Fool me once, shame on you, fool me twice, shame on me! ;-)

;-)

I've always like the Klingon aphorism:
Fool me once, shame on you. Fool me twice, prepare to die.

Once is happenstance. Twice is coincidence. The third time it's enemy action. This is twice you have meddled in my affairs, Mr. Bond...

"Yeah, I wasn't going to say anything until later this week, but after some long talks with Ghost, we've decided that due to all the changes in the website over the years it would be best to start everyone's ratings over as of Jan 1 2012."

Something similar to this would actually make a lot of sense. A category basing GR calculations specifically on GR in that month, or collected specifically within a year, would be very interesting and create more of a challenge and dynamic - as opposed to a system that balances in favour of older players who will have been on the website longer and therefore had more time to collect GR. You keep the 'All-Time' List for GR calcuated since the 'beginning' and perhaps have a new category that resets GR from each year - so we could get an accurate picture of gameplay in that year, rather than it being skewed with players who have amassed a great wealth of GR from many years.

Actually you could make it a "moving average" type, meaning each month you count the last 12 months only, resulting old games will disappear automatically.

Or, you could use ELO as intended, which involves keeping track of an uncertainty that makes sense along with the current system rather than inventing a new one that doesn't...

CS, I'm still confused by what you mean in saying that Elo ratings have two parameters. I understand what you want to achieve with the 2nd parameter but this isn't something that exists currently with Elo ratings.

Rolling averages don't work as each months GR is based upon the previous month. And long time.players don't have an advantage as they have to keep playing to not fall off and, in so doing, they are continually risking more and more.points as they play. In fact, the new guys get a boost by playing higher ranked GR if they win without risking as much to play. A rolling GR would require running every months GR for the entire time frame every time.it was calculated.

A reset would, however, Put everyone on the same playing field starting fresh once in a while. But if a new player is that good, they will quickly rise to the top ranks as is evidenced by the number of top 100 players who have been here less than a year.

Rolling 1 year ghost ratings would actually work but that wouldn't change much as older results are less relevant for your rating than newer ones, anyways.

Resetting the GR would highly discriminate (good) non-live players with a small number of concurrent games, as it probably takes them half a year or more to reach the number of games necessary to get them anywhere near their native rating while live players can achieve that within a month.

Good point Mr. A... you think they could do a "last 100 games" thing?

I don't think restarting the ratings is a good idea.

Just let inflation take care of the problem.

Sure, but again, if you played 300 games, 99% of your score are determined by the last 100 games. The first 200 results don't contribute much to your current score, so it shouldn't be worth it.

GR and ELO aren't bad rating systems for non-tournament situations. The only obvious 'improvement' I could think of is incorporating the standard deviation as CS suggested. Like in Microsoft's TrueSkill.

Personally I'd suggest to weight live games lower than non-live games, but that's mainly because I seem to fail at live games. :)

For the record, I'm not actually planning on a reset.

@Draugnar, actually the calculations are done from scratch each time. The old pearl script took 2 minutes to chug through 7 years worth of games, and longer each month you add on. My new java code takes about 10 seconds.

Somebody get rid of Grand Duke Feodor, his meteoric rise up the table despite being banned for the last 3 months is causing some consternation :-)

And remember that GR affects everyone in the game, so someone may find their game 100 in the past is another's game 300 in the past. GR has to be time frame based unless live games are separated from non-live and even then, a disparity occurs when comparing people who play one game at a time versus those who play 3 or 4 and those who play a dozen.or more.

CS, I think your envisioned system is something like the Glicko system (http://www.glicko.net/glicko.html).

Unfortunately for non-live games, the sample size is simply too small for most people for there to be some sort of reasonable convergence. Therefore, the best we can do - imho - is to use a rudimentary rewards/punishment system like Elo ratings.