I beat Telepath

[CRoberson]

Not really, I cheated.

This comesfrom a discussion that Bob, Zach and I had in October.
I had Telepath play without pondering and I always played the
move from its PV. I believe theory suggests that I can win lose or
draw doing this. I don't know what the odds are on this but it sounds
like an interesting experiment to code.

I have played 3 games in this manner (all as white). I won 2 and
lost 1. Here is one of the games.

[Event "Computer chess game"]
[Site "HEPHAESTUS"]
[Date "2009.01.11"]
[Round "?"]
[White "Owner"]
[Black "Telepath Version 5.014"]
[Result "1-0"]
[ECO "C47"]
[Opening "Four Knights"]
[Time "20:08:01"]
[Variation "Scotch, Main Line, 7...d5"]
[WhiteElo "2350"]
[TimeControl "180+2"]
[Termination "normal"]
[PlyCount "141"]
[WhiteType "human"]
[BlackType "program"]

1.e4 e5 2.Nf3 Nc6 3.d4 exd4 4.Nxd4 Nf6 5.Nc3 Bb4 6.Nxc6 bxc6 7.Bd3 d5 8.e5 Qe7
{(Qd8e7 00 Nf6g4 Bc1f4 Bb4xc3 b2xc3 g7g5 Bf4g3 Ng4xe5 Rf1e1) +0.17/11 7} 9.O-O Ng4
{(Nf6g4 Bc1f4 Bb4xc3 b2xc3 g7g5 Bf4g3 Ng4xe5 Rf1e1 f7f6) -0.07/11 16} 10.Bf4 O-O {(00
Qd1e2 Rf8e8 Ra1e1 Qe7c5 h2h3 Qc5d4 g2g3 Bb4xc3) -0.06/10 8} 11.Qe2 g5 {(g7g5 Bf4g3
Rf8e8 f2f4 Bb4xc3 b2xc3 g5xf4 Rf1xf4 Ng4xe5) +0.06/10 9} 12.Bg3 h5 {(h7h5 h2h3 h5h4
Bg3xh4 g5xh4 h3xg4 Qe7g5 f2f3 Bb4c5+ Kg1h1) -0.30/10 7} 13.h3 h4 {(h5h4 Bg3xh4 g5xh4
h3xg4 Qe7g5 Bd3f5 Bc8xf5 g4xf5 Qg5xf5 Nc3a4 h4h3) -0.29/12 8} 14.Bxh4 gxh4 {(g5xh4
h3xg4 Qe7g5 Bd3f5 Bc8xf5 g4xf5 Qg5xf5 Nc3a4 h4h3 f2f4 Ra8b8) -0.28/12 9} 15.hxg4 Qg5
{(Qe7g5 Bd3f5 Bc8xf5 g4xf5 Bb4xc3 b2xc3 Qg5xf5 f2f4 Rf8e8 Ra1b1) -0.24/12 15} 16.Bf5
Bxf5 {(Bc8xf5 g4xf5 Qg5xf5 Nc3a4 h4h3 f2f4 Ra8b8 Qe2d3 Qf5xd3 c2xd3 h3xg2) -0.29/11
3} 17.gxf5 Qxf5 {(Qg5xf5 f2f4 Bb4c5+ Kg1h1 Qf5g6 Nc3a4 Bc5d4 c2c3 Bd4b6 Na4xb6)
-0.03/10 7} 18.f4 Bc5+ {(Bb4c5+ Kg1h1 Rf8b8 Nc3a4 Bc5e7 Qe2d3 Qf5g4 Kh1g1 Kg8h8 Ra1d1)
