/dev: Matchmaking in 2024

In our last Dev Update we got quite a few questions from players across the globe about matchmaking, so today we’re going to take a look at how the matchmaking changes this year have landed and, more importantly, what we’re currently working on. 

Negative LP Accounts

Earlier in the year, we had an issue causing accounts to lose more LP than they gained, even if they were performing well. We made some changes to remove Negative LP for all tiers and as of Split 2, we’ve resolved these cases, except for short periods of time when players are losing a bunch of games at Tier boundaries. We also reduced queue times for Iron-Silver players by about a minute and LP gaps within team by about 1.5 divisions to bring the average LP gap between teams to around 1 division (eg. Silver 3 to Silver 2). 

These changes landed even better in China, as it reduced the LP gaps in Iron - Silver from 7 divisions to 1,  and decreased average queue times by 30 seconds. We also forced “winrate harvesting teams” to play against each other :) (winrate harvesting teams use manipulated accounts to maintain 100% winrate over hundreds of games). 

LP Difference, Blue and Red Side

Another matchmaking space we revisited was Red vs Blue side balancing. In 14.17, we made some changes to reduce the average LP difference between teams from 2 divisions to 1. We also made changes to make blue and red side more balanced, especially in high MMR. These changes also reduced the incidence of certain high rank players placing repeatedly on the same side.

Autofill Diffs

Autofill is one of our greatest challenges. It’s a necessity because the reality is some roles are more popular than others especially in different regions, but if everyone always got their primary role queue times would be much longer.

When we took another look at this space, we found that our current configurations to ensure autofill parity in some portion of games were resulting in too many games with mismatched amounts of autofilled players. In 14.17, we shipped an improvement that increased our rate of 0:0 and 1:1 games, and effectively eliminated  2:0 (and worse) games. 

We try to prioritize either 0:0 or 1:1 autofills on teams and try to limit swaps that would give one team an advantage. In some cases however, we need to have autofill imbalances, such as 1-0 or 2-1 to prevent queue times getting too long. 

New Account Seeding

Earlier in the year, we also had a few issues causing some players to be placed a lot higher than intended. So we improved the seeding formulas to more accurately place players based on their skill level. We also added a 10 normal game requirement before playing ranked, so now we’ll have a much more accurate read on a player’s skill level, rather than bypassing our checks by playing ARAM and bot games. 

Our aim is for new players to be placed with similarly skilled players  and achieve a ~50% winrate. While we’re currently sitting at a ~46% winrate for new players, with future adjustments and tweaks we expect to be much closer to 50% in the future. 

The Future

While we’ve been able to make some solid improvements in matchmaking this year, rest assured, there’s still more to come. Here are some of the matchmaking issues that are currently on our radar. Solving each of these is very much a balancing act of finding the right tradeoffs between queue times and match quality, both of which are critically important for the health of League.

Seeding Smurfs:

We’ve seen much better placements with the 10 normal game requirement to play Ranked which is helping us seed new accounts into ranks where they average 50% win rates.

However, smurf accounts for higher skill players still take too long to get to their true rating, sowe’re investigating ways of improving that with the goal of reducing the amount of smurfs in skill brackets well below their actual level.

We’ve also got an evaluation underway on a system called TrueSkill2 at present. That could potentially help out meaningfully with things like better placing new accounts and detection of smurfs. As of now we’re actively testing its effectiveness and determining how good of a fit it is for League. We’ll let you all know once we have an update on what that testing has shown. 

Skill Level Consistency Across Queues

We have a large majority of players whose queue of choice is Ranked solo queue, and while that isn’t an issue, we want to create some degree of parity between queues. At the moment, when high ranked players play normal or flex games occasionally, it can cause quite imbalanced matches as their solo queue rating isn’t referenced for normal or flex. The goal here is to take their overall skill level into consideration for matchmaking so games aren’t excessively lopsided.

Leveling Role Queue Disparities

As mentioned in the table above, there’s a pretty wide variance of least popular role by region. As a result, a one size fits all regional solution doesn’t really exist… right now. We’re on the hunt for better matchmaking or incentive solutions here that can level out some of the extreme disparities and lower queue times without compromising on match quality.

Autofill

We’re still trying to find optimizations for which roles get autofilled. For example, if we have to make a 1v0 autofilled game, should we make some concessions to the team with an autofilled player? Alternatively, if there’s 1v1 autofill, should we try to match them in the same role?

Our early explorations have found that sometimes we can do this, but sometimes we can’t without increasing queue times (which is a non-ideal tradeoff to make), so we’re still investigating potential solutions for this issue.

Match Balance

While we’re always balancing making games more equal in terms of skill level, as noted earlier we always need to find the right balance between weighing the importance of how close skill levels should be against how fast queues need to be.

One area we’re looking at is how we’re matching players based on the role they queued up for and their lane opponent’s skill level. Imagine a world where each player is close in skill level to the opponent they’ll be playing against most within a game. While we have a few concerns to work through here, we think this is a promising area to look for improvements in.

Return to Ranked

Lastly, we still want to improve the “return to Ranked” experience for players coming back to the game. Currently, players returning to ranked players have a ~46% win rate on average, but there’s still room for improvement there as we’d like for that number to be closer to 50%.

Matchmaking may be a complicated space, but having high quality matchmaking is an important part of making League the most competitive game in the world. For that reason, we’re constantly looking for opportunities to improve it for players. We’ll keep you all posted on any big updates as we roll them out, but for now thanks for playing with us!