Amazing results of LeelaKnightOdds-Dev

[lkaufman]

Thank you Mr. Larry Kaufman for your information and teachings. Of course I'm happy to be #2 on the LeelaQueenOdds Top 100 Contenders leaderboard in 3-0 time controls per battle. This situation motivates me and produces happiness; No matter what place I occupy or will occupy in the table, I will continue with the help of the Programmer of my soul in the war to be a better human every day. I have several questions that I would like you to help me resolve, please: 1- According to this table, what Elo Lichess has LeelaQueenOdds?; 2- What Elo Fide would LeelaQueenOdds have? 3- What Elo Lichess and Fide do I have as a result of these tests and game atmosphere? On the other hand, I would like to suggest, if Lichess is willing to implement it, that the system, in all game environments and variables, produce a list of the hundred best human players facing modules; each module, a list, or at least the most famous modules. This table of positions that I suggested should be objective, simply the table would be and reflect the result of the confrontations of man against machine, without caring at all about the person who faces them, be it an amateur, be it the one or the other, world champion of human chess, be it Gary Kasparov, or be it Mr. "Xhuman". This system of measuring the power of the game of man against machines will open the projection and threshold of contenders. This table will be an accessibility motivation to achieve what is accessible by everyone. In addition, an interesting consequence will occur: The level of anti-machine humans will have to be endorsed or remain in the shadows.

1. Near the top, you can see the assumed rating for LeelaQueenOdds at every time control and with each color. So at 3+0, it shows 2247 when it has White and 2197 when it has Black. These numbers will be refined soon with more data.
2. FIDE ratings run about 185 below Lichess blitz ratings for the same players in that range, so 2062 as White and 2012 as Black.
3. Your predicted Lichess blitz rating is shown as 2191 (rounded), which should correspond to about 2006 FIDE.

[Fritz 0]

Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.

Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.

...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.

In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

Good evening Mr. Larry Kaufman. On November 3, 2024, I played a very considerable number of games at a time control level of 3 minutes per game, with LeelaQueenOdds, the result is the following: 95 games in total. LeelaQueenOdds had 33 wins, 20 losses, and 42 draws. Consequently, on November 3, 2024, the computer defeated me 54 to 41.
On November 4, 2024, 3 matches were controlled and timed for 3 minutes, and the result in the final standings is, in my favor, 2 wins and 1 draw. On the other hand, 1 bullet game and the computer and I are equal. My opinion about LeelaQueenOdds, in the current state of the art, is that the LeelaQueenOdds computer has substantially improved its level. Today I ended up exhausted, but good.

So overall your results before the update were about 60% which is +70 elo, and after the update results are about -40 elo for you, so LeelaQueenOdds has improved about 110 elo against you. But you are probably getting better at playing queen odds with experience, so the actual improvement is probably more than this. Maybe 150 elo would be a good guess and reasonably close to our expectations.

I do not know if father improved but the first thing that I think about if you prepare for a better result is to prepare a book to memorize and it seems that father did not try to do it.

I see that he tried a lot of different moves against 1.d4 after winning with d5 in the following game

https://lichess.org/xnNxV6RV

1)a5 draw
https://lichess.org/y7TqLvmB/black

2)Nf6 loss
https://lichess.org/dSMOXY9X/black

3)d6 draw
https://lichess.org/jVeTnMIl/black

I wonder why not to use the following strategy of memorizing the moves that you played and won and do it part of your book when if you do not win simply see what you need to fix in your book based on analysis of the game.

Of course humans have a limited memory but I am sure they can at least memorize the first move.

I added several options on move 1 for all the odds tonite, so this won't be quite so easy to do. All the regulars who play the queen odds bot are seeing results drop by a hundred elo or so since the upgrade on Nov.2.

1)I see that there are several options at move 1 but there are only 21 legal moves for white at move 1 without a queen and memorizing a reply for everyone of them is not so hard.