-0.21/10 7} 19.Kh1 Kh8 {(Kg8h8 Nc3a4 Bc5b6 Na4xb6 a7xb6 Qe2d3 Qf5g4 Qd3f3 Qg4f5 Qf3h3)
-0.04/11 8} 20.Na4 Bb6 {(Bc5b6 Qe2d3 Qf5g4 Na4xb6 a7xb6 Qd3f3 Qg4f5 Qf3c3 c6c5 Kh1g1
Rf8g8) -0.16/12 8} 21.Qd3 Qg4 {(Qf5g4 Qd3f3 Rf8g8 Na4xb6 c7xb6 Qf3xg4 Rg8xg4 Rf1f2
Ra8g8 Ra1d1 Rg4g3) -0.13/12 6} 22.Qf3 Rfg8 {(Rf8g8 Qf3xg4 Rg8xg4 Na4xb6 c7xb6 Rf1f2
Ra8g8 Ra1d1 Rg4g3 Kh1h2 a7a5) -0.08/13 10} 23.Qxg4 Rxg4 {(Rg8xg4 Na4xb6 c7xb6 Rf1f2
Ra8g8 Kh1h2 Rg4g3 Ra1d1 a7a5 Kh2g1) -0.05/11} 24.Nxb6 cxb6 {(c7xb6 Ra1e1 Ra8e8 Rf1f3
Re8g8 Re1e2 Kh8h7 Kh1g1 Rg4g3 Rf3d3 h4h3) -0.14/12 7} 25.Rae1 Rag8 {(Ra8g8 Rf1f2
Rg8e8 Kh1h2 Re8e6 Re1d1 Kh8g7 b2b3 Rg4g3 c2c4 d5xc4) -0.29/12 5} 26.Rf2 Re8 {(Rg8e8
Kh1g1 Rg4g3 Re1d1 Kh8g7 Rd1d3 Kg7h6 Rf2e2 a7a5 a2a4 Re8g8) -0.13/12 6} 27.Kg1 Rg3
{(Rg4g3 Kg1h2 Kh8g8 f4f5 Rg3g4 e5e6 f7xe6 Rf2e2 e6e5 Re2xe5 Re8xe5) -0.37/12 5} 28.Kh2
Kg8 {(Kh8g8 Rf2f3 Rg3g6 Kh2h3 Kg8f8 Re1d1 Rg6h6 Rf3c3 a7a5 a2a4 Re8e7) -0.37/12 5}
29.Rf3 Rg6 {(Rg3g6 Kh2h3 Kg8f8 f4f5 Rg6h6 e5e6 c6c5 Rf3e3 Re8e7 e6xf7 Re7xe3+ Re1xe3)
-0.53/12 5} 30.Kh3 Kf8 {(Kg8f8 f4f5 Rg6g5 e5e6 Re8e7 Rf3f2 Rg5h5 Re1f1 a7a5 e6xf7
Kf8xf7) -0.67/11 4} 31.f5 Rg5 {(Rg6g5 e5e6 a7a5 Rf3f2 Rg5h5 Rf2f4 Re8e7 Re1e5 Rh5g5
Rf4xh4) -0.70/12 7} 32.e6 Rg3+ {(Rg5g3+ Rf3xg3 h4xg3 Kh3g4 f7xe6 Re1xe6 Re8xe6 f5xe6
Kf8e7 Kg4f5 c6c5 e6f7) -0.37/11 4} 33.Rxg3 hxg3 {(h4xg3 Kh3g4 f7xe6 Re1xe6 Re8xe6
f5xe6 Kf8e7 Kg4f5 a7a5 Kf5e5 a5a4 c2c3) -0.38/14 7} 34.Kxg3 fxe6 {(f7xe6 a2a4 e6xf5
Re1xe8+ Kf8xe8 Kg3f4 a7a6 Kf4xf5 b6b5 a4xb5 a6xb5 g2g4 Ke8f7) -0.13/13 4} 35.a4 exf5
{(e6xf5 Re1xe8+ Kf8xe8 Kg3f4 a7a6 Kf4xf5 b6b5 a4a5 Ke8f7 g2g4 c6c5 g4g5) -0.07/12 2}
36.Rxe8+ Kxe8 {(Kf8xe8 Kg3f4 a7a6 Kf4xf5 Ke8f7 g2g4 b6b5 g4g5 b5xa4 g5g6+ Kf7g7)
-0.14/11} 37.Kf4 d4 {(d5d4 b2b4 c6c5 b4xc5 b6xc5 Kf4xf5 c5c4 Kf5e4 d4d3 c2xd3 c4c3 Ke4e3
