I believe spd_iv is the 3rd player to ever reach 2500 in LG Twixt. Grandmaster level. He's also raked #1 now, and our reigning champ (twixt.ch.34.1.1).

(The other two were Maciej and Gyorgy.)

Congratulations! It must be the pretzels.

Need a like button

thx, but it is only becouse many top players are not playing now. i believe that Klaus reached 2500 in november 2005.

If the rating system is consistent, it shouldn't make much difference if fewer games are with top players

theoretically if more games are with top players than there is higher probability of loosing a game and rating. (when you lose 1 game against player who has the same rating then you have to win about 5-6 games with players who has 300-400 less points to be at the same level)

ELO is a well-balanced system. Against a whole bunch of lower-rated opponents, you'll gain points in small increments, and lose points in large chunks. Someone you can only get 2 points from can take 30 points away from you. Over time, this will average out to no gain.

Alan, I don't believe you. Why not let me win a game against you and then I will believe you.

For a player it is always better to lose first - and then win.

@Alan: _Over time, this will average out to no gain._
It is false. It would be obvious if if the _K_ factor was not constant. However, with _K_ factor constant the outcome of the rating formula calculetad for a player is rounded - more probable - or truncated (it isn't clearly explained in the FAQ) to an integer. I never tried to calculate an average rating in any game but I belive I isn't 1500. Should there be any result of such calculation let us know it. I'm courious if it's above or below 1500.

_ELO is a well-balanced system._
If _K_ is not constant.

That's true, ELO assumes you don't have any control over the order of your games. By resigning your hopeless games quickly, you maximize your rating, if your opponents are not resigning their lost games just as quickly. But I don't think you can gain much from this, and whatever you gain starts getting washed away pretty quickly as more losses and wins are accounted for.

Rounding could cut either way. I don't know how LG does rounding. You'd have to have that all figured out to benefit from it..

Non-constant K is usually used to get new players up to the right level quickly, and keep established player ratings stable. (Glicko does this.) I don't understand how non-constant K could balance the system. In fact, it can cause the average rating to drift from the starting point (1500 on LG), especially with an experienced player with a low K and a newbie with a high K.

LG, being strict about ELO as it is, always subtracts the same amount from the loser as it does from the winner, so if you average the ratings for all the players who ever played a particular game, it has to be 1500. The player lists on LG, however, drop off inactive users. You would need the complete list.

The ELO system is based on several assumptions. For example, whatever the strengths of the players involved, the difference between their ratings, independently from the absolute ratings themselves, should indicate the expected result of a game between them. There have probably been statistical analyses of the results of many games of chess which show a strong correlation between this assumption and actual game results, so the assumption there would be justified. Are tens of thousands of games enough to draw any meaningful conclusions about whether this same assumption is valid for Twixt? Am I trying to wheedle Alan into writing a Ruby sqripte? You be the judge.

Also, one factor to consider is the overall quality of play of all players as their ratings become established, when the rating system is first implemented. If the field initially had overall weaker play, and players established their ratings, and then there is an influx of many strong players who start at 1500, it seems reasonable that would contribute to a sort of “ratings deflation” as the average rating of 1500 would tend to indicate stronger ability than it used to, over the course of a few years. Then if many of those stronger players exit the playing field, the absolute value of a 1500 rating would tend to decrease, meaning more players would achieve it, and a ratings inflation would ensue. Does this make sense?

I believe I read in some earlier thread that the scores on this site are handled as floating point numbers internally. There are finitely many of them only, so still there are issues with snapping the 'real' values to the grid but the effect should'nt be relevant.

While it is true that earlier losses are better for the score, the observation is that those have the highest scores that have the most wins. _To win is the best strategy for a high score._ In that sense the system is consistent.

One could take different measures, e.g. let w be the number of wins, l the number of losses and d = (w-l), then w/l * d is quite a good measure.