GLICKO formula is needs revision IMMEDIATELY!

Once you have played many many games, if you lose a game you will lose 6 points but if you win you only win 5 points. This makes it extremely hard because is a negative formula where you are losing more than you are wining in each game.

This happens with players that are very close to you in rating of course.

example: lichess.org/bm5AkChImprY

I have been in the other side of the board too losing 6 and winning only 5.

What was the point of showing the game? @rampageJackson

The argument you have presented is not statistically significant...

"is a negative formula where you are losing more than you are wining in each game."

If that was true we would observe a slow but steady decrease in overall ratings. I don't think this is the case.

Two guesses:
1) Ratings are not integers but floating point values even though they are shown as integer. So when one person loses 1 point more than the other wins - provided both have equal ratings and an equal deviation value - he was near the rounding limit. (I didn't bother to look this up in the source code.)

2) You are suffering from confirmation bias, meaning you will note it when your assumption comes true but you don't if it doesn't.

You are absolutely correct, @Katzenschinken . The rating change shown there is the change to the whole number part of the floating-point rating, dropping any decimals.

Under the Glicko-2 system, if you and your opponent have equal RDs, equal ratings, and equal ratings volatility (sigma) values, you will each gain and lose the same amount in a head-to-head result.

If both of your ratings were 2300.9 going into the game and the rating change for both was 5.5, the winner would have 2306.4 and show a +6, while the loser would have 2295.4 and show a -5.

And that's before even analyzing the effect of a different sigma value, which would indeed introduce an imbalance, with a larger rating change for the more inconsistent player.

So this is resolved? there is no issue and everything is perfect based on the theory above. No facts whatsoever. You gotta love the science and scientists.

Because based on facts I have seen many times ratings go up 5 and lose 6. That is an observable fact not theory. At the very list someone in the dev list can take a look perhaps?

@rampageJackson The source code is publicly available. If you think there is an issue with having the Glicko calculation not implemented properly you can look it up yourselves and check it.

If you think there is a problem with the Glicko formula itself: This is also publicly available. The author of that formula has his own webpage and certainly welcomes every feedback. Well-founded feedback that is, of course.

Oh, so you are challenging the Glicko system? What do you recommend changing in the formula?
http://www.glicko.net/glicko/glicko.pdf
The equation is used for post-game rating is at step 2 in page 3.

Maybe to make it more transparent we can show the decimal in the ratings so instead of 2340 show 2340.6 etc?

@rampageJackson

This feature is already (partly) available. If you click on one time format on your profile (say blitz), it not only tells you that your current rating deviation in blitz is 45.78, it also tells you your current rating to the nearest two decimal places (in your case atm 2341.02).

I believe Lichess should keep showing ratings in their rounded form by default. I find it unlikely the minute difference will bother many players and integers are a bit easier to read than messy floating point numbers. But maybe you can convince (@)thibault to give players the ability to disable the rounding of the rating in their settings?