2)I wonder if you tested Leela with queen odds also against weak chess engines and if there is a similiar elo improvement against weak engines.

Of course I expect it to lose against everything with 2500 CCRL blitz rating list but it may have chances at least against part of the engines below 2000 in the CCRL rating list if not against all of them

https://computerchess.org.uk/ccrl/404/

Yes, I did test Leela at queen odds vs weak/crippled engines, and got similar (somewhat larger) elo gains. The best match for the current LeelaQueenOdds using Komodo 14.1 skill levels is level 15 with -100 Contempt to seek trades, pretty even match. Not sure where that would fit on the CCRL list, either blitz or Rapid, but it is a more suitable opponent than a random engine that may not know to simplify when ahead.

It sounds a little bit strange. I've just run a 4000+ game match between Komodo 14.1 level 15 and Dragon 3.3 Elo 1800 and Komodo 14.1 level 15 lost by 115 Elo. Since Dragon 3.3 Elo settings are made for 10'+5'', is it possible that LeelaQueenOdds is just about 1685 at fast Rapid? It would mean that it's below 1500 at Classical, which seems too low to me. Am I missing something here?

[Uri Blass]

Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.

Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.

...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.

In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

Good evening Mr. Larry Kaufman. On November 3, 2024, I played a very considerable number of games at a time control level of 3 minutes per game, with LeelaQueenOdds, the result is the following: 95 games in total. LeelaQueenOdds had 33 wins, 20 losses, and 42 draws. Consequently, on November 3, 2024, the computer defeated me 54 to 41.
On November 4, 2024, 3 matches were controlled and timed for 3 minutes, and the result in the final standings is, in my favor, 2 wins and 1 draw. On the other hand, 1 bullet game and the computer and I are equal. My opinion about LeelaQueenOdds, in the current state of the art, is that the LeelaQueenOdds computer has substantially improved its level. Today I ended up exhausted, but good.

So overall your results before the update were about 60% which is +70 elo, and after the update results are about -40 elo for you, so LeelaQueenOdds has improved about 110 elo against you. But you are probably getting better at playing queen odds with experience, so the actual improvement is probably more than this. Maybe 150 elo would be a good guess and reasonably close to our expectations.

...for my taste, it would be very emblematic, if at the blitz game levels without game increase, 5-0 and 3-0, LeelaQueenOdds acquires a GM level of game and performance.

Mr. Larry Kaufman... I would like to give you an idea for your study and analysis, although you consider it pertinent... if you observe my anti-machine thinking system, you will conclude that it comes from strategic resources of David against Goliath, and yes it has given time-tested results. Likewise, if LeelaQueenOdds acts in the manner of a block symphony with anti-GM bias, in my opinion the battlefront will replace the deficiencies in supplies and inputs attributable to the queen.

You will be very glad to hear that a Leaderboard system is now online for LeelaQueenOdds at https://marcogio9.github.io/LeelaQueenOdds-Leaderboard/ . You are number 2 on the list now! It is still being refined and I'm not sure when it is being updated, but it gives ratings based on what rating you should achieve on LiChess blitz playing 3'2" games, based on your queen odds games! It considers what time limits you play at, what color you play, and your results. It should be reasonably accurate once there are enough games with the upgraded bot.

I played 4 games at time control 3+2 against leelaqueen odds and won 2.5-1.5
I wonder if there is a minimal number of games to be included in the list.

Here is my last game

https://lichess.org/BKObn8YT

Comments:
1)I knew that 1.d4 d5 2.Nf3 Nf6 3.c3 c5 is theory in regular chess so I was not afraid to play this line when white has not a queen.
2)4...cxd4 is the simplest trading pawns.
3)I played 6...Bg4 to trade the bishop.
I saw that if 7.Nh4 then g5 and 7.Ng1 is passive so lc0 decided to allow me to trade the bishop(not sure if Ng1 was practically better because being passive is also a problem.