Ke8d7 g2g4 Kd7e6) -0.77/16 5} 38.b4 Ke7 {(Ke8e7 Kf4xf5 Ke7d6 g2g4 c6c5 b4xc5+ b6xc5
Kf5e4 Kd6e6 g4g5 a7a5 g5g6 Ke6f6 Ke4d5 Kf6xg6 Kd5xc5) -0.97/17 4} 39.Kxf5 Kd6 {(Ke7d6
g2g4 a7a5 b4xa5 b6xa5 g4g5 c6c5 g5g6 Kd6e7 Kf5e5 Ke7f8) -1.06/17 5} 40.g4 c5 {(c6c5
b4xc5+ b6xc5 a4a5 Kd6d5 g4g5 Kd5d6 g5g6 Kd6e7 Kf5e5 Ke7f8 Ke5d5 Kf8g7 Kd5xc5 Kg7xg6
a5a6) -1.11/16 3} 41.bxc5+ bxc5 {(b6xc5 g4g5 c5c4 Kf5e4 d4d3 c2xd3 c4c3 Ke4e3 Kd6e5
d3d4+ Ke5f5 Ke3d3 Kf5xg5 d4d5 Kg5f5 Kd3xc3 Kf5e5) -1.36/17 7} 42.g5 c4 {(c5c4 Kf5e4
d4d3 c2xd3 c4c3 Ke4e3 Kd6e5 d3d4+ Ke5f5 Ke3d3 Kf5xg5 Kd3xc3 Kg5f5 Kc3c4 Kf5e4 d4d5)
-1.36/17 6} 43.Ke4 d3 {(d4d3 c2xd3 c4c3 Ke4e3 Kd6e5 d3d4+ Ke5f5 Ke3d3 Kf5xg5 Kd3xc3
Kg5f5 Kc3c4 Kf5e4 d4d5 Ke4e5 a4a5 Ke5d6 Kc4d4 a7a6) -1.46/19 4} 44.cxd3 c3 {(c4c3
Ke4e3 Kd6e5 d3d4+ Ke5f5 Ke3d3 Kf5xg5 Kd3xc3 Kg5f5 Kc3c4 Kf5e4 Kc4c5 Ke4f4 d4d5 Kf4e5
d5d6 Ke5e6) -1.80/18 3} 45.Ke3 Ke5 {(Kd6e5 d3d4+ Ke5f5 Ke3d3 c3c2 Kd3xc2 Kf5xg5 a4a5
Kg5f5 Kc2d3 Kf5f4 d4d5 Kf4e5 Kd3c4 Ke5d6 a5a6) -1.46/16} 46.d4+ Kf5 {(Ke5f5 Ke3d3
c3c2 Kd3xc2 Kf5xg5 Kc2d3 Kg5f5 Kd3c4 Kf5e4 d4d5 Ke4e5 Kc4c5 Ke5f5 a4a5 Kf5e5 d5d6)
-1.80/17} 47.Kd3 c2 {(c3c2 Kd3xc2 Kf5xg5 Kc2d3 Kg5f6 Kd3c4 a7a6 Kc4c5 Kf6f5 a4a5 Kf5e4
Kc5b6 Ke4xd4) -2.48/18 1} 48.Kxc2 Kxg5 {(Kf5xg5 Kc2d3 Kg5f5 Kd3c4 Kf5e6 a4a5 Ke6d6
Kc4b5 Kd6c7 Kb5a6 Kc7b8 d4d5 Kb8a8 d5d6 Ka8b8 d6d7 Kb8c7 Ka6xa7 Kc7xd7) -2.71/20 1}
49.Kd3 Kf5 {(Kg5f5 Kd3c4 Kf5e6 Kc4c5 Ke6d7 Kc5b5 Kd7c7 Kb5a6 Kc7b8 a4a5 Kb8a8 d4d5
Ka8b8 d5d6 Kb8c8 Ka6xa7 Kc8d7 Ka7b6 Kd7e6 a5a6 Ke6xd6 a6a7 Kd6e6) -10.21/23 3} 50.Kc4
Ke6 {(Kf5e6 Kc4c5 Ke6d7 Kc5b5 Kd7c7 Kb5a6 Kc7b8 d4d5 Kb8a8 a4a5) -10.21/19} 51.Kc5
Kd7 {(Ke6d7 Kc5b5 Kd7c7 Kb5a6 Kc7b8 d4d5 Kb8a8 d5d6 Ka8b8 a4a5 Kb8c8 Ka6xa7 Kc8d7
Ka7b6 Kd7xd6 a5a6 Kd6e5 a6a7 Ke5f5 a7a8Q Kf5g6) -10.23/23 3} 52.Kb5 Kc7 {(Kd7c7 Kb5a6
Kc7b8 d4d5 Kb8a8 d5d6 Ka8b8 a4a5 Kb8c8 Ka6xa7 Kc8d7 Ka7b6 Kd7xd6 a5a6 Kd6d5 a6a7
