Fide equivalent "slow blitz" rating list

[lkaufman]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I meant your oinion on whether the conversion formula of Lichess blitz 1584 = FIDE 1503 is accurate or not, for humans. Regarding Monchester, I think you are playing the right engine; it seems incredibly bad, but apparently humans playing 5' + 3" blitz on lichess make enough blunders to allow it to perform at 1584, and I'm pretty sure that this would not drop more than a class (200 elo) at 15' + 10". It does a 4 ply search, and humans with such Lichess ratings playing fast can't even do that.

[Carlos777]

About Monchester, I am approx 1650 at blitz in Lichess. Despite playing "awful", it's difficult to me to capitalize my advantage at the end of the game. It doesn't make obvious blunders and manage its time better than me. I played 3'+2" against it, having time control problems. At 15'+10", I wouldn't have any problem with it.

[Odd Gunnar Malin]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I believe that there are no players with rating 500 in lichess.
Here is the rating distribution in lichess.
I see that the minimal rating is 600
https://lichess.org/stat/rating/distribution/rapid

If you talk about fide then the minimal rating is 1000

Aren't we are talkin playing strength. All ratinglist with floor are inacurate near the floor. Until a few years ago we had a floor of 400 (hidden upto 600) in Norway. While I didn't play against 500 players, I played many games against around 800 players and they was clearly much stronger than the Monchester I dowloaded. I would say the Fide list are are inaccurate upto nearly 1400.

[Odd Gunnar Malin]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I meant your oinion on whether the conversion formula of Lichess blitz 1584 = FIDE 1503 is accurate or not, for humans. Regarding Monchester, I think you are playing the right engine; it seems incredibly bad, but apparently humans playing 5' + 3" blitz on lichess make enough blunders to allow it to perform at 1584, and I'm pretty sure that this would not drop more than a class (200 elo) at 15' + 10". It does a 4 ply search, and humans with such Lichess ratings playing fast can't even do that.

I'm not sure. It seems to to vary too much. Look f.ex. at this guy: https://lichess.org/@/tkst98
I know him personally and he is just a lttle bit below me in otb blitz but his Fide rating is 1418. At least for me Blitz and classical are too different thing. To be good in blitz have much to do with how early you started playing. I started late (28) and missing the 'auto pilot' and many of my losses is with queen up. He is maybe 'underrated' but not by 400 points. And I'm sure he would win 10-0 against Monchester.

By the way I started going through Carlos's list and I see what you mean. Comming down to Delimiter and this is the first engine that only had minor bug (inverted scoring when playing black). Most of the above I couldn't find binaries, Giuchess made illegal move when playing black.

[Uri Blass]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I meant your oinion on whether the conversion formula of Lichess blitz 1584 = FIDE 1503 is accurate or not, for humans. Regarding Monchester, I think you are playing the right engine; it seems incredibly bad, but apparently humans playing 5' + 3" blitz on lichess make enough blunders to allow it to perform at 1584, and I'm pretty sure that this would not drop more than a class (200 elo) at 15' + 10". It does a 4 ply search, and humans with such Lichess ratings playing fast can't even do that.

I'm not sure. It seems to to vary too much. Look f.ex. at this guy: https://lichess.org/@/tkst98
I know him personally and he is just a lttle bit below me in otb blitz but his Fide rating is 1418. At least for me Blitz and classical are too different thing. To be good in blitz have much to do with how early you started playing. I started late (28) and missing the 'auto pilot' and many of my losses is with queen up. He is maybe 'underrated' but not by 400 points. And I'm sure he would win 10-0 against Monchester.

By the way I started going through Carlos's list and I see what you mean. Comming down to Delimiter and this is the first engine that only had minor bug (inverted scoring when playing black). Most of the above I couldn't find binaries, Giuchess made illegal move when playing black.

I agree that blitz and long time control are different but I do not see a reason why people who start early should be relatively stronger in blitz.
Of course blitz is also about using premove that is not for OTB games but I see no reason why using it has to do something with starting early.

I guess that being stronger at blitz may be about learning to be fast in finding many good moves in many positions and if your relative advantage is being able to memorize the best move in many chess positions then you should be able to be relatively stronger in blitz.

I guess that some chess course that teach humans to play the right moves in many chess positions without thinking(and not cases when the right move is dependent on a lot of calculations but simply about patterns that you know) may be efficient to teach human players to be better blitz players.

