Why I believe using 50% winrate as a metric for matchmaking sucks:

Why I believe using 50% winrate as a metric for matchmaking sucks:

To explain I will need to introduce some control terms. I will try to bring this back to league as much as possible.

Proportional Control: This control scheme corrects for errors proportionally to the size of the error that occurred. For example; if someones winrate jumped to 60% a proportional control scheme would aim to push the winrate down 10% by adjusting its matchmaking to give that player a 40% winrate. As you can expect, this scheme leads to a hill-and-valley approach where you rarely are at 50% but you tend to stay within some predefined bounds (say 55% to 45%)

Integral Control: This control scheme corrects for the area between the error signal and the correct winrate. Take for example winning 60 of 100 games or winning 30 of 50 games. Under proportional control these are both 60% winrates and corrected the same. Under Integral control, because the error signal is sustained twice as long (100 vs 50 games) it is corrected twice as hard. This approach can catch extreme outliers but will generally not respond forcefully enough to small error rates unless they are sustained over an incredibly long period of time.

Derivative Control: This control scheme corrects based on the slope of the error signal away from the correct winrate. In this scheme the faster you move away from the correct winrate the more intense the correction signal. This scheme completely fails to pick up players with consistent winrates either above or below 50%. If you are not moving away from the correct winnrate there is no correction signal sent. Because this control scheme is completely incapable of correcting for some types of errors it is rarely used solo and instead is combined with proportional and/or integral as a failsafe to correct for rapidly deviating signals.

Okay, so why does all this matter. This matters cause I do not believe that matchmaking implements these control schemes in a manner that supports player growth. This has to come down to the metric of establishing a 50% winrate. If wins and loses are your only metric than both proportional and integral control will restore 50% over time. They will accomplish this by providing you with games you are almost certain to win (if you are below 50%) or games you are almost guaranteed to lose (if you are above 50%). The degree to which your victory or loss is assured will be determined by whichever control scheme is implemented. As I mentioned above, this will vary by control scheme.

If they use proportional control you will rarely sit at 50% and instead will go on hot and cold streaks as the system hands you impossible or guaranteed victories.

If they use integral control correction signals will be lagged as your error builds up over time and will only result in guaranteed loses or victories if your winrate maintains a difference from 50% over a long period of time or after a large magnitude shift from this winrate.

If they use derivative control you would be almost guaranteed a loss after a winstreak (and vice versa for a losing streak).

The issue here is that ONLY victories or loses will result in a correction signal being sent if winrate is the metric used to establish error signal deviation. This leads to the "elo hell" concept as you have no personal agency over what your error signal will be. Teammate dc's, congrats! you now get a positive correction signal from the system and your next games should be easier.

Cool story, bro so what... I would instead propose shifting the metric to match long-term performance of the role. For example, mids traditionally have X% of team gold, x% Kill participation, x% deaths of team and so on and so forth. In this situation, your individual performance is what is analyzed and the correction signal sent is entirely in your control. Match or outperform your roles average performance? Your opponents would get harder. No impact from other members of the team aside from the fact of how their role and yours intersect (ie kill participation).

This could even be adjusted by league, so you would have to outperform bronze mids on average to move from bronze to silver.

Additional benefit: Smurfs would be easier to identify as their performance could be more easily separated from win/loss.

TL;DR 50% winrate metric sucks as an analysis tool and contributes to the creation of the "elo hell" concept. Assign a different metric that is more in the individual players control to foster an environment of improvement over blaming.