Kd5e6 a7a8Q Ke6f6 Kb6c5 Kf6g6 Kg6g5 Kc5d4 Kg5f6) -10.42/25 2} 53.Ka6 Kb8 {(Kc7b8 d4d5
Kb8a8 d5d6 Ka8b8 a4a5 Kb8c8 Ka6xa7 Kc8d7 Ka7b6 Kd7xd6 a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q
Ke6f6 Kf6e7 Kb6c5 Ke7f7 Kf7f6 Kc5d4 Kf6f7) -10.43/26 3} 54.d5 Kc7 {(Kb8c7 Ka6xa7
Kc7d6 a4a5 Kd6xd5 Ka7b6 Kd5d6 a5a6 Kd6d5 a6a7 Kd5c4 a7a8Q Kc4d4 Kd4e4 Kb6c5+ Ke4f5
Kf5f6 Kc5d4 Kf6g7 Kg7f7) -10.47/23 2} 55.Kxa7 Kd6 {(Kc7d6 a4a5 Kd6xd5 Ka7b6 Kd5d6
a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q Ke6f6 Kf6e6 Kb6c5 Ke6e5 Ke5e4 Ke4f4 Kf4g4 Kc5d5 Kg4h5)
-10.50/22 1} 56.a5 Kxd5 {(Kd6xd5 Ka7b6 Kd5d6 a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q Ke6f6 Kf6e6
Kb6c5 Ke6e5 Ke5e4 Ke4f4) -10.50/20 1} 57.Kb6 Kd6 {(Kd5d6 a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q
Ke6f6 Kf6e6 Ke6e5) -10.48/11} 58.a6 Kd5 {(Kd6d5 a6a7 Kd5e6 Kb6c5 Ke6d7 a7a8Q Kd7e6
Ke6f5 Kf5e4 Ke4e3 Ke3f4) -10.50/16} 59.a7 Ke6 {(Kd5e6 a7a8Q Ke6f6 Kf6e6 Ke6e5 Kb6c5
Ke5e4 Ke4e3 Ke3e4) -10.50/14} 60.a8=Q Kf6 {(Ke6f6 Qa8f3+ Kf6g5 Kb6c5 Kg5g6 Qf3g4+
Kg6f6 Kc5d4 Kf6f7 Kd4e5 Kf7f8 Qg4g6 Kf8e7 Qg6g7+ Ke7e8 Ke5d6) -10.58/14 2} 61.Qf3+
Kg6 {(Kf6g6 Qf3g4+ Kg6f6 Kb6c5 Kf6e5 Qg4g5+ Ke5e6 Kc5d4 Ke6d7 Qg5g6 Kd7d8 Kd4c5
Kd8d7 Qg6d6+) -10.57/12} 62.Qg4+ Kf6 {(Kg6f6 Kb6c5 Kf6e5 Qg4g6 Ke5f4 Kc5d4 Kf4f3
Qg6e4+ Kf3f2 Qe4e3+ Kf2g2 Kd4e4 Kg2f1 Qe3d2) -10.61/15 6} 63.Kc5 Kf7 {(Kf6f7 Qg4f5+
Kf7g7 Qf5g5+ Kg7f8 Kc5d5 Kf8f7 Kd5d6 Kf7f8 Qg5e7+ Kf8g8 Kd6d7 Kg8h8) -10.61/12 4}
64.Qf5+ Kg8 {(Kf7g8 Qf5g5+ Kg8h8 Kc5d6 Kh8h7) -109.83/11 5} 65.Qg5+ Kf8 {(Kg8f8 Kc5d4
Kf8f7 Kd4d5 Kf7e8 Qg5g8+ Ke8d7 Qg8g7+ Kd7d8 Kd5d6 Kd8e8) -109.82/11 3} 66.Kd4 Ke8
{(Kf8e8 Qg5g7 Ke8d8 Kd4d5 Kd8e8 Kd5d6 Ke8d8) -109.96/10} 67.Qg7 Kd8 {(Ke8d8) -10.47/1}
68.Kd5 Kc8 {(Kd8c8 Kd5d6 Kc8b8) -109.98/4} 69.Kd6 Kb8 {(Kc8b8 Kd6c6 Kb8a8) -109.99/5}
70.Kc6 Kc8 {(Kb8c8) -109.99/2} 71.Qf8# {Mate} 1-0

