Navara vs Leela Knight Odds Match

[Chris Formula]

looks like queen-based endgame is the key to beating Leela on knight odds game?

[AdminX]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

[lkaufman]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

[Chessqueen]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

Does your Dragon has a good chance of drawing against GM Navara Knight Odds at time control of 3'+2"

Anyway I believe it is more interesting to see a match of Rook Odds versus a Master human player
[pgn][Event "Computer chess game"]
[Site "DESKTOP-4QNC0GS"]
[Date "2024.03.30"]
[Round "?"]
[White "Stockfish-windows-x86-64-avx2"]
[Black "Kingfisher.v1.1.1"]
[Result "1/2-1/2"]
[BlackElo "2200"]
[Time "11:47:02"]
[WhiteElo "3645"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1"]
[Termination "normal"]
[PlyCount "36"]
[WhiteType "program"]
[BlackType "program"]

1. e4 {(e2-e4 d7-d5 e4-e5 Bc8-f5 g2-g4 Bf5-d7 d2-d4 c7-c5 c2-c4 Bd7-c6
Bc1-g5 c5xd4 Qd1xd4) -4.21/27 10} Nc6 {(Nb8-c6 d2-d4 d7-d5 e4-e5 f7-f6
Nb1-c3 f6xe5 d4xe5 d5-d4 Nc3-e4 Bc8-f5 Ne4-c5 Nc6xe5 Bf1-b5+ c7-c6 Nc5xb7)
+5.64/16 12} 2. d4 {(d2-d4 e7-e6 d4-d5 e6xd5 e4xd5 Bf8-d6 Ng1-f3 Ng8-f6
d5xc6 d7xc6 Bf1-e2 O-O O-O h7-h6 Nb1-c3 Rf8-e8 Be2-d3 a7-a5 a2-a3 Bc8-d7
h2-h3 a5-a4) -3.89/19 0} d5 {(d7-d5) +5.68/15 2} 3. Nc3 {(Nb1-c3 e7-e5
Nc3xd5 Ng8-f6 Bf1-c4 Nc6xd4 c2-c3 Nf6xd5 e4xd5 Nd4-f5 Bc4-b3 Bf8-e7 h2-h4
O-O Ng1-f3 e5-e4 Nf3-g5 Nf5-h6 O-O Bc8-f5 Bc1-e3 Rf8-e8 Rf1-e1 Be7-d6
Qd1-c2) -4.14/29 13} Nf6 {(Ng8-f6 e4-e5 Nf6-e4 Ng1-f3 e7-e6 Bf1-d3 Bf8-b4
Bc1-d2 Bb4xc3 b2xc3 f7-f6 O-O Ne4xd2 Qd1xd2 Bc8-d7) +5.86/17 12} 4. e5
{(e4-e5 Nf6-g8 Bf1-b5 e7-e6 Ng1-f3 Ng8-e7 h2-h4 a7-a6 Bb5-d3 Nc6-b4 h4-h5
Nb4xd3+ Qd1xd3 Bc8-d7 g2-g4 Qd8-c8 h5-h6 g7xh6 g4-g5) -3.70/22 2} Ne4
{(Nf6-e4 Ng1-e2 f7-f6 Ne2-f4 Ne4xc3 b2xc3 g7-g6 e5xf6 e7xf6 Qd1-f3 Nc6-e7