4)I knew that 12...Ne4 lose a pawn but I think that it is a good deal to lose 4 against 3 in this game(I use 1.3,5,9 values to make things simple).
Before the trade my material value was 35 against 26 of white and the ratio is 35/26 when after the trade I have 31 against 23
I did not calculate it during the game but 31/23=1.3478...>35/26=1.34615... so I basically improved my material ratio even if I consider bishop to be equal to a knight and I consider the bishop to be stronger.

5)17...e5 was a good move to get more trades and I prepared it earlier.

6)21...Bxc3 was another good move.

material ratio before Bxc3 was 27/19=1.421...
After the game line 21...Bxc3 22.bxc3 Rxe1+ Rxe1 Qxc3 material ratio is 19/12=1.5833...
white could play 22.Rec1 that I expected but after Qxf3 gxf3 Bxb2 Rab1 Bxc1 Rxc1 material ratio is even worse 15/9=1.6666...

I did not plan Qxf3 but also the line 22...Bxb2 Rxc6 bxc6 is good enough with material ratio 18/12=1.5 that is better than 1.421

7)I considered 24...Qxe1+ that is also winning but no reason to reduce material ratio from 19/12 to 10/7 that is smaller than 1.5

8)I considered 28...Qxe3 as the simplest way to win.
material ratio was 19/11=1.7272.. before this move and only 10/6=1.6666... after the move but I already saw that white has nothing to do to stop the a pawn because the king is not close.

9)33...a1=Q is of course a safe win for black but I already saw that I can safely get a queen and prevent white to sacrifice the knight for it

10)37.Na1 is practically a bad move and 37.Nd4 was better and maybe give some practical chances against significantly weaker players than me but I am sure I could win also against 37.Nd4

You are on the list, number 22 at 1692. You start at 1600 and move gradually away from that, so it does take a lot of games to get to anywhere near 2000.

It seems that it does not update the games immediately because after 2 more games when I got 1.5 out of 2 in 3+2 my rating is still 1692.61 and I am number 22.

[Uri Blass]

In the last game I do not understand lc0's sacrifice and if they are the best practical chance.

https://lichess.org/jLtE0wZR/black#27

Mistakes of lc0.
1)5.d4 that not only lose a pawn but also allow me to trade knights.
White get some initiative but basically I am not afraid of initiative and if I do not something specific I take material.

2)14.Bxc7 lose material

3)16.Nxf7 another move that lose material so I was happy to sacrifice my queen for a rook and even a blunder of me by 21...Nf4 did not prevent me winning easily.

Edit:I see finally that my rating improved based on the last 2 games to 1746.98 and place 14.

[lkaufman]

Mr. Larry Kaufman.
This performance in the fight of man against machine with odds is really wonderful.
I consider that in man-machine duels with odds, that in some way, there are real probabilities on the part of the human being to acquire progress in shortening the distances.
If at any time you wish or consider that it is useful for you to participate in man-versus-machine duels, I would be attentive to an invitation, in exchange for the sole pleasure of playing chess, without a prize of any kind.

Thank you. Your games have been helpful in highlighting the issue of allowing early repetition draws, we're working on ways to reduce them. If you want to know what I would consider most useful, I would like to see you play games with LeelaQueenForKnight, at whatever time limit allows you to win a decent percentage of the games. It's most helpful when players pick a handicap that is a bit difficult for them but where the win/loss ratio is not too one-sided. My guess is that you'll need a fairly slow blitz or even Rapid TC to do this; it's WAY harder than actual queen odds, though easier for the human than rook odds.

...excellent odds!! I have stated, ( 20 games ) I believe that as human beings playing against odds, we have the possibility of learning more about the odds and thus increase the probability of improving the results... I have a question; What could be the estimated elo of this odds? Thank you Mr. Larry Kaufman.

In LiChess blitz games that bot performs around 2400 Lichess level, which is considered to be equivalent to 2235 FIDE rating. Of course it would be lower at longer time controls, but you are playing blitz, so the above is appropriate for you.

