Suggestion for improving SR Algorithm

Competitive Discussion
1

This is not me complaining but more of a white paper situation analyzing an issue with the algorithm in SR calculations.

For those of you not aware there are 3 main factors Determing the SR.

  1. Win Or Lose ( Determines on Positive or Negative gain )
  2. Predictive match analysis. During MM the game decides which team is likely going to win. That team essentially gets less SR gain per win and more SR loss if they lose. The opposite is true for the opposing team.
  3. Individual Performance - is also a factor in the Degree or SR gain or Loss.

So now that is out of the way, the main issue is with Leavers and Steam Rolls.
If the game considers you the “Underdog team” - the game is betting on you to lose. So the amount of SR Loss is less.

But the issue comes how can a person perform well in a situation where they are against overwhelming odds?

If for example, you have a leaver on the team - you are already in a 5 v 6 situation. So the focus fire against individual players is already much higher. Confirming a kill in this situation is fundamentally much harder. So by circumstances, the chance of you performing “well” is going to be less likely an option.

Again in a Steam Roll Situation - you are getting steamrolled because you and your team cannot confirm kills. Again individual performance again is not going to be a great indicator in this situation.

So how do we resolve this issue?

We change how the Algorithm fundamentally works.

If the game ends in less than 5 minutes and the average scores between the teams show a difference of more than 50%, personal SR performance is not calculated. The losing team does not lose SR.

Why is not lose SR?

For 2 reasons.
People who purposefully tank their rank do so to lower their rank. By throwing the match for an easy steamroll they will constantly lose sr. By having 0 SR loss in steam rolls there is no longer any way for them to purposefully tank their rank.

The second reason is - the team that lost does not necessarily mean they got suddenly worse. It just means they were placed against overwhelming odds. This will allow there be a less dramatic shift in ranks and people will not feel bad at all for losses.

But then how do you lose SR? and What is the new Algorythm?

So the way that SR is calculated needs to be different. No longer is it going to be weighted on predicted win and loss. Instead, it will be a match performance matrix.

It will first compare the average scores on the teams and make sure the performance between the two teams is similar or not. If it is similar to a degree then the match was balanced.

If the teams were balanced based on actual performance analysis not predictive analysis. Then the game can give positive and negative SR. As well as calculate individual performance.

Yet one change I recommend for individual performance is factor in time played. After all you can’ expect a person who played Mercy as an emergency healer for 5 minutes to have 8000 heals. It should only expect them to have roughly 1000 - 1500 instead.

I think these changes to where SR loss only happens if the performance between teams are near even would really balance the system and make horrific losses far less painful.

Anyone is free and welcome to comment.

2

Can we stop with all these “losing teams should not lose sr” threads? It’s never going to happen and just makes you look like an entitled child looking for an 8th place ribbon. Learn how to lose and improve on it.

2 Likes
3

Actually, my suggestion does not remove losing SR.
It only lowers SR if the performance between two teams are even.

That is the Difference between Predictive Analysis and Performance-based Analysis. One is subjective, one is more Objective ( not perfectly but far more than the other )

My Suggestion also solves intentional throwing to derank.
As now people who throw matches will have a much harder time to lower their rank purposefully.

Also, it removes the punishment for having griefers on your team who intentionally throw a match.

Or in other words it solves multiple problems.

4

How will your system determine legitimate steamrolls? I just played a game like 20 minutes ago where my team got annihilated. We didn’t have a thrower. We had 5 dps mains and our tanks and supports were completely terrible. We lost in 5 minutes. We deserved to lose sr because the other team was better than us.

1 Like
5

First of all you have to keep in mind there is no such thing as legitimate and illegitimate steamroll. The game is putting weight in the fact you reached your current rank because of your skill.

So it will only lower SR when the performance between the teams are even. If there is an even performance between the teams - it shows the match was fair. That means one team did not perform as well hence their rank should be lowered.

If one team gets utterly destroyed it shows the match was not fair at all. So the losing team instead gets a pass - while the winning team gets a small minor boost. The game at which point will try to shuffle the teams to try and create to teams where the final performance has only a deviation of 33%. If it does SR can be better awarded correctely.

This will prevent drastic shifts in SR and better put people in the correct SR level.

6

I can tell we will never come to an agreement on this. There are absolutely games where one team will just completely outplay the other. Not every game between closely skilled teams will come down to the last second. Some guys have amazing games where they play out of their mind and carry. Some guys make critical mistakes that snowball. Your system will not punish players for performing poorly and even worse, not reward someone for performing great. That is the worst thing you can do to a competitive system.

We’ll just agree to disagree.

7

That is not what a performance-based analysis is - simply waiting for overtime.
I think you are imagining a system with your own ideas and trying to argue aginst it.

The performance-based system instead looks at the player’s scores - not how much the objective was completed.

It looks at;

  • The Average K:D Ratio of each team
  • The Average time a player is alive between deaths
  • The Average amount of healing between teams
  • The average amount blocked between each team
  • The accuracy of characters between each team.

Essentially all the performance analytics not related to the objective.

For example, say the Losing team could not even move the cart 10 feet.
The game doesn’t care.
Instead, it will look at how well each team performed - and if the scores are near even. That means it was not a steamroll but a fair match where one team outplayed the other. Doesn’t matter if one team when 99% to the objective or only 1%.

It is the comparison of scores between the two teams that matter in the end…

8

It sounds like you’re arguing for the team that got destroyed to not lose performance based sr. In this case the winning team will always have much greater stats, thrower or no thrower.

9

Yes - which is why there is a margin of reasonable separation.
For example, if the winning team had 30 kills and the other team had 3 kills.
It is obviously a steamroll.

If the Winning team had 30 kills and Losing team had 22 kills.
It would be within a reasonable seperation to be considered even.

There will of course be more stats to look at, but this is the most basic of ideas.