Bf1-d3 c7-c6 Bc1-a3 Bf8-g7 Ke1-d2 O-O) +5.91/17 12} 5. Nxe4 {(Nc3xe4 d5xe4
c2-c3 Bc8-f5 Bf1-b5 a7-a6 Bb5xc6+ b7xc6 h2-h4 h7-h5 Qd1-a4 Ra8-b8 Ng1-e2
Bf5-d7 Ne2-f4 e7-e6) -3.59/23 4} dxe4 {(d5xe4 Bf1-b5 Bc8-d7 Bb5-c4 g7-g6
Ng1-e2 Bd7-f5 c2-c3 Nc6-a5 Bc4-b3 Bf8-g7 Bc1-f4 O-O b7-c6) +5.87/16 12} 6.
c3 {(c2-c3 Bc8-f5 Ng1-e2 e7-e6 Ne2-g3 Bf5-g6 Bf1-b5 a7-a6 Bb5-a4 b7-b5
Ba4-b3 h7-h5 O-O h5-h4 Ng3-e2 Nc6-a5 Ne2-f4 Bg6-f5 Bb3-c2 c7-c6 g2-g4
h4xg3/ep f2xg3) -3.55/27 10} Be6 {(Bc8-e6 Bf1-b5 a7-a6 Bb5xc6+ b7xc6 Qd1-c2
f7-f6 Ng1-e2 f6xe5 Qc2xe4 Qd8-d5 Qe4xe5 Qd5xe5 d4xe5 Be6xa2 Ke1-d1)
+5.85/15 13} 7. Ne2 {(Ng1-e2 Be6xa2 Qd1-c2 Ba2-c4 Qc2-a4 Bc4-d5 Ne2-f4
e7-e6 Bf1-e2 Bf8-e7 O-O O-O Rf1-d1 a7-a6 Qa4-c2 b7-b5 b2-b3 Kg8-h8 h2-h4
Be7xh4 c3-c4 Nc6-b4 Qc2-b1 Bd5-b7 d4-d5 e6xd5 g2-g3 Bh4-g5 Nf4xd5) -3.67/31
31} Bxa2 {(Be6xa2 Qd1-c2 Ba2-c4 Ne2-f4 Bc4xf1 Rh1xf1 g7-g5 Nf4-e2 Qd8-d5
Bc1xg5 Rh8-g8 Ne2-f4 Nc6xd4 c3xd4 Qd5xd4 Bg5-h4) +5.84/16 13} 8. Nf4
{(Ne2-f4 Qd8-d7 Bf1-e2 Nc6xe5 d4xe5 Qd7xd1+ Be2xd1 g7-g6 e5-e6 O-O-O h2-h4
f7-f5 Bd1-e2 Bf8-g7 c3-c4 b7-b5 c4xb5 Rd8-d6 h4-h5 Ba2xe6 b5-b6 a7xb6 h5xg6
h7xg6 Rh1xh8+ Bg7xh8) -3.73/26 7} a6 {(a7-a6 Qd1-a4 Ba2-b1 Qa4-b3 e7-e6
Bf1-e2 g7-g5 Nf4-h5 Bf8-e7 Bc1-e3 b7-b5 O-O Bb1-d3 Qb3-d1 Bd3xe2 Qd1xe2
c7xc6) +6.25/16 13} 9. e6 {(e5-e6 Ba2xe6 Nf4xe6 f7xe6 Bf1-c4 Qd8-d6 O-O
e6-e5 d4-d5 Nc6-d8 Bc4-a2 g7-g6 c3-c4 Qd6-d7 Rf1-e1 Nd8-f7 Re1xe4 Bf8-g7
h2-h4 c7-c6 h4-h5 c6xd5 c4xd5 Nf7-d6 Re4-g4 Ra8-c8 Bc1-g5 Rh8-f8 Bg5-e3
Rf8-f6 Rg4-b4 b7-b5) -3.45/27 11} fxe6 {(f7xe6 b2-b3 e6-e5 Nf4-e6 Qd8-d6
d4-d5 Nc6-a5 Bc1-a3 Qd6xa3 Ne6xc7+ Ke8-d8 Nc7xa8 Qa3xb3 Qd1xb3 Na5xb3)
+6.90/15 13} 10. b3 {(b2-b3 Qd8-d6 Bf1-c4 b7-b5 Bc4xe6 g7-g6 d4-d5 Bf8-h6
g2-g3 Nc6-d8 O-O Rh8-f8 Be6-h3 Bh6xf4 Bc1xf4 Rf8xf4 g3xf4 Qd6xf4 Qd1-c2
Nd8-f7 Qc2xa2 Nf7-e5 Bh3-g2 Ne5-g4 Rf1-e1 e4-e3 f2-f3 Qf4xh2+ Kg1-f1 Qh2-f4
Qa2-e2 Ra8-d8 Bg2-h3 Ng4-f6 c3-c4 b5xc4 b3xc4 Qf4-h4 Kf1-g2 Qh4-g5+ Kg2-h1)