[Dirt]

I had Telepath play without pondering and I always played the move from its PV.

I think you were quite lucky. Weren't you playing at a disadvantage of one ply?

[bob]

Not really, I cheated.

This comesfrom a discussion that Bob, Zach and I had in October.
I had Telepath play without pondering and I always played the
move from its PV. I believe theory suggests that I can win lose or
draw doing this. I don't know what the odds are on this but it sounds
like an interesting experiment to code.

I have played 3 games in this manner (all as white). I won 2 and
lost 1. Here is one of the games.

Here's my thinking. You are copying the PV from ply 2 on for your move. So you are searching exactly one ply shallower than Telepath each and every move. This is worth somewhere in the 60-100 Elo range depending on the program and the depth. Which means that you should win a game reasonably often. If you were using a 2-ply less deep search, which would be around 200 Elo optimistically, then you should still win one of every four games approximately. Just being down less than 100 Elo means something like 1.5-2.5 for a 4 game match, fairly close. Probably closer than one would first expect.

[Event "Computer chess game"]
[Site "HEPHAESTUS"]
[Date "2009.01.11"]
[Round "?"]
[White "Owner"]
[Black "Telepath Version 5.014"]
[Result "1-0"]
[ECO "C47"]
[Opening "Four Knights"]
[Time "20:08:01"]
[Variation "Scotch, Main Line, 7...d5"]
[WhiteElo "2350"]
[TimeControl "180+2"]
[Termination "normal"]
[PlyCount "141"]
[WhiteType "human"]
[BlackType "program"]

