@blodir said in In the current system, rating 1v1 games is borderline rating manipulation:
--> any reason that has to do with "I want to raise/lower my rating through tournaments" is the very definition of rating manipulation. Seeing as their tournament performance doesn't reflect their performance in the games that global rating is used for. This response also goes for maudlin's reply. I get that there's a problem with returning players, but trying to "fix" rating through manipulation is not the way.
I don't know what exactly is the best way to fix the system (no small amount due to not having read the paper on trueskill), so I can only talk about the problems, not really on how to fix them. That being said, it seems like there's an assumption in the system that 1v1 and teamgames are comparable ie. 1v1 performance gives a good description of teamgame performance, which might be true for shooter games (idk I don't play them), but it is not true for supcom.
This is the fulcrum of it. You have mistake “the point of global rating” with “what I use global rating for.”
At no point in time was global defined as teamgame rating. At no point in time was global defined as map gen rating. At no point in time was global defined as 5v5 rating. That’s the way you use it.
Someone can theoretically only use global to play custom 1v1 games, then a global that only has 1v1 data is accurate. Someone can occasionally play 1v1 games, then a global where it occasionally accounts for 1v1 games is accurate. Someone joins your 5v5 map gen with 1v1 tourney rating, the game will now be inaccurate. You join with your 5v5 map gen rating, the tourney game will now be inaccurate. The only solution is making a rating for every FAF defined “skill zone” which only accommodates half a dozen high rated dudes at best while drastically increasing and prolonging one of the worst parts of FAF for everyone else (being an unknown entity in the rating environment).
Then you may argue that 1v1 weighs more than 5v5, but of course it does. I have won games where I’ve done literally nothing just because my team outclassed my enemy. That cannot happen in 1v1. And if you actually hold factors equal, a 1v1 is likely going to result in something like double the point swing that a teamgame would in the logic of the system. The bigger issue is you cannot hide a 2600 with 1600s against 1800s and a 2200 in a 1v1. It’s easy to hit 95% balance quality or whatever even if you’re 500 rating (read: smth like a 95% win probability) higher than the next best player.
Arguing the 1v1 rating is inaccurate but the 5v5 rating is accurate is also strange when there is also the rationale that the hard part of global is actually rising in rating. Keeping a rating is much easier. Precisely because of the dynamic I said above, it’s just a less extreme version of what Sid is currently doing.
The convo reminds me of the problems Chess has with massive rating discrepancies in their Elo system that FIDE tries to resolve. They first made an addition that caps any rating discrepancy as though there was a 400 rating discrepancy in order to not make it a waste of time for high rated players to play anyone less (this results in .8 points per win). They recently passed a new rule that only lets you utilize this rule once per tournament and beyond that, the games are counted as the original Elo calculation intends, which is essentially a <.01 point gain. That was to stop people artificially increasing Elo against way worse players due to the original rule.
Rather that fuck with 1v1, one can make the entirely different argument to unrank high rating discrepancy teamgames (sorta similar to the situation FIDE found itself in) since 1v1 is closer to an accurate metric of skill in general and the logic of teamgame rating calculations is just derived from the 1v1 system. All rating systems tend to suck when dealing with ridiculous rating range discrepancy, hard for a person to actually win every game except 1 in 300 games against somebody 800ish rating less, particularly due to how humans aren’t programs that stop learning.
Because I feel like some dude is going to point out that I contradict myself by saying that this task of consistently beating 800 rating worse players is hard but gaining rating in teamgames against them is easy, I will reiterate that the problem is you can make a teamgame where the odds of beating them is 50/50 or 60/40 rather than the .3/99.7 of a 1v1.