Losing ladder rating when fighting really low-rated players

[FunkOff]

It's happened twice now (in recent memory). I've just searched ladder. Gotten somebody ridiculously low rated, beat them, lost rating for it.

See attached photo. I beat this guy who was rated ~200. (I'm ~1300 atm.)

I can understand not gaining points, but why LOSE them?

[Cuddles]

It's QAI adapting it's rating algorithms. Statistically you where trending to be over rated. This is a correction to the anomaly.

[codepants]

For my own education, how does QAI determine if someone is "trending to be over rated?"

On the FAF wiki, here's what it says about rank:

- Your rating increases when you win a rated game, and decreases when you lose a rated game. In-game performance, beyond win or loss, has no effect.
-Your rating, the rating of your opponent, your rating's volatility, and the volatility of your opponent's rating determine how much your rating changes.

I see nothing about statistical trending -- just the game at hand. It also explicitly states that "your rating increases when you win a rated game."

Even Elo rating, which FAF rating (I believe) is based off of, takes only into account:

Rnew and Rold are the player's new and old rating respectively, Di is the opponent's rating minus the player's rating, W is the number of wins, L is the number of losses, C = 200 and K = 32.

Assuming I am reading that correctly, the only elements not related to the game at hand are W and L, so you'd have to have already lost more games than you won to lose score on a game you just won.

[FtXCommando]

FAF uses trueskill not ELO.

Trueskill does not account for “overratedness” but rather uncertainty.

[codepants]

Well, the Trueskill wiki claims,

mu always increases after a win and always decreases after a loss

where mu is the player's rank.

So the question still stands: why did FunkOff's rating go down if they won?

[ThomasHiatt]

Your rating is mean - (3 * deviation). You played someone with a very high deviation so the rating system is uncertain about how to interpret the result so your deviation increases slightly. Deviation has 3x more effect on displayed rating than mean, so even if mean increased slightly the displayed rating will be lower due to the change in deviation. It is explained here https://wiki.faforever.com/index.php?ti ... kill_works under the section "Why did I lose points/don't gain point with a win ?" It is almost incomprehensible though since it was probably written by someone who doesn't understand English or care about proofreading.

[ConditionZero]

Think of it this way

Your displayed rating was not well known by the system, but rather a 'range'. The system displays a number in the middle. As you play more games, the 'range' that your rating is will decrease, becoming more accurate. If the middle of that range is lower than the original middle of the less accurate 'range', then the displayed rating will be less than before. Your rating hasn't decreased exactly, it became more accurate.

This happens more often for new players. Their range could be wildly inaccurate, and display a middle number of say 1600, but as they play more games, even against lower rated players, that range decrease significantly and alters that middle number thus decreasing the displayed rating.