Good evening Mr. Larry Kaufman. On November 3, 2024, I played a very considerable number of games at a time control level of 3 minutes per game, with LeelaQueenOdds, the result is the following: 95 games in total. LeelaQueenOdds had 33 wins, 20 losses, and 42 draws. Consequently, on November 3, 2024, the computer defeated me 54 to 41.
On November 4, 2024, 3 matches were controlled and timed for 3 minutes, and the result in the final standings is, in my favor, 2 wins and 1 draw. On the other hand, 1 bullet game and the computer and I are equal. My opinion about LeelaQueenOdds, in the current state of the art, is that the LeelaQueenOdds computer has substantially improved its level. Today I ended up exhausted, but good.

So overall your results before the update were about 60% which is +70 elo, and after the update results are about -40 elo for you, so LeelaQueenOdds has improved about 110 elo against you. But you are probably getting better at playing queen odds with experience, so the actual improvement is probably more than this. Maybe 150 elo would be a good guess and reasonably close to our expectations.

I do not know if father improved but the first thing that I think about if you prepare for a better result is to prepare a book to memorize and it seems that father did not try to do it.

I see that he tried a lot of different moves against 1.d4 after winning with d5 in the following game

https://lichess.org/xnNxV6RV

1)a5 draw
https://lichess.org/y7TqLvmB/black

2)Nf6 loss
https://lichess.org/dSMOXY9X/black

3)d6 draw
https://lichess.org/jVeTnMIl/black

I wonder why not to use the following strategy of memorizing the moves that you played and won and do it part of your book when if you do not win simply see what you need to fix in your book based on analysis of the game.

Of course humans have a limited memory but I am sure they can at least memorize the first move.

I added several options on move 1 for all the odds tonite, so this won't be quite so easy to do. All the regulars who play the queen odds bot are seeing results drop by a hundred elo or so since the upgrade on Nov.2.

1)I see that there are several options at move 1 but there are only 21 legal moves for white at move 1 without a queen and memorizing a reply for everyone of them is not so hard.

2)I wonder if you tested Leela with queen odds also against weak chess engines and if there is a similiar elo improvement against weak engines.

Of course I expect it to lose against everything with 2500 CCRL blitz rating list but it may have chances at least against part of the engines below 2000 in the CCRL rating list if not against all of them

https://computerchess.org.uk/ccrl/404/

Yes, I did test Leela at queen odds vs weak/crippled engines, and got similar (somewhat larger) elo gains. The best match for the current LeelaQueenOdds using Komodo 14.1 skill levels is level 15 with -100 Contempt to seek trades, pretty even match. Not sure where that would fit on the CCRL list, either blitz or Rapid, but it is a more suitable opponent than a random engine that may not know to simplify when ahead.

It sounds a little bit strange. I've just run a 4000+ game match between Komodo 14.1 level 15 and Dragon 3.3 Elo 1800 and Komodo 14.1 level 15 lost by 115 Elo. Since Dragon 3.3 Elo settings are made for 10'+5'', is it possible that LeelaQueenOdds is just about 1685 at fast Rapid? It would mean that it's below 1500 at Classical, which seems too low to me. Am I missing something here?

According to the website, LeelaQueenOdds as White is rated 1955 at 15'10" (not "fast rapid", but perhaps the right tc or close enough). Using the converter tables https://chessgoals.com/rating-compariso ... sschesscom, that looks like about 1687 chess.com Rapid, which was the standard for the Komodo Dragon settings, and fairly close to FIDE (which would be 1733 FIDE by these tables). So no big disparity, other than the question of just what time control in Rapid best fits the Komodo Dragon ratings. A faster tc would raise the estimate, but I also think that we are currently overestimating the rating gain from the upgrade, so maybe everything will agree pretty well soon. The main point is that Lichess ratings run higher than chess.com ratings around the 2000 level.

