Use geometric mean for matchmaking

UgyseMered

Hi,

I think most of the player base can agree that the queue times are far from optimal now. Apart from the diminishing number of killers, the current matchmaking system plays a major role for it. The problem is:

• High rank survivor queues are very long on average
• Low rank killer times are even worse

The current system will find a killer with rank related to the highest survivor rank, completely disregarding all the other survivor ranks - which is bad because 3/4 of survivors are not considered at all when determining the opponent's level. This is not only leads to the depletion of high rank killer pools, it also makes the general match experience a lot worse for imbalanced teams. My suggestion: use geometric mean to determine killer rank as it will give an average closer to the highest rank but still consider every player. It is also not expensive computationally and is easy to implement. Some examples to give intuition why would this help:

• Survivor lobby: 2, 15, 15, 15.
  • Killer rank with current matchmaking: 2
  • Killer rank with old matchmaking: 12
  • Killer rank with geometric mean: 9

Here the old matchmaking system would be quite ok - the killer has solid chance to meet a survivor of his level, and not be bullied. The current matchmaking on the other hand (with rank 2 killer) will make the match a nightmare for the lower rank survivors. Using geometric mean would give a better experience for the survivors.

• Survivor lobby: 2, 3, 16, 18.
  • Killer rank with current matchmaking: 2
  • Killer rank with old matchmaking: 10
  • Killer rank with geometric mean: 6

Here the old matchmaking system will make the match a nightmare for the killer - the two high rank survivor can destroy and bully the green rank killer - even more so as lobbies like this are groups of 2 SWF. New matchmaking system favors killer, they will finish chases really fast on the lower rank survivors - making the game not fun once again for new players. Using geometric mean would help both cases.

• Survivor lobby: 6, 7, 8, 9.
  • Killer rank with current matchmaking: 6
  • Killer rank with old matchmaking: 8
  • Killer rank with geometric mean: 7

As the team is not imbalanced, every ranking system gives a quite good result. Geometric mean suffices in these cases too.

Also keep in mind that this suggested method would pull killers from lower ranks too, so the high rank killer pool wouldn't get that much focus. In the above examples, instead of getting two red rank killers and one purple, lobbies would get 3 purple rank (or 2 purple and 1 green as killer of an average rank 9 in the first example can easily be rank 11 too for example).

TLDR: Using geometric mean would improve survivor rank que times, lower rank killer que times, and survivor game satisfaction.

Some blah blah with math intuition:

I think this is because the survivor rank doesn't correlate linearly with survivor skill, rather exponentially. So the old system's way of using arithmetic mean is flawed, as it assumes additive relation between ranks. That is: Rank 18 is twice as unexperienced as rank 9. Which is obviously not true, a rank 18 player on any side (but i think it's more unforgiving for killers) has a lot less experience that a rank 9, not just twice. Geometric mean would assume a multiplicative relation of ranks, which is quite true - not perfect of course but closer to the truth.

The current system of only considering the highest rank is obviously very flawed too, as it completely disregards information - which is bad design.

Arisquelle

I came here from my other post. I agree that I like this method for better ranked matches but there is also the problem with survivors leaving and entering lobbies. At first the ranking might be accurate but then one person leaves and the game eventually puts in another survivor and their rank doesn't seem to be factored into the equation.