-3.80/25 12} e5 {(e6-e5 Qd1-h5+ g7-g6 Nf4xg6 h7xg6 Qh5xh8 Ba2xb3 Bc1-h6
Ke8-f7 Qh8-h7+ Kf7-f6 d4xe5+ Nc6xe5 Bh6-f4 e7-e6) +7.17/15 13} 11. Qh5+
{(Qd1-h5+ g7-g6 Nf4xg6 h7xg6 Qh5xg6+ Ke8-d7 d4-d5 Kd7-c8 d5xc6 Ba2xb3
Bf1-e2 b7xc6 Bc1-e3 e7-e6 h2-h4 Kc8-b7 Qg6xe4 a6-a5 Qe4-b1 a5-a4 h4-h5
Qd8-d5 O-O Rh8-g8 g2-g4 a4-a3 c3-c4 Rg8xg4+ Be2xg4 Qd5xc4 Bg4-f3 a3-a2)
-3.68/27 11} g6 {(g7-g6 Nf4xg6 h7xg6 Qh5xh8 Ba2xb3 Bc1-h6 Ke8-f7 Qh8-h7+
Kf7-f6 Qh7-h8+ Kf6-f7) +7.03/21 13} 12. Nxg6 {(Nf4xg6 h7xg6 Qh5xg6+ Ke8-d7
d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 c3-c4 Qd8-d6 Qg6-g4+ Qd6-d7 Qg4-g3
Bb3-c2 O-O e7-e6 h2-h4 a6-a5 Bc1-e3 Bf8-b4 h4-h5 Kc8-b7 Qg3xe5 Bc2-d3
Rf1-d1 Rh8-g8 h5-h6) -3.93/29 16} hxg6 {(h7xg6 Qh5xh8 Ba2xb3) +7.03/20 13}
13. Qxg6+ {(Qh5xg6+ Ke8-d7 d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 O-O
Rh8-g8 Qg6-h5 e7-e6 Bc1-e3 Bf8-d6 c3-c4 Qd8-e8 Qh5-h3 Qe8-g6 g2-g3 a6-a5
Rf1-a1 c6-c5 Qh3-g2 a5-a4 Ra1-a3 c7-c6 Qg2-f1 Kc8-c7 Qf1-a1 Rg8-b8)
-3.92/26 4} Kd7 {(Ke8-d7 d4-d5 Nc6-a5 Qg6-e6+ Kd7-e8 Qe6xe5 Rh8-h4 Bc1-g5
Rh4-h7 Qe5xe4 Rh7-g7 Bg5-h6 Ba2xb3 Bh6xg7 Bf8xg7 Qe4-g6+ Ke8-f8 Qg6-f5+
Bg7-f6) +7.08/19 14} 14. d5 {(d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 O-O
Rh8-g8 Qg6-f5+ e7-e6 Qf5-h7 Bf8-d6 g2-g3 Qd8-e8 Qh7xe4 Bb3-d5 Qe4-e3 Qe8-g6
f2-f3 Qg6-c2 Bc1-d2 Qc2-f5 c3-c4 Qf5-f4 Qe3-c3 Qf4-d4+ Qc3xd4 e5xd4 c4xd5
c6xd5 f3-f4 Rg8-f8 Be2-g4 Kc8-d7 f4-f5 e6xf5 Bg4xf5+ Rf8xf5 Rf1xf5 Kd7-e6
Rf5-f1 a6-a5 Rf1-e1+ Ke6-d7 h2-h4) -3.91/26 5} Na5 {(Nc6-a5 Qg6-e6+ Kd7-e8
Qe6-g6+ Ke8-d7) +7.08/20 14} 15. Qf5+ {(Qg6-f5+ Kd7-e8 Qf5-g6+ Ke8-d7)
0.00/65 2} Ke8 {(Kd7-e8 Qf5-g6+) +7.08/19 3} 16. Qg6+ {(Qf5-g6+ Ke8-d7
Qg6-e6+ Kd7-e8) 0.00/57 2} Kd7 {(Ke8-d7 Qg6-e6+) +7.08/21 3} 17. Qg4+
{(Qg6-g4+ Kd7-e8 Qg4-g6+ Ke8-d7) 0.00/59 3} Ke8 {(Kd7-e8 Qg4-g6+) +7.08/20
3} 18. Qg6+ {(Qg4-g6+) 0.00/65 3} Kd7 {(Ke8-d7) 0.00/64 1 3-fold
repetition} 1/2-1/2[/pgn]

[lkaufman]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

Does your Dragon has a good chance of drawing against GM Navara Knight Odds at time control of 3'+2"

Anyway I believe it is more interesting to see a match of Rook Odds versus a Master human player
[pgn][Event "Computer chess game"]
[Site "DESKTOP-4QNC0GS"]
[Date "2024.03.30"]
[Round "?"]
[White "Stockfish-windows-x86-64-avx2"]
[Black "Kingfisher.v1.1.1"]
[Result "1/2-1/2"]
[BlackElo "2200"]
[Time "11:47:02"]
[WhiteElo "3645"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1"]
[Termination "normal"]
[PlyCount "36"]
[WhiteType "program"]
[BlackType "program"]

1. e4 {(e2-e4 d7-d5 e4-e5 Bc8-f5 g2-g4 Bf5-d7 d2-d4 c7-c5 c2-c4 Bd7-c6
Bc1-g5 c5xd4 Qd1xd4) -4.21/27 10} Nc6 {(Nb8-c6 d2-d4 d7-d5 e4-e5 f7-f6
Nb1-c3 f6xe5 d4xe5 d5-d4 Nc3-e4 Bc8-f5 Ne4-c5 Nc6xe5 Bf1-b5+ c7-c6 Nc5xb7)
+5.64/16 12} 2. d4 {(d2-d4 e7-e6 d4-d5 e6xd5 e4xd5 Bf8-d6 Ng1-f3 Ng8-f6
d5xc6 d7xc6 Bf1-e2 O-O O-O h7-h6 Nb1-c3 Rf8-e8 Be2-d3 a7-a5 a2-a3 Bc8-d7
h2-h3 a5-a4) -3.89/19 0} d5 {(d7-d5) +5.68/15 2} 3. Nc3 {(Nb1-c3 e7-e5
Nc3xd5 Ng8-f6 Bf1-c4 Nc6xd4 c2-c3 Nf6xd5 e4xd5 Nd4-f5 Bc4-b3 Bf8-e7 h2-h4
O-O Ng1-f3 e5-e4 Nf3-g5 Nf5-h6 O-O Bc8-f5 Bc1-e3 Rf8-e8 Rf1-e1 Be7-d6
Qd1-c2) -4.14/29 13} Nf6 {(Ng8-f6 e4-e5 Nf6-e4 Ng1-f3 e7-e6 Bf1-d3 Bf8-b4
Bc1-d2 Bb4xc3 b2xc3 f7-f6 O-O Ne4xd2 Qd1xd2 Bc8-d7) +5.86/17 12} 4. e5
{(e4-e5 Nf6-g8 Bf1-b5 e7-e6 Ng1-f3 Ng8-e7 h2-h4 a7-a6 Bb5-d3 Nc6-b4 h4-h5
Nb4xd3+ Qd1xd3 Bc8-d7 g2-g4 Qd8-c8 h5-h6 g7xh6 g4-g5) -3.70/22 2} Ne4
{(Nf6-e4 Ng1-e2 f7-f6 Ne2-f4 Ne4xc3 b2xc3 g7-g6 e5xf6 e7xf6 Qd1-f3 Nc6-e7
Bf1-d3 c7-c6 Bc1-a3 Bf8-g7 Ke1-d2 O-O) +5.91/17 12} 5. Nxe4 {(Nc3xe4 d5xe4
c2-c3 Bc8-f5 Bf1-b5 a7-a6 Bb5xc6+ b7xc6 h2-h4 h7-h5 Qd1-a4 Ra8-b8 Ng1-e2
Bf5-d7 Ne2-f4 e7-e6) -3.59/23 4} dxe4 {(d5xe4 Bf1-b5 Bc8-d7 Bb5-c4 g7-g6
Ng1-e2 Bd7-f5 c2-c3 Nc6-a5 Bc4-b3 Bf8-g7 Bc1-f4 O-O b7-c6) +5.87/16 12} 6.
c3 {(c2-c3 Bc8-f5 Ng1-e2 e7-e6 Ne2-g3 Bf5-g6 Bf1-b5 a7-a6 Bb5-a4 b7-b5
Ba4-b3 h7-h5 O-O h5-h4 Ng3-e2 Nc6-a5 Ne2-f4 Bg6-f5 Bb3-c2 c7-c6 g2-g4
h4xg3/ep f2xg3) -3.55/27 10} Be6 {(Bc8-e6 Bf1-b5 a7-a6 Bb5xc6+ b7xc6 Qd1-c2
f7-f6 Ng1-e2 f6xe5 Qc2xe4 Qd8-d5 Qe4xe5 Qd5xe5 d4xe5 Be6xa2 Ke1-d1)
+5.85/15 13} 7. Ne2 {(Ng1-e2 Be6xa2 Qd1-c2 Ba2-c4 Qc2-a4 Bc4-d5 Ne2-f4
e7-e6 Bf1-e2 Bf8-e7 O-O O-O Rf1-d1 a7-a6 Qa4-c2 b7-b5 b2-b3 Kg8-h8 h2-h4
Be7xh4 c3-c4 Nc6-b4 Qc2-b1 Bd5-b7 d4-d5 e6xd5 g2-g3 Bh4-g5 Nf4xd5) -3.67/31
31} Bxa2 {(Be6xa2 Qd1-c2 Ba2-c4 Ne2-f4 Bc4xf1 Rh1xf1 g7-g5 Nf4-e2 Qd8-d5
Bc1xg5 Rh8-g8 Ne2-f4 Nc6xd4 c3xd4 Qd5xd4 Bg5-h4) +5.84/16 13} 8. Nf4
{(Ne2-f4 Qd8-d7 Bf1-e2 Nc6xe5 d4xe5 Qd7xd1+ Be2xd1 g7-g6 e5-e6 O-O-O h2-h4
f7-f5 Bd1-e2 Bf8-g7 c3-c4 b7-b5 c4xb5 Rd8-d6 h4-h5 Ba2xe6 b5-b6 a7xb6 h5xg6
h7xg6 Rh1xh8+ Bg7xh8) -3.73/26 7} a6 {(a7-a6 Qd1-a4 Ba2-b1 Qa4-b3 e7-e6
Bf1-e2 g7-g5 Nf4-h5 Bf8-e7 Bc1-e3 b7-b5 O-O Bb1-d3 Qb3-d1 Bd3xe2 Qd1xe2
c7xc6) +6.25/16 13} 9. e6 {(e5-e6 Ba2xe6 Nf4xe6 f7xe6 Bf1-c4 Qd8-d6 O-O
e6-e5 d4-d5 Nc6-d8 Bc4-a2 g7-g6 c3-c4 Qd6-d7 Rf1-e1 Nd8-f7 Re1xe4 Bf8-g7
h2-h4 c7-c6 h4-h5 c6xd5 c4xd5 Nf7-d6 Re4-g4 Ra8-c8 Bc1-g5 Rh8-f8 Bg5-e3
Rf8-f6 Rg4-b4 b7-b5) -3.45/27 11} fxe6 {(f7xe6 b2-b3 e6-e5 Nf4-e6 Qd8-d6
d4-d5 Nc6-a5 Bc1-a3 Qd6xa3 Ne6xc7+ Ke8-d8 Nc7xa8 Qa3xb3 Qd1xb3 Na5xb3)
+6.90/15 13} 10. b3 {(b2-b3 Qd8-d6 Bf1-c4 b7-b5 Bc4xe6 g7-g6 d4-d5 Bf8-h6
g2-g3 Nc6-d8 O-O Rh8-f8 Be6-h3 Bh6xf4 Bc1xf4 Rf8xf4 g3xf4 Qd6xf4 Qd1-c2
Nd8-f7 Qc2xa2 Nf7-e5 Bh3-g2 Ne5-g4 Rf1-e1 e4-e3 f2-f3 Qf4xh2+ Kg1-f1 Qh2-f4
Qa2-e2 Ra8-d8 Bg2-h3 Ng4-f6 c3-c4 b5xc4 b3xc4 Qf4-h4 Kf1-g2 Qh4-g5+ Kg2-h1)
-3.80/25 12} e5 {(e6-e5 Qd1-h5+ g7-g6 Nf4xg6 h7xg6 Qh5xh8 Ba2xb3 Bc1-h6
Ke8-f7 Qh8-h7+ Kf7-f6 d4xe5+ Nc6xe5 Bh6-f4 e7-e6) +7.17/15 13} 11. Qh5+
{(Qd1-h5+ g7-g6 Nf4xg6 h7xg6 Qh5xg6+ Ke8-d7 d4-d5 Kd7-c8 d5xc6 Ba2xb3
Bf1-e2 b7xc6 Bc1-e3 e7-e6 h2-h4 Kc8-b7 Qg6xe4 a6-a5 Qe4-b1 a5-a4 h4-h5
Qd8-d5 O-O Rh8-g8 g2-g4 a4-a3 c3-c4 Rg8xg4+ Be2xg4 Qd5xc4 Bg4-f3 a3-a2)
-3.68/27 11} g6 {(g7-g6 Nf4xg6 h7xg6 Qh5xh8 Ba2xb3 Bc1-h6 Ke8-f7 Qh8-h7+
Kf7-f6 Qh7-h8+ Kf6-f7) +7.03/21 13} 12. Nxg6 {(Nf4xg6 h7xg6 Qh5xg6+ Ke8-d7
d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 c3-c4 Qd8-d6 Qg6-g4+ Qd6-d7 Qg4-g3
Bb3-c2 O-O e7-e6 h2-h4 a6-a5 Bc1-e3 Bf8-b4 h4-h5 Kc8-b7 Qg3xe5 Bc2-d3
Rf1-d1 Rh8-g8 h5-h6) -3.93/29 16} hxg6 {(h7xg6 Qh5xh8 Ba2xb3) +7.03/20 13}
13. Qxg6+ {(Qh5xg6+ Ke8-d7 d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 O-O
Rh8-g8 Qg6-h5 e7-e6 Bc1-e3 Bf8-d6 c3-c4 Qd8-e8 Qh5-h3 Qe8-g6 g2-g3 a6-a5
Rf1-a1 c6-c5 Qh3-g2 a5-a4 Ra1-a3 c7-c6 Qg2-f1 Kc8-c7 Qf1-a1 Rg8-b8)
-3.92/26 4} Kd7 {(Ke8-d7 d4-d5 Nc6-a5 Qg6-e6+ Kd7-e8 Qe6xe5 Rh8-h4 Bc1-g5
Rh4-h7 Qe5xe4 Rh7-g7 Bg5-h6 Ba2xb3 Bh6xg7 Bf8xg7 Qe4-g6+ Ke8-f8 Qg6-f5+
Bg7-f6) +7.08/19 14} 14. d5 {(d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 O-O
Rh8-g8 Qg6-f5+ e7-e6 Qf5-h7 Bf8-d6 g2-g3 Qd8-e8 Qh7xe4 Bb3-d5 Qe4-e3 Qe8-g6
f2-f3 Qg6-c2 Bc1-d2 Qc2-f5 c3-c4 Qf5-f4 Qe3-c3 Qf4-d4+ Qc3xd4 e5xd4 c4xd5
c6xd5 f3-f4 Rg8-f8 Be2-g4 Kc8-d7 f4-f5 e6xf5 Bg4xf5+ Rf8xf5 Rf1xf5 Kd7-e6
Rf5-f1 a6-a5 Rf1-e1+ Ke6-d7 h2-h4) -3.91/26 5} Na5 {(Nc6-a5 Qg6-e6+ Kd7-e8
Qe6-g6+ Ke8-d7) +7.08/20 14} 15. Qf5+ {(Qg6-f5+ Kd7-e8 Qf5-g6+ Ke8-d7)
0.00/65 2} Ke8 {(Kd7-e8 Qf5-g6+) +7.08/19 3} 16. Qg6+ {(Qf5-g6+ Ke8-d7
Qg6-e6+ Kd7-e8) 0.00/57 2} Kd7 {(Ke8-d7 Qg6-e6+) +7.08/21 3} 17. Qg4+
{(Qg6-g4+ Kd7-e8 Qg4-g6+ Ke8-d7) 0.00/59 3} Ke8 {(Kd7-e8 Qg4-g6+) +7.08/20
3} 18. Qg6+ {(Qg4-g6+) 0.00/65 3} Kd7 {(Ke8-d7) 0.00/64 1 3-fold
repetition} 1/2-1/2[/pgn]

[lkaufman]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

Does your Dragon has a good chance of drawing against GM Navara Knight Odds at time control of 3'+2"

Anyway I believe it is more interesting to see a match of Rook Odds versus a Master human player
[pgn][Event "Computer chess game"]
[Site "DESKTOP-4QNC0GS"]
[Date "2024.03.30"]
[Round "?"]
[White "Stockfish-windows-x86-64-avx2"]
[Black "Kingfisher.v1.1.1"]
[Result "1/2-1/2"]
[BlackElo "2200"]
[Time "11:47:02"]
[WhiteElo "3645"]
[TimeControl "180+2"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/1NBQKBNR w Kkq - 0 1"]
[Termination "normal"]
[PlyCount "36"]
[WhiteType "program"]
[BlackType "program"]

1. e4 {(e2-e4 d7-d5 e4-e5 Bc8-f5 g2-g4 Bf5-d7 d2-d4 c7-c5 c2-c4 Bd7-c6
Bc1-g5 c5xd4 Qd1xd4) -4.21/27 10} Nc6 {(Nb8-c6 d2-d4 d7-d5 e4-e5 f7-f6
Nb1-c3 f6xe5 d4xe5 d5-d4 Nc3-e4 Bc8-f5 Ne4-c5 Nc6xe5 Bf1-b5+ c7-c6 Nc5xb7)
+5.64/16 12} 2. d4 {(d2-d4 e7-e6 d4-d5 e6xd5 e4xd5 Bf8-d6 Ng1-f3 Ng8-f6
d5xc6 d7xc6 Bf1-e2 O-O O-O h7-h6 Nb1-c3 Rf8-e8 Be2-d3 a7-a5 a2-a3 Bc8-d7
h2-h3 a5-a4) -3.89/19 0} d5 {(d7-d5) +5.68/15 2} 3. Nc3 {(Nb1-c3 e7-e5
Nc3xd5 Ng8-f6 Bf1-c4 Nc6xd4 c2-c3 Nf6xd5 e4xd5 Nd4-f5 Bc4-b3 Bf8-e7 h2-h4
O-O Ng1-f3 e5-e4 Nf3-g5 Nf5-h6 O-O Bc8-f5 Bc1-e3 Rf8-e8 Rf1-e1 Be7-d6
Qd1-c2) -4.14/29 13} Nf6 {(Ng8-f6 e4-e5 Nf6-e4 Ng1-f3 e7-e6 Bf1-d3 Bf8-b4
Bc1-d2 Bb4xc3 b2xc3 f7-f6 O-O Ne4xd2 Qd1xd2 Bc8-d7) +5.86/17 12} 4. e5
{(e4-e5 Nf6-g8 Bf1-b5 e7-e6 Ng1-f3 Ng8-e7 h2-h4 a7-a6 Bb5-d3 Nc6-b4 h4-h5
Nb4xd3+ Qd1xd3 Bc8-d7 g2-g4 Qd8-c8 h5-h6 g7xh6 g4-g5) -3.70/22 2} Ne4
{(Nf6-e4 Ng1-e2 f7-f6 Ne2-f4 Ne4xc3 b2xc3 g7-g6 e5xf6 e7xf6 Qd1-f3 Nc6-e7
Bf1-d3 c7-c6 Bc1-a3 Bf8-g7 Ke1-d2 O-O) +5.91/17 12} 5. Nxe4 {(Nc3xe4 d5xe4
c2-c3 Bc8-f5 Bf1-b5 a7-a6 Bb5xc6+ b7xc6 h2-h4 h7-h5 Qd1-a4 Ra8-b8 Ng1-e2
Bf5-d7 Ne2-f4 e7-e6) -3.59/23 4} dxe4 {(d5xe4 Bf1-b5 Bc8-d7 Bb5-c4 g7-g6
Ng1-e2 Bd7-f5 c2-c3 Nc6-a5 Bc4-b3 Bf8-g7 Bc1-f4 O-O b7-c6) +5.87/16 12} 6.
c3 {(c2-c3 Bc8-f5 Ng1-e2 e7-e6 Ne2-g3 Bf5-g6 Bf1-b5 a7-a6 Bb5-a4 b7-b5
Ba4-b3 h7-h5 O-O h5-h4 Ng3-e2 Nc6-a5 Ne2-f4 Bg6-f5 Bb3-c2 c7-c6 g2-g4
h4xg3/ep f2xg3) -3.55/27 10} Be6 {(Bc8-e6 Bf1-b5 a7-a6 Bb5xc6+ b7xc6 Qd1-c2
f7-f6 Ng1-e2 f6xe5 Qc2xe4 Qd8-d5 Qe4xe5 Qd5xe5 d4xe5 Be6xa2 Ke1-d1)
+5.85/15 13} 7. Ne2 {(Ng1-e2 Be6xa2 Qd1-c2 Ba2-c4 Qc2-a4 Bc4-d5 Ne2-f4
e7-e6 Bf1-e2 Bf8-e7 O-O O-O Rf1-d1 a7-a6 Qa4-c2 b7-b5 b2-b3 Kg8-h8 h2-h4
Be7xh4 c3-c4 Nc6-b4 Qc2-b1 Bd5-b7 d4-d5 e6xd5 g2-g3 Bh4-g5 Nf4xd5) -3.67/31
31} Bxa2 {(Be6xa2 Qd1-c2 Ba2-c4 Ne2-f4 Bc4xf1 Rh1xf1 g7-g5 Nf4-e2 Qd8-d5
Bc1xg5 Rh8-g8 Ne2-f4 Nc6xd4 c3xd4 Qd5xd4 Bg5-h4) +5.84/16 13} 8. Nf4
{(Ne2-f4 Qd8-d7 Bf1-e2 Nc6xe5 d4xe5 Qd7xd1+ Be2xd1 g7-g6 e5-e6 O-O-O h2-h4
f7-f5 Bd1-e2 Bf8-g7 c3-c4 b7-b5 c4xb5 Rd8-d6 h4-h5 Ba2xe6 b5-b6 a7xb6 h5xg6
h7xg6 Rh1xh8+ Bg7xh8) -3.73/26 7} a6 {(a7-a6 Qd1-a4 Ba2-b1 Qa4-b3 e7-e6
Bf1-e2 g7-g5 Nf4-h5 Bf8-e7 Bc1-e3 b7-b5 O-O Bb1-d3 Qb3-d1 Bd3xe2 Qd1xe2
c7xc6) +6.25/16 13} 9. e6 {(e5-e6 Ba2xe6 Nf4xe6 f7xe6 Bf1-c4 Qd8-d6 O-O
e6-e5 d4-d5 Nc6-d8 Bc4-a2 g7-g6 c3-c4 Qd6-d7 Rf1-e1 Nd8-f7 Re1xe4 Bf8-g7
h2-h4 c7-c6 h4-h5 c6xd5 c4xd5 Nf7-d6 Re4-g4 Ra8-c8 Bc1-g5 Rh8-f8 Bg5-e3
Rf8-f6 Rg4-b4 b7-b5) -3.45/27 11} fxe6 {(f7xe6 b2-b3 e6-e5 Nf4-e6 Qd8-d6
d4-d5 Nc6-a5 Bc1-a3 Qd6xa3 Ne6xc7+ Ke8-d8 Nc7xa8 Qa3xb3 Qd1xb3 Na5xb3)
+6.90/15 13} 10. b3 {(b2-b3 Qd8-d6 Bf1-c4 b7-b5 Bc4xe6 g7-g6 d4-d5 Bf8-h6
g2-g3 Nc6-d8 O-O Rh8-f8 Be6-h3 Bh6xf4 Bc1xf4 Rf8xf4 g3xf4 Qd6xf4 Qd1-c2
Nd8-f7 Qc2xa2 Nf7-e5 Bh3-g2 Ne5-g4 Rf1-e1 e4-e3 f2-f3 Qf4xh2+ Kg1-f1 Qh2-f4
Qa2-e2 Ra8-d8 Bg2-h3 Ng4-f6 c3-c4 b5xc4 b3xc4 Qf4-h4 Kf1-g2 Qh4-g5+ Kg2-h1)
-3.80/25 12} e5 {(e6-e5 Qd1-h5+ g7-g6 Nf4xg6 h7xg6 Qh5xh8 Ba2xb3 Bc1-h6
Ke8-f7 Qh8-h7+ Kf7-f6 d4xe5+ Nc6xe5 Bh6-f4 e7-e6) +7.17/15 13} 11. Qh5+
{(Qd1-h5+ g7-g6 Nf4xg6 h7xg6 Qh5xg6+ Ke8-d7 d4-d5 Kd7-c8 d5xc6 Ba2xb3
Bf1-e2 b7xc6 Bc1-e3 e7-e6 h2-h4 Kc8-b7 Qg6xe4 a6-a5 Qe4-b1 a5-a4 h4-h5
Qd8-d5 O-O Rh8-g8 g2-g4 a4-a3 c3-c4 Rg8xg4+ Be2xg4 Qd5xc4 Bg4-f3 a3-a2)
-3.68/27 11} g6 {(g7-g6 Nf4xg6 h7xg6 Qh5xh8 Ba2xb3 Bc1-h6 Ke8-f7 Qh8-h7+
Kf7-f6 Qh7-h8+ Kf6-f7) +7.03/21 13} 12. Nxg6 {(Nf4xg6 h7xg6 Qh5xg6+ Ke8-d7
d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 c3-c4 Qd8-d6 Qg6-g4+ Qd6-d7 Qg4-g3
Bb3-c2 O-O e7-e6 h2-h4 a6-a5 Bc1-e3 Bf8-b4 h4-h5 Kc8-b7 Qg3xe5 Bc2-d3
Rf1-d1 Rh8-g8 h5-h6) -3.93/29 16} hxg6 {(h7xg6 Qh5xh8 Ba2xb3) +7.03/20 13}
13. Qxg6+ {(Qh5xg6+ Ke8-d7 d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 O-O
Rh8-g8 Qg6-h5 e7-e6 Bc1-e3 Bf8-d6 c3-c4 Qd8-e8 Qh5-h3 Qe8-g6 g2-g3 a6-a5
Rf1-a1 c6-c5 Qh3-g2 a5-a4 Ra1-a3 c7-c6 Qg2-f1 Kc8-c7 Qf1-a1 Rg8-b8)
-3.92/26 4} Kd7 {(Ke8-d7 d4-d5 Nc6-a5 Qg6-e6+ Kd7-e8 Qe6xe5 Rh8-h4 Bc1-g5
Rh4-h7 Qe5xe4 Rh7-g7 Bg5-h6 Ba2xb3 Bh6xg7 Bf8xg7 Qe4-g6+ Ke8-f8 Qg6-f5+
Bg7-f6) +7.08/19 14} 14. d5 {(d4-d5 Kd7-c8 d5xc6 Ba2xb3 Bf1-e2 b7xc6 O-O
Rh8-g8 Qg6-f5+ e7-e6 Qf5-h7 Bf8-d6 g2-g3 Qd8-e8 Qh7xe4 Bb3-d5 Qe4-e3 Qe8-g6
f2-f3 Qg6-c2 Bc1-d2 Qc2-f5 c3-c4 Qf5-f4 Qe3-c3 Qf4-d4+ Qc3xd4 e5xd4 c4xd5
c6xd5 f3-f4 Rg8-f8 Be2-g4 Kc8-d7 f4-f5 e6xf5 Bg4xf5+ Rf8xf5 Rf1xf5 Kd7-e6
Rf5-f1 a6-a5 Rf1-e1+ Ke6-d7 h2-h4) -3.91/26 5} Na5 {(Nc6-a5 Qg6-e6+ Kd7-e8
Qe6-g6+ Ke8-d7) +7.08/20 14} 15. Qf5+ {(Qg6-f5+ Kd7-e8 Qf5-g6+ Ke8-d7)
0.00/65 2} Ke8 {(Kd7-e8 Qf5-g6+) +7.08/19 3} 16. Qg6+ {(Qf5-g6+ Ke8-d7
Qg6-e6+ Kd7-e8) 0.00/57 2} Kd7 {(Ke8-d7 Qg6-e6+) +7.08/21 3} 17. Qg4+
{(Qg6-g4+ Kd7-e8 Qg4-g6+ Ke8-d7) 0.00/59 3} Ke8 {(Kd7-e8 Qg4-g6+) +7.08/20
3} 18. Qg6+ {(Qg4-g6+) 0.00/65 3} Kd7 {(Ke8-d7) 0.00/64 1 3-fold
repetition} 1/2-1/2[/pgn]

There have already been many games played by LeelaRookOdds on LiChess against masters up thru International Masters. There have been a large number by one rather strong IM, Joan Fluvia (FIDE 2445, was over 2500), mostly at 3' + 5", with the score 29 wins for Leela, 17 wins for Fluvia, and 3 draws. The games are all there to be viewed.

[Chessqueen]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

There have already been many games played by LeelaRookOdds on LiChess against masters up thru International Masters. There have been a large number by one rather strong IM, Joan Fluvia (FIDE 2445, was over 2500), mostly at 3' + 5", with the score 29 wins for Leela, 17 wins for Fluvia, and 3 draws. The games are all there to be viewed.

My Next question is, after observing different Odds, how much Elo is worth to give a knight odds, would you estimate around 700 elo points versus GMs at 3+2" time control; Whereas Leela without its Knight might be only rated around 2740 at the most in comparison to Human FIDE rating?

[lkaufman]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

There have already been many games played by LeelaRookOdds on LiChess against masters up thru International Masters. There have been a large number by one rather strong IM, Joan Fluvia (FIDE 2445, was over 2500), mostly at 3' + 5", with the score 29 wins for Leela, 17 wins for Fluvia, and 3 draws. The games are all there to be viewed.

My Next question is, after observing different Odds, how much Elo is worth to give a knight odds, would you estimate around 700 elo points versus GMs at 3+2" time control; Whereas Leela without its Knight might be only rated around 2740 at the most in comparison to Human FIDE rating?

I think that the match showed that a lot depends on preparation and motivation. It seems that with motivation and preparation (he reviewed 250 LeelaKnightOdds games!) he is a slight favorite at 3'2", maybe it would be almost balanced if only g1 odds. Probably his FIDE blitz rating of 2688 is the most relevant one for this (and it's near the average of his rapid and classical ratings). so LeelaKnightOdds is roughly 2650 level at 3'2" against a very well-prepared, motivated opponent, perhaps more like 2750 against a casual opponent. We don't really know what FIDE rating Leela would have on good hardware like this but it might be close to 1000 above the 2650 figure, depending on what assumptions one makes about how it would get a FIDE rating in theory. Probably it will be possible to improve odds performance by better settings or better nets or better opening book choices.

[Chessqueen]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

There have already been many games played by LeelaRookOdds on LiChess against masters up thru International Masters. There have been a large number by one rather strong IM, Joan Fluvia (FIDE 2445, was over 2500), mostly at 3' + 5", with the score 29 wins for Leela, 17 wins for Fluvia, and 3 draws. The games are all there to be viewed.

My Next question is, after observing different Odds, how much Elo is worth to give a knight odds, would you estimate around 700 elo points versus GMs at 3+2" time control; Whereas Leela without its Knight might be only rated around 2740 at the most in comparison to Human FIDE rating?

I think that the match showed that a lot depends on preparation and motivation. It seems that with motivation and preparation (he reviewed 250 LeelaKnightOdds games!) he is a slight favorite at 3'2", maybe it would be almost balanced if only g1 odds. Probably his FIDE blitz rating of 2688 is the most relevant one for this (and it's near the average of his rapid and classical ratings). so LeelaKnightOdds is roughly 2650 level at 3'2" against a very well-prepared, motivated opponent, perhaps more like 2750 against a casual opponent. We don't really know what FIDE rating Leela would have on good hardware like this but it might be close to 1000 above the 2650 figure, depending on what assumptions one makes about how it would get a FIDE rating in theory. Probably it will be possible to improve odds performance by better settings or better nets or better opening book choices.

It would have been interesting to see Leela playing half of the games as Black, but with 3+2" whereas GM David Navarra playing White with all the pieces only have 3+1"

[lkaufman]

8.5 to 4.5 David Wins

Quote Navara "At least now I knew I were playing a engine!"

The first 3 games (1.5 each) were his practice games last week, the actual match score was 7 to 3 for David. He won decisively until dropping the time limit to 3'+1", where he lost 1.5 to 0.5. But he went 4.5 to 1.5 at 3'2", so it looks like the fair time limit is between those two. He scored better with b1 odds (4 to 1) than with g1 odds (3 to 2). Maybe for a future match most or all games need to be g1 odds, b1 looks too easy for the human. David prepared well for the match, he said he reviewed 250 LeelaKnightOdds games. He played quite well, only one real blunder of a piece which didn't even cost him the game, he still drew. Leela seemed to play a bit too "normally", but this may be mostly a consequence of the openings, it rarely got openings with serious attacking chances. The opening book was turned off for the final four games; it's not clear if that helped or hurt. He did much better than pure calculations predicted, perhaps due to his preparation or perhaps it's just that the elo model based on standard chess is not so accurate for predicting results in these odds games.

There have already been many games played by LeelaRookOdds on LiChess against masters up thru International Masters. There have been a large number by one rather strong IM, Joan Fluvia (FIDE 2445, was over 2500), mostly at 3' + 5", with the score 29 wins for Leela, 17 wins for Fluvia, and 3 draws. The games are all there to be viewed.

My Next question is, after observing different Odds, how much Elo is worth to give a knight odds, would you estimate around 700 elo points versus GMs at 3+2" time control; Whereas Leela without its Knight might be only rated around 2740 at the most in comparison to Human FIDE rating?

I think that the match showed that a lot depends on preparation and motivation. It seems that with motivation and preparation (he reviewed 250 LeelaKnightOdds games!) he is a slight favorite at 3'2", maybe it would be almost balanced if only g1 odds. Probably his FIDE blitz rating of 2688 is the most relevant one for this (and it's near the average of his rapid and classical ratings). so LeelaKnightOdds is roughly 2650 level at 3'2" against a very well-prepared, motivated opponent, perhaps more like 2750 against a casual opponent. We don't really know what FIDE rating Leela would have on good hardware like this but it might be close to 1000 above the 2650 figure, depending on what assumptions one makes about how it would get a FIDE rating in theory. Probably it will be possible to improve odds performance by better settings or better nets or better opening book choices.

It would have been interesting to see Leela playing half of the games as Black, but with 3+2" whereas GM David Navarra playing White with all the pieces only have 3+1"

An anonymous GM rated 2958 blitz on LIChess (handle Brown_player) played a lot of blitz games at 3'2" with LeelaKnightOdds after the Navara match but came out behind. So David really did very well to win the series at 3'2", presumably due to his excellent preparation, attitude, and form. A funny development today; some unrated anonymous player was suspended for repeatedly cheating against LeelaKnightOdds, but despite such cheating and the knight odds, he still lost every game (9 games!). Now if someone can't win or draw a single game starting a knight up using a computer to help him, he must be a REALLY bad player (or he uses a really bad engine)!