[Father]

Thank you Mr. Larry Kaufman for your information and teachings. Of course I'm happy to be #2 on the LeelaQueenOdds Top 100 Contenders leaderboard in 3-0 time controls per battle. This situation motivates me and produces happiness; No matter what place I occupy or will occupy in the table, I will continue with the help of the Programmer of my soul in the war to be a better human every day. I have several questions that I would like you to help me resolve, please: 1- According to this table, what Elo Lichess has LeelaQueenOdds?; 2- What Elo Fide would LeelaQueenOdds have? 3- What Elo Lichess and Fide do I have as a result of these tests and game atmosphere? On the other hand, I would like to suggest, if Lichess is willing to implement it, that the system, in all game environments and variables, produce a list of the hundred best human players facing modules; each module, a list, or at least the most famous modules. This table of positions that I suggested should be objective, simply the table would be and reflect the result of the confrontations of man against machine, without caring at all about the person who faces them, be it an amateur, be it the one or the other, world champion of human chess, be it Gary Kasparov, or be it Mr. "Xhuman". This system of measuring the power of the game of man against machines will open the projection and threshold of contenders. This table will be an accessibility motivation to achieve what is accessible by everyone. In addition, an interesting consequence will occur: The level of anti-machine humans will have to be endorsed or remain in the shadows.

1. Near the top, you can see the assumed rating for LeelaQueenOdds at every time control and with each color. So at 3+0, it shows 2247 when it has White and 2197 when it has Black. These numbers will be refined soon with more data.
2. FIDE ratings run about 185 below Lichess blitz ratings for the same players in that range, so 2062 as White and 2012 as Black.
3. Your predicted Lichess blitz rating is shown as 2191 (rounded), which should correspond to about 2006 FIDE.

Thank you very much Mr. Larry Kaufman for your message. I want to express to you that today has been a very happy day for me, as a result of the dynamics that your Post about the Odds has generated. For the first time in my life, I am observing that a door is opening that leads to a path of equity and a new world, in the markings of the power of the arm-wrestling game between humans against chess machines. I myself have been a victim for many years, more than thirty-eight years in total, since my confrontations against Chess Challenger Level IV 1986 that served as inspiration for me to sow the seeds of the purest anti-machine game system. It was then in the year 2002, when in the first half of the year, in person to machine, I achieved the best results of men against machines in my country. I then beat GM, IM and experts, against Fritz 7. After that, I then beat Fritz 8 in weeb by sacrificing my queen with white pieces on move 5 of the match, but it seemed then that very few cared. So then, I have continued alone fighting against windmills, and since 2017, against AI Wind Whirlwinds. But I have never given up. And today, thank you, Mr. Larry Kaufman, I have seen a little light at the end of the tunnel... I don't care if I am the one who enjoys the joys of success or not, on the other hand, it makes me happy that success is democratized to the combat guidelines. Opening the Odds to the general public will catapult the science game. It will happen in terms of analogues, a parallel to that of Bitcoin against material circulating money. The tables of the Top 100 on computers in their standard form as Odds, is light. It will be a fight of the elos against the egos. The true and greatest battle of man against machines. Those who do not participate in the war will not be able to wear the uniform of new warriors.

[Uri Blass]

I do not like games with no increment because I like to think in chess so I used only 3+2.
I won 2 additional games so the score now is 6-2 for(won 5 games drew 2 and lost 1 game).

I believe that 2149.56 of LeelaQueenOdds as white 3+2 (Reference) is too high and I believe that humans with this rating should be clearly stronger than LeelaQueenOdds as white 3+2 if they play correctly and are ready for bad trades in order to simplify.

[Marcus91]

Here’s how the leaderboard works: Each player starts with a rating of 1600. The Elo rating is adjusted after every game similarly to how FIDE calculates Elo. The 'K' factor is set to 40 for the first 20 games, then to 20 until reaching 100 games, and finally to 10 for all games after that. More data will improve the accuracy of the LeelaQueenOdds ratings and adjustments for different time controls. Currently, the leaderboard is not running on a dedicated server, so updates are sporadic, but this will be resolved soon

[Father]

I do not like games with no increment because I like to think in chess so I used only 3+2.
I won 2 additional games so the score now is 6-2 for(won 5 games drew 2 and lost 1 game).

