This is the much needed follow up to our previous blog where the ranks were reset.

We have been regularly analysing our rank  data over the past two months to find out areas where the system could be improved upon. Based on our findings, there were two areas that needed improvement: Lobby changes/Modifiers and Rank Ranges.

Before we begin – please go a through brief overview of how our system works and has been working since Day 1 of Matchmaking. The Glicko2 system is what we base our matchmaking algorithm on.

You earn/lose your Glicko  based on your own Glicko compared to the other team’s average Glicko when you play a game. Eg. if your own Glicko/Rating is 1500 and you face an opponent team with an average Glicko Rating  of  1300. On winning, you would get very few Glicko Points  which is decided by the Glicko2 Algorithm. Losing against the same opponent, you will lose a lot more Glicko points compared to if you won against that opponent. You get the general idea of how Glicko2 operates. This has been in place since Day 1 and has never been changed or modified and is still in effect.

The MM system tries to find the best possible match compared to your Glicko Rating. If it isn’t able to find the exact match similar to your Glicko Rating – it will try to compensate by  mixing and matching different ranks searching at that point to average out both teams’ ratings. This has been in place since Day 1 and has never been changed or modified and is still in effect.

The Lobby Problem – (Changes to Modifiers)

What are Modifiers?

Bonus and Penalty modifiers. These were added in the system to reward high performers and punish low performers. Modifiers made no change for average performers. 

In the pre-reset MM model, the modifier did not care about the lobby sizes and average Glicko difference between the teams. This caused a huge influx of 3+ lobbies farming these modifiers and causing the bloat in the system which ultimately led to the reset. 

In the post-reset MM model, the modifier simply omitted the 3+ lobbies from getting any bonus whatsoever and substantially increased the eligibility criteria for bonus from 1.2 to 1.4 times of the average team IR for Solos/Duos. We realized that these changes made rank up extremely hard since getting bonus modifiers became extremely difficult. And to top it all – Avg Glicko between teams was not considered while rewarding bonus or punishing players.

The Fix  

Working on  a hybrid approach by putting primary focus on the Glicko difference for the reward/punishing mechanisms and secondary focus on lobby compositions. 

The new system does away with the wall in place for bonus modifiers when 3+ lobbies are playing. Every lobby now irrespective of size is eligible for modifiers whether it be bonus or penalty. The catch in this case is that the Glicko difference between the two teams will define the exact requirements to be eligible for those two modifiers.

After the reset the bonus modifier was set at your IR being more than 1.4 of the average Team IR. 

However, now, the requirements for the modifiers scale with opponent strength and take into account how much lower or higher your team’s average rating was compared to the opponents. If all things considered equal and team A and team B are exactly same rating wise – the requirement would be 1.2 * Avg Team IR. Which is more or less in line with the old system.

In the new system modifier number scales with the opponent strength, which means if you are facing a tough opponent your requirements for bonus goes down and for easier opponents, the requirements go up.  3+ lobbies. due to having an edge in terms of teamwork and coordination will have to work with slightly tougher requirements when facing equal opponents compared to Solo/Duos. However in principle if you are facing hard opponents irrespective of lobby composition, requirements are more or less the same when it comes to the modifiers. 

Note: This change was live a month ago and we have seen positive results in terms of expected results. Leading to the second change.

The Rank Problem – (Changes to MMR ranges)

Data we collected over the past month after the Lobby changes showed us that our Glicko Rating distribution was actually pretty evenly distributed.

This made two things clear that the lobby fixes and the Glicko2 Algorithm are actually working fine in maintaining the bell curve across the rating bucket.  But one more change was made when we reset the ranks. We had also increased the MMR range between ranks. To make it more challenging for users to rank up.

The Problem? 

The ranges were too wide to accurately assess a players actual rank. Two players in the same rank could be more than 199 points apart. This combined with all the modifications we have made to the rank behaviour, a difference of 199 can mean a lot considering how challenging the current system is. 

The Squash?

Instead of 3 ranks within a specific MMR range, we have squashed  to include 5 ranks. All ranks are not created equal,  there will be ranks that will require more effort than others. 

This helps us get a diverse group of ranks while accurately representing a players skill level through his SoStronk Rank, also aiding in improved MM now that the Matchmaking algorithm has more ranks to work with.

What does this all mean?

To put it simply, ranking up is now more forgiving and slightly easier while still presenting a challenge. We hope this gets rid of a lot of the frustration in trying to rank up in the extremely competitive environment which SoStronk offers.

Spread the joy

This site uses Akismet to reduce spam. Learn how your comment data is processed.