[Uri Blass]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I meant your oinion on whether the conversion formula of Lichess blitz 1584 = FIDE 1503 is accurate or not, for humans. Regarding Monchester, I think you are playing the right engine; it seems incredibly bad, but apparently humans playing 5' + 3" blitz on lichess make enough blunders to allow it to perform at 1584, and I'm pretty sure that this would not drop more than a class (200 elo) at 15' + 10". It does a 4 ply search, and humans with such Lichess ratings playing fast can't even do that.

I am not sure because the rapid rating and the blitz rating is not the same.
I suspect that usually people have higher rapid rating relative to blitz rating(except the top players).

[lkaufman]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I thought of a way to estimate more accurately the FIDE ratings weaker engines would earn at 15' + 10". I went back to Safrad 2.2, it's a more normal engine that gets better with more time just like humans do. I looked at all the Rapid games vs. humans in the last few months, but I looked up the blitz ratings, not the Rapid ratings, for the humans, since we have a way to convert these blitz ratings (on average, not for each person of course) to FIDE equivalent ratings. I set the minimum time limit to 10' + 5" or equivalent, discarded all the wins against players under 1300 Lichess blitz, and limited the number of games allowed for any one opponent. This left 23 games with a performance rating of 1613 on the Lichess blitz scale (but for Rapid games), which converts to 1527 FIDE by the formula, almost a full class below the performance at 5' + 3" blitz, reasonable enough. My testing indicates that at Rapid Safrad should rate about 500 elo above Monchester on the human FIDE scale, which would put Monchester just over 1000, the lowest possible FIDE rating. So roughly speaking 1000 CCRL means about 1500 FIDE Rapid, but the scale is much compressed when going from engine lists to human. Based on my results and this estimation, we need an engine as strong as Gull3 (on one thread) to be evenly matched with Magnus Carlsen in Rapid. I don't knowo how accurate the methodology is, but that's what I get.

[Chessqueen]

Mybe his point was that his score suggest that theire rating is more like below 1400 fide. But I think the sample is too smal.

I think you are much closer to the 1500 FIDE level than Uri is, so what is your own opinion on the matter?

Are we talking about the Monchester Günter linked to?
Try to play against it. If I should guess I would say 500 in 15+10, then I'm nice to it. Maybe it is some other engine that have played the rating up.

I thought of a way to estimate more accurately the FIDE ratings weaker engines would earn at 15' + 10". I went back to Safrad 2.2, it's a more normal engine that gets better with more time just like humans do. I looked at all the Rapid games vs. humans in the last few months, but I looked up the blitz ratings, not the Rapid ratings, for the humans, since we have a way to convert these blitz ratings (on average, not for each person of course) to FIDE equivalent ratings. I set the minimum time limit to 10' + 5" or equivalent, discarded all the wins against players under 1300 Lichess blitz, and limited the number of games allowed for any one opponent. This left 23 games with a performance rating of 1613 on the Lichess blitz scale (but for Rapid games), which converts to 1527 FIDE by the formula, almost a full class below the performance at 5' + 3" blitz, reasonable enough. My testing indicates that at Rapid Safrad should rate about 500 elo above Monchester on the human FIDE scale, which would put Monchester just over 1000, the lowest possible FIDE rating. So roughly speaking 1000 CCRL means about 1500 FIDE Rapid, but the scale is much compressed when going from engine lists to human. Based on my results and this estimation, we need an engine as strong as Gull3 (on one thread) to be evenly matched with Magnus Carlsen in Rapid. I don't knowo how accurate the methodology is, but that's what I get.

Mr. Kaufman, briefly tell me what I did wrong here against this 2200 rated player, beside losing on time?
[pgn][Event "Rated Blitz game"]
[Site "https://lichess.org/7FdYKJld"]
[Date "2022.01.09"]
[Round "?"]
[White "Popemil"]
[Black "ChessLearner1979"]
[Result "1-0"]
[BlackElo "1500"]
[ECO "A00"]
[Opening "Dunst (Sleipner-Heinrichsen-Van Geet) Opening"]
[Variation "1...Nf6 2.Nf3"]
[WhiteElo "2213"]
[TimeControl "180+0"]
[Termination "normal"]
[PlyCount "79"]
[WhiteType "human"]
[BlackType "human"]