I believe that 2149.56 of LeelaQueenOdds as white 3+2 (Reference) is too high and I believe that humans with this rating should be clearly stronger than LeelaQueenOdds as white 3+2 if they play correctly and are ready for bad trades in order to simplify.

Congratulations Uri Blass. In Good Time!! Welcome to the battlefield!! You are Welcome to the "Promised Land." Now you can fight to reach your fair position in the standings. If you or someone else displaces me from my second place then I will be happy to share with new fighters, I will be obliged to play again, I will return to war after 40 years of pilgrimage through the certainty of life, where I never lacked the protection of "my Programmer" and to the point that in extreme situations I was able to dine on the rich "Manna." In case, Mr. Uri Blass, the "LeelaQueenOdds" dinner would be insufficient for you, there are other delicious dishes on the menu... "LeelanightOdds" is just an example!! This last breakfast I am referring to is delicious, but it kicks us hard against our brain. If the horseback steak breakfast would be then still "light", then we are waiting for the whole horse to run in the Derby, that is, "Leela in its Standard and full mode." Here that computer goes from being a horse to being a lion. But even so, I will battle with my shield and my sword, behind the wall, and if necessary, I will go out and jump over the wall seeking to achieve my dreams, and freedom. Chains cannot hold us free spirits, we only know war to achieve it, we love the indomitable spirit. Free will is our north, determination is part of our path, the prize, and high heaven, our goal. The position tables will then be the maximum marker, where our position will be written higher and firmer than in other media with tinted ink.

[Uri Blass]

I do not like games with no increment because I like to think in chess so I used only 3+2.
I won 2 additional games so the score now is 6-2 for(won 5 games drew 2 and lost 1 game).

I believe that 2149.56 of LeelaQueenOdds as white 3+2 (Reference) is too high and I believe that humans with this rating should be clearly stronger than LeelaQueenOdds as white 3+2 if they play correctly and are ready for bad trades in order to simplify.

Congratulations Uri Blass. In Good Time!! Welcome to the battlefield!! You are Welcome to the "Promised Land." Now you can fight to reach your fair position in the standings. If you or someone else displaces me from my second place then I will be happy to share with new fighters, I will be obliged to play again, I will return to war after 40 years of pilgrimage through the certainty of life, where I never lacked the protection of "my Programmer" and to the point that in extreme situations I was able to dine on the rich "Manna." In case, Mr. Uri Blass, the "LeelaQueenOdds" dinner would be insufficient for you, there are other delicious dishes on the menu... "LeelanightOdds" is just an example!! This last breakfast I am referring to is delicious, but it kicks us hard against our brain. If the horseback steak breakfast would be then still "light", then we are waiting for the whole horse to run in the Derby, that is, "Leela in its Standard and full mode." Here that computer goes from being a horse to being a lion. But even so, I will battle with my shield and my sword, behind the wall, and if necessary, I will go out and jump over the wall seeking to achieve my dreams, and freedom. Chains cannot hold us free spirits, we only know war to achieve it, we love the indomitable spirit. Free will is our north, determination is part of our path, the prize, and high heaven, our goal. The position tables will then be the maximum marker, where our position will be written higher and firmer than in other media with tinted ink.

I played additional 10 games.

Relatively a bad day because I won only 5.5-4.5.

Note that I take advantage of the fact that the machine is bluffing.

I mean that Leela make sometimes bad sacrifices so I do not believe lc0's sacrifices but calculate so there are cases when I think for a move more than 30 seconds(with 2 seconds increment it is not a risk that is too big).

I also play sacrifices that lose material but in my case they are for good reason to simplify.

Note that place in the standing does not measure ability unless the players played many games and 20 games are clearly not enough.