|
|
|
Author
| Message |
|---|
|
|
Three bugs in the implementation of the rating system...
1. [edit -- this point is incorrect because, as pointed out by Blutigeroo, below, I'd not noticed that Bergman's opponent had a provisional rating when this game was played.] I was idly looking through user Bergman's games (wondering when he last lost one) when I came across this game, a draw between him (then rated 4501) and a player rated 1947.
http://www.queenalice.com/game.php?id=811139
According to the description of the rating system at
http://main.uschess.org/content/view/7520/393/
the following should happen.
1.1. The rating difference is more than 350 points, so it is treated as being exactly 350 points.
1.2. Bergman's rating is over 2400 so his new rating is calculated as 4501 + 0.02*(-350) = 4494
1.3. The opponent's rating is under 2100 so his new rating is calculated as 1947 + 0.04*350 = 1961
But, as you can see from the game, neither player's rating changed as a result of the draw: Bergman remained at 4501 and his opponent at 1947. This is incorrect and absurd -- a 1947 player played well above his rating to get a draw with a 4501 so his rating should increase; the 4501 played well below his rating to only draw against a sub-2000 so his rating should go down.
2. One of my current opponents has a provisional rating. His first game
http://www.queenalice.com/game.php?id=726085
was a draw against a player rated 2083. According to the description of the rating system, my opponent (DNWheeler) should have been given a rating of 2083. Instead, he was given a rating of 1100.
3. The description of the rating system for provisionally-rated players on this site at
http://www.queenalice.com/about.php
says, "Special rule: if the opponent is rated 400 or more points below or 400 or more points above, then the rating will not be affected no matter the outcome."
This is a mistaken interpretation of the USCF correspondence rating system (linked above). The USCF page says that your provisional rating does not change if you beat a player rated 400 or more points less than you or lose to a player rated 400 or more more than you. In particular, this means that if you beat a player rated 400 points more than you, you should gain a lot of rating points -- and quite rightly. However, on Queen Alice, you gain no rating points as a provisional player if you beat somebody rated 400 points more than you.
This is completely wrong. Imagine that Gary Kasparov decides to start playing on Queen Alice. In his first game, he accepts a challenge from a player rated 1000, which he wins. Gary's rating is now 1400. In his next 24 games, Gary beats players rated 2500. But Queen Alice says, "No! These players are rated more than 400 points above you so they don't count towards your rating!" This is absurd -- Gary has beaten 24 players rated 2500 but, instead of having a rating of about 2900, he has a rating of 1400. The whole point of a provisional rating is that it should quickly reach approximately the right value. It isn't supposed to stay extremely low just because somebody was unlucky enough to play a low-rated player in their first game.
This has badly affected DNWheeler's rating -- after being given a rating 983 points too low for his first game, he's been stuck under 1450 because this second error means that his wins against higher-rated players aren't doign anything to lift his rating.
Remember: the rating is supposed to be a guide to how strong a player is. Under normal circumstances, if you're winning a lot of games, your rating should be higher than that of the people you're beating.
|
|
|
Good point but are these three items not all the same one? You may not have noticed
that AFA1961 was provisional at the time of that game with Bergman.
So you really are just pointing out the apparent misinterpretation of the USCF rules in regards to the "Special Rule" on Queen Alice.
I've never looked at any of the rules before but on first glance, I'd have to agree with you.
On the other hand, miguel may have had a good reason for implementing that rule for online play.
Perhaps the protected ratings have something to do with it and/or making account switching more difficult.
|
|
|
You're right -- I'd not noticed that AFA1961 was still provisional in the game I linked in point 1. So, in that game, it's correct that Bergman's rating didn't go down (it was protected) and AFA1961's rating not going up is the same misinterpretation of the USCF rule as in point 3.
But point 2 is different. If unrated player A beats rated player B, then A's rating after the game is supposed to be B's rating plus 400; if the game was drawn, it's supposed to be B's rating; if A loses, it's supposed to be B's rating minus 400. That didn't happen in the DNWheeler game I linked.
|
|
|
Yes sorry, I looked at the entire sequence of DNWheeler's games.
I forgot you had specifically pointed out the first game although there is a significant group that fall under the same special rule.
While that first game goes beyond the special rule, it is related and I can see why it would be
important not to give the opponent's rating for that first game.
If you accept the special rule, then you must also accept the first game condition.
Otherwise a expert player can play Bergman to win or draw and then win 24 games against the lower ranked players.
Thereby achieving a rating even higher than Bergman. Bad enough to have just one of those!
|
|
|
I think there is a pretty strong contingent that believe the rating system is flawed. The Bergman problem could be very easily eliminated (or at least reduced) by getting rid of the "350 rule," which awards at least one point for a win, no matter what the rating difference. The "400 rule" for provisionally rated players could also use some work. I gather that it was instituted when some highly rated players seemed to be losing deliberately to their provisional friends. But it does have the undesireable consequence of making your first game critical to your short-term future, because if you lose that one you are stuck with an 1100 rating, and can only get ahead by playing (and beating of course) players whose ratings are < 1500. As for the DNWheeler case, I agree that it seems wrong, even if you accept the "special rule," because he ended up with the same rating as he would have if he'd lost.
|
|
|
| If you accept the special rule, then you must also accept the first game condition. |
I don't accept the "special rule". The point of provisional ratings is that they should rapidly move to roughly the right value but the "special rule" means that your provisional rating is actually stuck at around the value determined by your first game. Not only that but, because of the "first game condition", the value determined by your first game is completely bogus. Drawing a game against a 2000-rated player does not suggest that your rating should be 1100 -- it suggests your rating should be about 2000!
Yes, the existence of people like bergman with crazy-high ratings means that somebody could end up with a rating of 4000+ just by losing their first game. But the correct way to deal with that is to prevent people from getting crazy 4500+ ratings, not by messing around with the provisional ratings in a way that makes them completely meaningless.
Let's not throw out the baby with the bathwater...
|
| 1 2 3 Next |
|
|
|