1. Nc3 {[%clk 0:03:00]} Nf6 {[%clk 0:03:00]} 2. Nf3 {[%clk 0:02:58]} Nc6?!
{A05 Zukertort Opening} (2. .. d5 3. d4 c5 4. Bf4 Nc6 5. e3 Bg4 6. Be2 e6
7. O-O) 3. d4 {[%clk 0:02:56]} d5 {[%clk 0:02:55]} 4. Bf4 {[%clk 0:02:54]}
a6 {[%clk 0:02:51]} 5. e3 {[%clk 0:02:53]} Bf5 {[%clk 0:02:48]} 6. Bd3
{[%clk 0:02:51]} Qd7? {[%clk 0:02:44]} (6. .. Bg4 7. h3 Bh5 8. g4 Bg6 9.
Ne5 e6 10. Nxg6 hxg6 11. a3) 7. Ne5 {[%clk 0:02:45]} Nxe5 {[%clk 0:02:41]}
8. dxe5 {[%clk 0:02:44]} Ne4 {[%clk 0:02:36]} 9. O-O? {[%clk 0:02:28]} (9.
Bxe4) 9. .. Qc6? {[%clk 0:02:28]} (9. .. O-O-O) 10. Bxe4?! {[%clk 0:02:19]}
(10. Ne2 e6) 10. .. Bxe4?! {[%clk 0:02:25]} (10. .. dxe4 11. Ne2 e6 12. c4
Bg4 13. h3 Bxe2 14. Qxe2 O-O-O 15. Qh5 Rd7 16. Rad1 g6 17. Qh4) 11. f3
{[%clk 0:02:11]} Bg6 {[%clk 0:02:20]} 12. Qxd5 {[%clk 0:02:08]} Qxd5 {[%clk
0:02:18]} 13. Nxd5 {[%clk 0:02:06]} O-O-O {[%clk 0:02:13]} 14. e4 {[%clk
0:02:04]} e6 {[%clk 0:02:07]} 15. Nc3 {[%clk 0:02:01]} Bb4 {[%clk 0:02:01]}
16. Rfd1?! {[%clk 0:02:00]} (16. Nd1 f5 17. exf6 gxf6 18. Nf2 Be7 19. Rfd1
Rdg8 20. Rd3 h5 21. Kf1 Be8 22. b3 e5) 16. .. Bxc3 {[%clk 0:01:58]} 17.
bxc3 {[%clk 0:01:59]} c6?? {[%clk 0:01:45]} (17. .. Rxd1+ 18. Rxd1 Rd8 19.
Rd3 f5 20. exf6 gxf6 21. c4 f5 22. e5 Be8 23. Kf2 Ba4 24. Rc3) 18. Bg5
{[%clk 0:01:57]} Rd7?! {[%clk 0:01:40]} (18. .. Rxd1+ 19. Rxd1 c5 20. Kf2
Re8 21. Bh4 h6 22. Ke3 Bh7 23. f4 h5 24. c4 Bg6 25. h3) 19. Rxd7 {[%clk
0:01:55]} Kxd7 {[%clk 0:01:39]} 20. Rd1+?! {[%clk 0:01:54]} (20. Rb1) 20.
.. Kc8 {[%clk 0:01:29]} 21. c4 {[%clk 0:01:52]} h6?! {[%clk 0:01:27]} (21.
.. c5 22. Kf2) 22. Bh4?! {[%clk 0:01:51]} (22. Be7 Re8 23. Bc5 Rd8 24.
Rxd8+ Kxd8 25. Bf8 f5 26. exf6 gxf6 27. Bxh6 e5 28. c5 Bf7) 22. .. Bh7?!
{[%clk 0:01:23]} (22. .. c5) 23. Be7 {[%clk 0:01:47]} Re8 {[%clk 0:01:18]}
24. Bd6 {[%clk 0:01:44]} Rd8 {[%clk 0:01:14]} 25. c5 {[%clk 0:01:42]} b6?
{[%clk 0:01:13]} (25. .. f6 26. exf6) 26. cxb6 {[%clk 0:01:40]} Kb7 {[%clk
0:01:09]} 27. Rb1 {[%clk 0:01:36]} f5 {[%clk 0:01:01]} 28. exf5 {[%clk
0:01:30]} Bxf5 {[%clk 0:01:00]} 29. Rb2 {[%clk 0:01:28]} Rd7?! {[%clk
0:00:49]} (29. .. c5) 30. c4 {[%clk 0:01:26]} Bd3 {[%clk 0:00:43]} 31. c5
{[%clk 0:01:23]} Bb5 {[%clk 0:00:37]} 32. Rb4 {[%clk 0:01:20]} a5 {[%clk
0:00:34]} 33. Rf4 {[%clk 0:01:18]} Kc8 {[%clk 0:00:17]} 34. a4 {[%clk
0:01:13]} Ba6 {[%clk 0:00:09]} 35. Rf8+ {[%clk 0:01:10]} Rd8 {[%clk
0:00:06]} 36. Rf7 {[%clk 0:01:08]} Rd7 {[%clk 0:00:05]} 37. Rf4 {[%clk
0:01:04]} Rd8 {[%clk 0:00:03]} 38. h4 {[%clk 0:01:03]} Rd7 {[%clk 0:00:02]}
39. h5 {[%clk 0:01:02]} Rd8 {[%clk 0:00:01]} 40. Kf2 {White wins on time.}
1-0[/pgn]

[Odd Gunnar Malin]

Here are some data I collected for Dragon 2.61. In the frustration of all illegal moves I had to trow in Hiarcs and Shredder. Even TSCP have illegal moves.