1.e4 e5 2.Nf3 Nc6 3.d4 exd4 4.Nxd4 Nf6 5.Nc3 Bb4 6.Nxc6 bxc6 7.Bd3 d5 8.e5 Qe7
{(Qd8e7 00 Nf6g4 Bc1f4 Bb4xc3 b2xc3 g7g5 Bf4g3 Ng4xe5 Rf1e1) +0.17/11 7} 9.O-O Ng4
{(Nf6g4 Bc1f4 Bb4xc3 b2xc3 g7g5 Bf4g3 Ng4xe5 Rf1e1 f7f6) -0.07/11 16} 10.Bf4 O-O {(00
Qd1e2 Rf8e8 Ra1e1 Qe7c5 h2h3 Qc5d4 g2g3 Bb4xc3) -0.06/10 8} 11.Qe2 g5 {(g7g5 Bf4g3
Rf8e8 f2f4 Bb4xc3 b2xc3 g5xf4 Rf1xf4 Ng4xe5) +0.06/10 9} 12.Bg3 h5 {(h7h5 h2h3 h5h4
Bg3xh4 g5xh4 h3xg4 Qe7g5 f2f3 Bb4c5+ Kg1h1) -0.30/10 7} 13.h3 h4 {(h5h4 Bg3xh4 g5xh4
h3xg4 Qe7g5 Bd3f5 Bc8xf5 g4xf5 Qg5xf5 Nc3a4 h4h3) -0.29/12 8} 14.Bxh4 gxh4 {(g5xh4
h3xg4 Qe7g5 Bd3f5 Bc8xf5 g4xf5 Qg5xf5 Nc3a4 h4h3 f2f4 Ra8b8) -0.28/12 9} 15.hxg4 Qg5
{(Qe7g5 Bd3f5 Bc8xf5 g4xf5 Bb4xc3 b2xc3 Qg5xf5 f2f4 Rf8e8 Ra1b1) -0.24/12 15} 16.Bf5
Bxf5 {(Bc8xf5 g4xf5 Qg5xf5 Nc3a4 h4h3 f2f4 Ra8b8 Qe2d3 Qf5xd3 c2xd3 h3xg2) -0.29/11
3} 17.gxf5 Qxf5 {(Qg5xf5 f2f4 Bb4c5+ Kg1h1 Qf5g6 Nc3a4 Bc5d4 c2c3 Bd4b6 Na4xb6)
-0.03/10 7} 18.f4 Bc5+ {(Bb4c5+ Kg1h1 Rf8b8 Nc3a4 Bc5e7 Qe2d3 Qf5g4 Kh1g1 Kg8h8 Ra1d1)
-0.21/10 7} 19.Kh1 Kh8 {(Kg8h8 Nc3a4 Bc5b6 Na4xb6 a7xb6 Qe2d3 Qf5g4 Qd3f3 Qg4f5 Qf3h3)
-0.04/11 8} 20.Na4 Bb6 {(Bc5b6 Qe2d3 Qf5g4 Na4xb6 a7xb6 Qd3f3 Qg4f5 Qf3c3 c6c5 Kh1g1
Rf8g8) -0.16/12 8} 21.Qd3 Qg4 {(Qf5g4 Qd3f3 Rf8g8 Na4xb6 c7xb6 Qf3xg4 Rg8xg4 Rf1f2
Ra8g8 Ra1d1 Rg4g3) -0.13/12 6} 22.Qf3 Rfg8 {(Rf8g8 Qf3xg4 Rg8xg4 Na4xb6 c7xb6 Rf1f2
Ra8g8 Ra1d1 Rg4g3 Kh1h2 a7a5) -0.08/13 10} 23.Qxg4 Rxg4 {(Rg8xg4 Na4xb6 c7xb6 Rf1f2
Ra8g8 Kh1h2 Rg4g3 Ra1d1 a7a5 Kh2g1) -0.05/11} 24.Nxb6 cxb6 {(c7xb6 Ra1e1 Ra8e8 Rf1f3
Re8g8 Re1e2 Kh8h7 Kh1g1 Rg4g3 Rf3d3 h4h3) -0.14/12 7} 25.Rae1 Rag8 {(Ra8g8 Rf1f2
Rg8e8 Kh1h2 Re8e6 Re1d1 Kh8g7 b2b3 Rg4g3 c2c4 d5xc4) -0.29/12 5} 26.Rf2 Re8 {(Rg8e8
Kh1g1 Rg4g3 Re1d1 Kh8g7 Rd1d3 Kg7h6 Rf2e2 a7a5 a2a4 Re8g8) -0.13/12 6} 27.Kg1 Rg3
{(Rg4g3 Kg1h2 Kh8g8 f4f5 Rg3g4 e5e6 f7xe6 Rf2e2 e6e5 Re2xe5 Re8xe5) -0.37/12 5} 28.Kh2
Kg8 {(Kh8g8 Rf2f3 Rg3g6 Kh2h3 Kg8f8 Re1d1 Rg6h6 Rf3c3 a7a5 a2a4 Re8e7) -0.37/12 5}
29.Rf3 Rg6 {(Rg3g6 Kh2h3 Kg8f8 f4f5 Rg6h6 e5e6 c6c5 Rf3e3 Re8e7 e6xf7 Re7xe3+ Re1xe3)
-0.53/12 5} 30.Kh3 Kf8 {(Kg8f8 f4f5 Rg6g5 e5e6 Re8e7 Rf3f2 Rg5h5 Re1f1 a7a5 e6xf7
Kf8xf7) -0.67/11 4} 31.f5 Rg5 {(Rg6g5 e5e6 a7a5 Rf3f2 Rg5h5 Rf2f4 Re8e7 Re1e5 Rh5g5
