Profile of Kieselberg

Jump back to quoted post17 May 2022, 14:42 PMBasedSecretary

im literally +12w with this shit

Jump back to quoted post27 Apr 2022, 02:57 AMDoomlord52

There's kind of 2 auto-match threads going now, but I'll add here as well: The issue isn't arranged teams, it's how they're handled. On paper, it should be possible to match any arranged team (except for the absolute best) against a random team; a 4-stack of 5,000 rank players will not beat a random collection of top 200s, for example. The issue is the handling of arranged teams is extremely naive right now.

Firstly, as Rosbone pointed out, any new combination of players is handled as a completely new team: 4x "top 10" players playing together? Totally new team with no way of knowing their skill level until they play 10 "placement" games. Then, if 1 person leaves the group, and a different "top 10" player joins, despite it being 75% of the same team, automatch says "no, this is an entirely new team. 10 placement matches please". This is pretty much the worst way of handling it. Instead, use the data we do have (maybe they played 1v1 a lot), and use that as a starting point. Even better, use an over-all "hidden ELO" score that tries to rate the players "average" skill level for all factions and all modes that they've played, if only to place them better when trying a new mode/team/etc.

Secondly, as pointed out by Rosbone and others in this thread, the match-making system can give some really terrible matches. While it does give even matches sometimes, in less popular modes, or in less popular regions/time zones, it gives some truly awful matches. Or perhaps sometimes it gives awful matches regardless of the situation. In any event, it clearly needs to be worked on, as the current system is just not very good, and it ends up giving a lot of players a poor experience.

Thirdly, even if the existing system worked flawlessly, the actual implementation of arranged teams is pretty bad. As-is, it assumes that a 4-stack is exactly equal to 4x randoms provided their ranks are the same. This is pretty flawed, as chances are that the 4-stack has voice communications, experience playing with each other in the past, and pre-made strategies/combinations that they like. Meanwhile, the 4x randoms have none of this, and might not even communicate in the same language. As a result, even if the "ranks" are the same, we need to assume that the 4-stack will be playing much, much better. Perhaps something like:

This would mean that a full pre-made team would match with players with 20% higher ELO than they actually have, and it would scale for a 2-player pre-made in a 4v4 game, for example (a 2-stack in 4v4 would give +10%).