Book used are the first 50 games in Günter's random 3M book played with switching side.


This give a performance between each Dragon Elo like this:

I'm using only matches where the score is between 20% and 80%. The performance differences can then be calculated with this simple formula:
800*(ScoreHight-ScoreLow)/games

I also did a selftest with 2374 games for each match which gave this result:

The book here are all games in Günter's book (1187*2).

In all list the selected games are marked with green (20-80%).
All games can be found here: http://polarchess.net/Filer/DragonGames.zip

To find engines I used:
https://rwbc-chess.de/download.htm
http://www.computer-chess.org/doku.php? ... nload_list
http://kirr.homeunix.org/chess/engines/ ... ollection/
Carlos777's list above
OliverBr list over bugfree engines for mac and linux
+google of course


It looks like there are some problem between 1600-1700-1800 in the elo setting of Dragon

The hard job with finding engines without problem and for wich level are done so it is easy to play more games against forregn engines to get a more accurate list if it is needed.

[lkaufman]

Here are some data I collected for Dragon 2.61. In the frustration of all illegal moves I had to trow in Hiarcs and Shredder. Even TSCP have illegal moves.

Book used are the first 50 games in Günter's random 3M book played with switching side.


This give a performance between each Dragon Elo like this:

I'm using only matches where the score is between 20% and 80%. The performance differences can then be calculated with this simple formula:
800*(ScoreHight-ScoreLow)/games

I also did a selftest with 2374 games for each match which gave this result:

The book here are all games in Günter's book (1187*2).

In all list the selected games are marked with green (20-80%).
All games can be found here: http://polarchess.net/Filer/DragonGames.zip

To find engines I used:
https://rwbc-chess.de/download.htm
http://www.computer-chess.org/doku.php? ... nload_list
http://kirr.homeunix.org/chess/engines/ ... ollection/
Carlos777's list above
OliverBr list over bugfree engines for mac and linux
+google of course


It looks like there are some problem between 1600-1700-1800 in the elo setting of Dragon

The hard job with finding engines without problem and for wich level are done so it is easy to play more games against forregn engines to get a more accurate list if it is needed.

This is very nice, thank you very much! It seems that the games are 1' + 1", which is fine for the purpose of comparing the levels, but must be kept in mind when we try to estimate human ratings from the data. Regarding the relative ratings of the levels, they are pretty much in line with intentions (engine vs. engine ratings tending to overestimate ratings on the human scale a bit), except that as you indicate, something is wrong around the Skill 17 (elo 1700) level, that level is clearly much closer to 1800 than to 1600, it is somehow unintentionally strong. We'll investigate. If we weaken it a bit everything is more or less reasonably spaced. As for interpreting the ratings in human terms, we have that Safrad 2.2 playing at this 1' + 1" appears to be about Dragon Elo 1550 level. However the Rapid Lichess games of Safrad (at 10' + 5" or slower) worked out to an estimated FIDE rating of only 1527, which means that if Safrad played 1' + 1" while the human played Rapid, it would be much weaker than 1527. On the other hand, its performance in blitz on Lichess worked out to 1922 Human Fide, which means that Dragon Elo 1550 would crush humans rated 1550 at 3 min blitz. At 5' + 3" Safrad performed human FIDE 1723, which might drop to something like 1550 if it had only 1' + 1" while the human had 5' + 3". So my conclusion is that for ratings in the 1500 to 1600 range Dragon Elo ratings play somewhat weaker than the intended 15' + 10" level of that rated human, but roughly like a human of that rating playing 5' + 3" "slow blitz". Most likely this is also roughly true for most of your tested range, let's say from 1200 to 2000 or so, with the exception of the 1700 anomaly, which by accident may actually play close to the intended level of a 1700 human playing Rapid! I don't feel very comfortable basing too much on the LiChess data of one engine, Safrad, but it seems the only really solid Rapid data available to me right now. If this all checks out, maybe we'll just accept that we are simulating slow blitz human play rather than Rapid human play in general; we could of course strengthen all the levels for the next release to correct this, but maybe it's just as well, perhaps most Dragon purchasers who play against the levels play Slow Blitz rather than Rapid, I don't know. I'll study your data more for more conclusions, thanks again!