 at gold 3. My 100 games account will climb much faster than my 300 games. This means i abandon an account every time an older one catches up, and so far its been too consistent to not be design. Is there no limit to how the system judges your improvement with games played?(i.e., judging it against a 300 game sample size instead of capping it at say the last 100 games sample size)

We match based on MMR, not Rank, so starting placements lower rank-wise wouldn't matter. If we started at a lower MMR, as I mentioned in the original post, it just ends up moving the middle downward and destabilizing more games along the way.

As for account age vs. climbing, I may be misunderstanding your example, but let's say you have a 300 game account that is settled in at Gold III. Then, you make a new account and play some games, and rise quickly to Gold III - much faster than the 300 games you played on the last account.

What you're not taking into account is that on the older account, you weren't necessarily at Gold III skill the entire time. Over the course of those 300 games, your skill changed and improved up to Gold III, and your MMR and Rank changed to follow. Now when you make a new account, your skill is starting at Gold III, and your MMR will adjust to that very quickly. Since you weren't the same player when you started the second account, you shouldn't necessarily have the same journey on the second account.

JimmyDuce

So it's your fault I get 30% WR people in my promos :P.

I know it's hard, and you get way too many suggestions, but why not weight initial MMR from norms? And also stop giving me 30% win rate people in my promos

Normals can be a part of the puzzle, but they don't perfectly correlate to skill in Ranked. It's probably a good thing for us to consider, but it's not the solution all by itself.

25chestnuts

Have you experimented with systems similar to what Hearthstone does? (where everyone starts at the bottom of the ladder, and rank up as times goes on?) Even if the queues are perturbed longer in this scenario, I am pretty sure that it feels less shitty to go from Iron IV to Bronze IV than from Gold IV to Bronze IV and more players will be likely to approach the ranked queue. Having more high skill players joining the queue (your words) may only be showing that effect.

Furthermore, given that MMR is already hidden, the overall deflation in the system should be transparent to the matchmaking. You only care to keep the relative distribution of rankings accurately.

I love Hearthstone personally, so I'm pretty familiar with their ranked system! It works very well for a 1v1 game, because you can resolve differences between skill and rank really easily when you only have two players facing off.

Let's say a given player is Rank 2 skill, but currently Rank 6. In Hearthstone, they'll play other Rank 6 players, and on average, be better - so they move up! Their opponents, as average Rank 6 players, will probably lose when they hit the Above average player, but that won't be the norm and they will even out over time.

In League, we have to worry about teammates. This means that if you are Plat Skill but currently in Bronze, and we match you into games of Bronze Skill, you probably win more. But so do 4 of your teammates, and 5 of your opponents lose a game they had no chance at winning. The amount of disruption each of these games creates is 9x what Hearthstone experiences, and means that the system destabilizes much faster as players get lucky streaks of being carried or unlucky streaks of getting stomped. This can happen today with smurfs, but adapting Hearthstone's system would make that considerably more common.

I'd also argue that, because of the lower inherent randomness in League, skill mismatches hurt way more. I can play a pro HS player and draw my way into a win sometimes, but I'm going to get wrecked by a pro league player 99 to 100 times out of 100. Expand that out to a full team, and that's a lot of people having unfun games.

Suavarino

I am in my 1st year of Ranked, And did placements at the beginning of the year and went 3-7 and was placed Iron 3 78LP

My friend https://br.op.gg/summoner/userName=kamatasyhi started a few months later, went 0-10 and was placed Bronze 1

Can you explain how this can happen?

Some of our placement logic has changed over the year, and we do use information from previous seasons to inform some of our placements, and the opponents you played in those games may change how your skill rating (and thus eventually your Rank) changed over that time. There's basically a lot of factors here that could make your experiences different.

getoastet

Stupid sounding suggestion, can't you take the (average mmr of all ranked players and add his normal mmr)/2? This would be a more personal placement which would make the first placements not so harsh for under average players and good players would have to climb less

Not stupid at all! Normal MMR doesn't necessarily run on the same range as Ranked MMR, though, so the math gets a bit wonkier. But the general idea of using Normal MMR to inform (but not dictate) starting ranked MMR is good!

sowydso

Can't normal game mmr help with gathering info about the player, if it's not already used of course

Certainly can!

hellyeah222

Thanks for the reply.

The thing that has always confused/bothered me is the fact that I can end up in a situation, where I'm playing my promotion series, for example, to get into Gold 1, BUT I am playing versus Plat players. And if I fail to win those games, the LP system determines that I am not good enough to be Gold 1, when in reality I was tested against Plat players. Do you see the disconnect here? My arbitrary solution would be to compose teams from within the same LP division, and if I am in promos to Gold 1, then I should be playing versus Gold 1 opponents.

I get that it's almost impossible to create such a system and that could lead to a multitude of problems which I am not considering right now, but can you imagine how my previously proposed example could be disheartening?

As for the other accounts, that wasn't really a complaint on how the new accounts work, cause I understand the principle. I use the additional accounts to play different roles, since it feels bad to practice something new at a level where I can't keep up yet.

Totally get the disconnect here - and something we'd like to improve in the future. It's unfortunately not as easy as guaranteeing you have Gold I opponents, because not all Gold I players have the same skill (including you - from what it sounds like, you may be climbing upward and have a higher MMR than your current Rank).

In general, we match on skill rating, not rank, because skill rating matching creates matches that play more fairly, even if they sometimes LOOK less fair. Let's hypothetically say you are a player with Platinum skill, playing in Gold II. If we match you against Gold Is in your promos, you will have a much greater likelihood of winning, because we honestly think you're better than that - but this also means we've made a game where 5 players are heavily disadvantaged. Instead, we try to make a game we think is fair for you based on skill rating, and then let your LP move more quickly for each win in order to get your rank and skill caught up.

Promos do create a complication for this system, though, because every win in a promo counts the same, regardless of who you're playing. There's a lot of good things promos do (creating high moments in the ranked season, breaking up the normal LP grind), but they do cause a lot of pain as well. They're not changing in Ranked 2020, but we're working on new ideas for them in the future.

Gkkiux

It's just that jumping ranks without promos seems like something that might happen in placements, not regular season. Unless I was placed into gold promos and the game just didn't show it, not sure if that's a thing

Yeah, I think this is a really good point - I'm going to look into how we can change this experience. It does seem silly for us to say, "Hey you're blasting off at the start of the season!" and then jump you a bunch AFTER that's done

Dingodogg

Thank you so much for sticking around replying to every comment, Im reading every single one and it's honestly refreshing to hear about this matter from Riot itself.

May I ask you to clear up a thing about mmr? If you're allowed to say it ofc. Can you confirm that LPs gains are exclusively dependant on win/lose at that given mmr (and how big is the difference between your personal mmr and game average mmr between the 10 players) and that win/lose is the only factor taken into consideration for mmr variations?

Basically, is win/lose the only stat that matters to the system? Aka game performance by stats (kda, vision score etc) are totally irrelevant? Im asking this because in s9 many people are still not sure about this and I've seen some people spreading what I think is misinformation, and there's no official source for people to actually verify this.

Thanks in advance, Im no native speaker, I hope it wasnt too confusing to read.

I'm definitely not getting to all of them - I still occasionally have to do work ><

We don't take KDA, Vision Score, etc. into account for MMR, because those statistics are gameable. We don't want to create a situation where your support decides their odds of winning the game are too low, so they instead focus on maxing out Vision score to save as much MMR/LP as possible.

It's this weird problem where those stats are really great for figuring out how good a player is - right up until the player realizes they're being measured on those stats.

KaisaPermanente

Yes that makes sense, however the example i have is the reverse. I started my first alternate account expecting to be lower than my main was at the time which was silver 3. When i finished leveling the account and finished placements, the account ranked at gold 3. I then climbed that account to gold 1 over the course of about 30-50 matches. When i tried to push my main account to gold afterwards, it was hardstuck at silver 2 and 3. I would switch back and forth between the accounts, and found i would win and climb much easier with the gold account than with the silver account.

Since then, ive abandoned two other accounts because a newer account climbed past the old one, while the old one seemed to be hardstuck. One of the things i noticed while climbing different accounts was the difference in LP gained and lost on the account with lots of games was always less, even with a much higher win streak(10-20 win streak vs 5-10).

If the system is built to always put me where i belong, logic would say i should always tank on my alt accounts over the course of the next hundred or so games, but its the opposite for me and a few of my friends. Obviously all anecdotal, but id be happy to dm you some of the accounts for an example.

Sure, if you want to send along the summoner names I can probably give you a better answer. There's a lot of components to Ranked so it can be hard to explain with incomplete info!