Yes, I agree. I think that’s the point I’m doing a very bad job of making actually. It’s very easy to justify how well the EGF system works - clearly a huge amount of time has been put into making it reliable, and continues to be put in to maintaining the quality of its data.
I just happen to think that there are other systems that have evolved differently and are equally reliable – within their sphere of responsibility. It was always a generally held belief (with a fair amount of data to support it) that a European dan going to America was on average about 1 dan rank higher in the AGA system. If people on OGS end up with a stable but different rank to EGF or AGA or anywhere else, I don’t think this contributes towards evidence that the ranks are unreliable, just anchored differently (and perhaps that’s the wrong word to use, but hopefully the meaning is clear).
I will re-iterate though that it’s beginning to look like the rolling rating system might be a mistake, despite the increased accuracy that I suspect it offers. The more recent discussions seem to imply that accuracy and reliability of the final number is not the only key consideration of a ranking system, particularly online!
Is this now the rule that my rating does not go down/up immediate when I lose/win but with a delay of several hours? If so its a change to the worse and should be rolled back
I think your profile page won’t always show instant updates to the ratings, there is some caching/delays involved. After the fact it’s hard to know what you saw because the ratings graph seems to include 1 point per day. Anyways in the screenshot now you can see your rating went down on June 29 on the basis of 14 rated games.
7 wins, 7 losses, but 12 games were vs weaker and 2 vs stronger, so overall 50% is not good enough.
ETA S_Alexandar’s link gives a lot more info, seems you did go down overall after those 3 wins in a row.
ok, thank you. Has the same information as the chart so information should be correct. I find this new system misleading. Might be that rating curve is now more smoothed but just updating the rating after every game with ELO formula would be preferable IMO. Now I am playing some kind of 15 round tournament without knowing it…
I have no idea, but I’d guess to gain a place in a marathon you have to run faster than the faster players, it’s established you can outrun the ones behind you.
So is it possible to have a scenario where you win 15 games in a row against weaker players, but go down in rank after each game because you slowly erase a previous 15 game win streak against stronger players?
I think if you win 15 games in a row rating will always go up. But suppose you lost 15 games against 10k and then start winning games against 9k. Rating will go down until you won perhaps 9-10 games in a row.
Edit: ok it’s not that simple because you need some wins against stronger players drop out of the sliding window but just shows how complicated this system is.
Not so much a fan of that solution as this proposed one: since each game is gonna be calculated 15 times, why not make an estimate that essentially calculates those 15 periods in advance, but only for display and matchmaking. Since game updates are calculated based on your rating, RD, and volatility from the time it was being played at, the adjustment of RD and volatility between ratings period shouldn’t matter, and any updates from future games should only affect your rating in the expected manner.
Of course, if all ratings are being calculated based on the ratings period it’s in, you can, for display, fill the rest of the period with “virtual games” that may affect volatility and RD, but not the rating shown.
I think the remaining issue is a windowing artifact. by adding a new game you are losing the oldest game in your window. if it was a worse loss than the new loss, then the total will go up by the difference.
I believe weighting the games by age in a more tapered fashion would dampen this out a bit.
not really… the “ratings period” is a period of games that updates your rating, when a game leaves your ratings period its effects don’t suddenly disappear, it’s just not creating new effects. The total does not go up the difference, it goes down by the effect of the new loss, plus whatever effects are in the other games within the period