Rf4xh4) -0.70/12 7} 32.e6 Rg3+ {(Rg5g3+ Rf3xg3 h4xg3 Kh3g4 f7xe6 Re1xe6 Re8xe6 f5xe6
Kf8e7 Kg4f5 c6c5 e6f7) -0.37/11 4} 33.Rxg3 hxg3 {(h4xg3 Kh3g4 f7xe6 Re1xe6 Re8xe6
f5xe6 Kf8e7 Kg4f5 a7a5 Kf5e5 a5a4 c2c3) -0.38/14 7} 34.Kxg3 fxe6 {(f7xe6 a2a4 e6xf5
Re1xe8+ Kf8xe8 Kg3f4 a7a6 Kf4xf5 b6b5 a4xb5 a6xb5 g2g4 Ke8f7) -0.13/13 4} 35.a4 exf5
{(e6xf5 Re1xe8+ Kf8xe8 Kg3f4 a7a6 Kf4xf5 b6b5 a4a5 Ke8f7 g2g4 c6c5 g4g5) -0.07/12 2}
36.Rxe8+ Kxe8 {(Kf8xe8 Kg3f4 a7a6 Kf4xf5 Ke8f7 g2g4 b6b5 g4g5 b5xa4 g5g6+ Kf7g7)
-0.14/11} 37.Kf4 d4 {(d5d4 b2b4 c6c5 b4xc5 b6xc5 Kf4xf5 c5c4 Kf5e4 d4d3 c2xd3 c4c3 Ke4e3
Ke8d7 g2g4 Kd7e6) -0.77/16 5} 38.b4 Ke7 {(Ke8e7 Kf4xf5 Ke7d6 g2g4 c6c5 b4xc5+ b6xc5
Kf5e4 Kd6e6 g4g5 a7a5 g5g6 Ke6f6 Ke4d5 Kf6xg6 Kd5xc5) -0.97/17 4} 39.Kxf5 Kd6 {(Ke7d6
g2g4 a7a5 b4xa5 b6xa5 g4g5 c6c5 g5g6 Kd6e7 Kf5e5 Ke7f8) -1.06/17 5} 40.g4 c5 {(c6c5
b4xc5+ b6xc5 a4a5 Kd6d5 g4g5 Kd5d6 g5g6 Kd6e7 Kf5e5 Ke7f8 Ke5d5 Kf8g7 Kd5xc5 Kg7xg6
a5a6) -1.11/16 3} 41.bxc5+ bxc5 {(b6xc5 g4g5 c5c4 Kf5e4 d4d3 c2xd3 c4c3 Ke4e3 Kd6e5
d3d4+ Ke5f5 Ke3d3 Kf5xg5 d4d5 Kg5f5 Kd3xc3 Kf5e5) -1.36/17 7} 42.g5 c4 {(c5c4 Kf5e4
d4d3 c2xd3 c4c3 Ke4e3 Kd6e5 d3d4+ Ke5f5 Ke3d3 Kf5xg5 Kd3xc3 Kg5f5 Kc3c4 Kf5e4 d4d5)
-1.36/17 6} 43.Ke4 d3 {(d4d3 c2xd3 c4c3 Ke4e3 Kd6e5 d3d4+ Ke5f5 Ke3d3 Kf5xg5 Kd3xc3
Kg5f5 Kc3c4 Kf5e4 d4d5 Ke4e5 a4a5 Ke5d6 Kc4d4 a7a6) -1.46/19 4} 44.cxd3 c3 {(c4c3
Ke4e3 Kd6e5 d3d4+ Ke5f5 Ke3d3 Kf5xg5 Kd3xc3 Kg5f5 Kc3c4 Kf5e4 Kc4c5 Ke4f4 d4d5 Kf4e5
d5d6 Ke5e6) -1.80/18 3} 45.Ke3 Ke5 {(Kd6e5 d3d4+ Ke5f5 Ke3d3 c3c2 Kd3xc2 Kf5xg5 a4a5
Kg5f5 Kc2d3 Kf5f4 d4d5 Kf4e5 Kd3c4 Ke5d6 a5a6) -1.46/16} 46.d4+ Kf5 {(Ke5f5 Ke3d3
c3c2 Kd3xc2 Kf5xg5 Kc2d3 Kg5f5 Kd3c4 Kf5e4 d4d5 Ke4e5 Kc4c5 Ke5f5 a4a5 Kf5e5 d5d6)
-1.80/17} 47.Kd3 c2 {(c3c2 Kd3xc2 Kf5xg5 Kc2d3 Kg5f6 Kd3c4 a7a6 Kc4c5 Kf6f5 a4a5 Kf5e4
Kc5b6 Ke4xd4) -2.48/18 1} 48.Kxc2 Kxg5 {(Kf5xg5 Kc2d3 Kg5f5 Kd3c4 Kf5e6 a4a5 Ke6d6
Kc4b5 Kd6c7 Kb5a6 Kc7b8 d4d5 Kb8a8 d5d6 Ka8b8 d6d7 Kb8c7 Ka6xa7 Kc7xd7) -2.71/20 1}
49.Kd3 Kf5 {(Kg5f5 Kd3c4 Kf5e6 Kc4c5 Ke6d7 Kc5b5 Kd7c7 Kb5a6 Kc7b8 a4a5 Kb8a8 d4d5
Ka8b8 d5d6 Kb8c8 Ka6xa7 Kc8d7 Ka7b6 Kd7e6 a5a6 Ke6xd6 a6a7 Kd6e6) -10.21/23 3} 50.Kc4
Ke6 {(Kf5e6 Kc4c5 Ke6d7 Kc5b5 Kd7c7 Kb5a6 Kc7b8 d4d5 Kb8a8 a4a5) -10.21/19} 51.Kc5
Kd7 {(Ke6d7 Kc5b5 Kd7c7 Kb5a6 Kc7b8 d4d5 Kb8a8 d5d6 Ka8b8 a4a5 Kb8c8 Ka6xa7 Kc8d7
Ka7b6 Kd7xd6 a5a6 Kd6e5 a6a7 Ke5f5 a7a8Q Kf5g6) -10.23/23 3} 52.Kb5 Kc7 {(Kd7c7 Kb5a6
Kc7b8 d4d5 Kb8a8 d5d6 Ka8b8 a4a5 Kb8c8 Ka6xa7 Kc8d7 Ka7b6 Kd7xd6 a5a6 Kd6d5 a6a7
Kd5e6 a7a8Q Ke6f6 Kb6c5 Kf6g6 Kg6g5 Kc5d4 Kg5f6) -10.42/25 2} 53.Ka6 Kb8 {(Kc7b8 d4d5
Kb8a8 d5d6 Ka8b8 a4a5 Kb8c8 Ka6xa7 Kc8d7 Ka7b6 Kd7xd6 a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q
Ke6f6 Kf6e7 Kb6c5 Ke7f7 Kf7f6 Kc5d4 Kf6f7) -10.43/26 3} 54.d5 Kc7 {(Kb8c7 Ka6xa7
Kc7d6 a4a5 Kd6xd5 Ka7b6 Kd5d6 a5a6 Kd6d5 a6a7 Kd5c4 a7a8Q Kc4d4 Kd4e4 Kb6c5+ Ke4f5
Kf5f6 Kc5d4 Kf6g7 Kg7f7) -10.47/23 2} 55.Kxa7 Kd6 {(Kc7d6 a4a5 Kd6xd5 Ka7b6 Kd5d6
a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q Ke6f6 Kf6e6 Kb6c5 Ke6e5 Ke5e4 Ke4f4 Kf4g4 Kc5d5 Kg4h5)
-10.50/22 1} 56.a5 Kxd5 {(Kd6xd5 Ka7b6 Kd5d6 a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q Ke6f6 Kf6e6
Kb6c5 Ke6e5 Ke5e4 Ke4f4) -10.50/20 1} 57.Kb6 Kd6 {(Kd5d6 a5a6 Kd6d5 a6a7 Kd5e6 a7a8Q
Ke6f6 Kf6e6 Ke6e5) -10.48/11} 58.a6 Kd5 {(Kd6d5 a6a7 Kd5e6 Kb6c5 Ke6d7 a7a8Q Kd7e6
Ke6f5 Kf5e4 Ke4e3 Ke3f4) -10.50/16} 59.a7 Ke6 {(Kd5e6 a7a8Q Ke6f6 Kf6e6 Ke6e5 Kb6c5
Ke5e4 Ke4e3 Ke3e4) -10.50/14} 60.a8=Q Kf6 {(Ke6f6 Qa8f3+ Kf6g5 Kb6c5 Kg5g6 Qf3g4+
Kg6f6 Kc5d4 Kf6f7 Kd4e5 Kf7f8 Qg4g6 Kf8e7 Qg6g7+ Ke7e8 Ke5d6) -10.58/14 2} 61.Qf3+
Kg6 {(Kf6g6 Qf3g4+ Kg6f6 Kb6c5 Kf6e5 Qg4g5+ Ke5e6 Kc5d4 Ke6d7 Qg5g6 Kd7d8 Kd4c5
Kd8d7 Qg6d6+) -10.57/12} 62.Qg4+ Kf6 {(Kg6f6 Kb6c5 Kf6e5 Qg4g6 Ke5f4 Kc5d4 Kf4f3
Qg6e4+ Kf3f2 Qe4e3+ Kf2g2 Kd4e4 Kg2f1 Qe3d2) -10.61/15 6} 63.Kc5 Kf7 {(Kf6f7 Qg4f5+
Kf7g7 Qf5g5+ Kg7f8 Kc5d5 Kf8f7 Kd5d6 Kf7f8 Qg5e7+ Kf8g8 Kd6d7 Kg8h8) -10.61/12 4}
64.Qf5+ Kg8 {(Kf7g8 Qf5g5+ Kg8h8 Kc5d6 Kh8h7) -109.83/11 5} 65.Qg5+ Kf8 {(Kg8f8 Kc5d4
Kf8f7 Kd4d5 Kf7e8 Qg5g8+ Ke8d7 Qg8g7+ Kd7d8 Kd5d6 Kd8e8) -109.82/11 3} 66.Kd4 Ke8
{(Kf8e8 Qg5g7 Ke8d8 Kd4d5 Kd8e8 Kd5d6 Ke8d8) -109.96/10} 67.Qg7 Kd8 {(Ke8d8) -10.47/1}
68.Kd5 Kc8 {(Kd8c8 Kd5d6 Kc8b8) -109.98/4} 69.Kd6 Kb8 {(Kc8b8 Kd6c6 Kb8a8) -109.99/5}
70.Kc6 Kc8 {(Kb8c8) -109.99/2} 71.Qf8# {Mate} 1-0

[bob]

I had Telepath play without pondering and I always played the move from its PV.

I think you were quite lucky. