All SchemingMind games (with the exception of Unrated Chess and Unrated Chess960) are rated using the Glicko rating system. The ratings are updated on the first day of every month. For new members of the site, their rating is set at 1500 (350).
What is the rating for
The main rating function is to help people find opponents of similar strength. It can also help to monitor one's own progress (am I improving?) and, finally, to compare with others.
Obtaining a high rating should not, nevertheless, be treated as an aim in itself. The true honors are in winning dropouts, tournaments, games against strong opponents.
The Glicko Rating System
The Glicko rating algorithm is similar to Elo (the algorithm used by FIDE to rate chess players), but improved over it. The main difference is, that while Elo assigns every player only one number - the rating (so Topalov is rated 2783, Anand 2779, etc), Glicko assigns two numbers: the rating, and the rating deviation (RD). This is displayed by giving the rating, followed by the RD in parentheses, like this: 1623 (76).
The first number is the "best guess" of the player's actual rating (the Glicko system assumes that we never know the exact player strength, but we just approximate it). The rating deviation (RD) describes how large the difference might be between the calculated rating and the actual player strength. There is a 95% chance that the actual player strength lies between (rating - 2RD) and (rating + 2RD).
For example, if some player is rated 1740 (20), then there is 95% chance that his actual strength is between 1700 and 1780. So, we know his strength fairly well. At the same time, if somebody is rated 1900 (250), then we know only that his true strength is likely to lie between 1400 and 2400 - a very rough estimation. The maximum value of RD is 350; that value is used for new players, whose strength is unknown.
How the rating change after a game
The rating is not changed immediately after a game. Instead, it is recalculated once per month, using results of all the games finished in the period.
Some online chess sites recalculate ratings immediately after the game. Whilst this gives immediate feedback to the players, Professor Glickman recommends that a rating period should optimally include five to ten games on average. Many organizations (including FIDE and ICCF) publish rating lists periodically.
The exact change of rating can vary, depending on the ratings and rating deviations of both players. You can remember the following practical rules:
- if your RD is high, a few games can change your rating by hundreds of points (this is good, thanks to this feature new players are quickly assigned a reasonable rating),
- if your RD is low, then your rating will be changing slowly (a win against a comparable opponent will give you about 10 rating points),
- a win against a stronger opponent is worth more than a win against somebody of equal strength, a win against a significantly weaker player is worth few or even no rating points (similarly for a loss: losing to a stronger opponent will not cost you much, losing to somebody weaker can cost a lot),
- results against players of low RD change your ratings more than results against players with high RD (so a loss against an underrated new player does not cost you much).
Rating of short games
Short games, even those which lasted 1 or 2 moves, are rated, just like any other games. But only if they ended normally (mate, resignation, draw agreement).
Games which timed out before move 4 are abandoned, so remain unrated.
New players' rating
New players are internally assigned rating 1500 (350). This rating is not displayed, though, instead the site shows information that the rating will be calculated on the next update.
First true rating is assigned on the end of the month during which the new player finished some games (usually it is no later than two months since joining the site).
Ratings recalculation
New ratings are calculated and are published on the beginning of every month (most often on the first day of the month, sometimes a day or two later). All games which were completed up to midnight GMT on the last day of the month are taken into account (so it does not matter whether ratings were recalculated on the first, or third day of the month, the result is the same).
While the ratings are recalculated, for a few hours ratings are unavailable (not displayed anywhere). This is normal, once the calculation is over, the (new) ratings are displayed again. Player profile pages show then Will be recalculated on the next update text, game pages do not show ratings at all, top player pages can be empty or incomplete.
New ratings are updated in the order of player id (players who joined site long time ago have their ratings calculated and published first), during recalculation one can observe best players list to be slowly filling.
Ratings unavailability
As it was said above, ratings are not available during rating recalculation.
Also, the rating is not displayed on places like top 100 players list, if the rating is not established (high RD), or the player is inactive.
SchemingMind rating versus FIDE (and other) ratings
While SchemingMind ratings are calibrated to use similar values to FIDE ratings, those ratings can not be directly compared. First, the rating is always a measure of how somebody performs within the specified group - and those groups are of course different. Moreover, the Glicko algorithm differs from Elo. Finally, correspondence chess is different from the over-the-board game, some people can play good correspondence games while blundering often on the board.
Imagine you take 100 novices, and one solid amateur player, let them play games and tournaments, and rate them. Sooner or later the only solid player will get a high rating, about - say - 2400. Does it mean that he is comparable to FIDE 2400 players? Of course not.
On the other hand, some FIDE-rated players play on SchemingMind, and in general the difference is not large (no more than 200 points in most cases). So, you can use the SchemingMind standard chess rating as a very rough estimation of player Elo.
The same reasoning applies to comparison with the ratings of other internet chess servers. There is no exact correspondence, but - for instance - I noticed that most FICS players tend to have SchemingMind rating similar to their FICS standard rating.
Rating of variant games
Variant games are rated using the same system as standard Chess, but separately (so there is a separate rating list for every variant). Ratings for Chess960 are displayed just like standard chess ratings. For other variants, ratings are displayed in fewer places, you will not see them on on the game board page or player profile page. To see a player's ratings in variants other than standard or 960, go to the Games link on top of their profile page. (To see all of your own variant ratings, go to View My Profile in the My Details menu, then click on Games.) Variant ratings are also shown on Pyramids and on the lists of titled players.
As fewer people play variant games, and those games are played less frequently than standard chess, variant ratings tend to be more roughly estimated. And there are significant differences comparing with ratings on other sites (for instance, there are some players who play CrazyHouse on SchemingMind and on FICS, the difference between those ratings can reach 500 points).
Average Rating
The average rating of all SchemingMind players should always be close to the starting rating (1500) - keeping such a balance is one of the characteristics of the algorithm.
Actual value is changing but keeps this requirement. For example, average standard chess rating of all SchemingMind players was 1484 in April 2008.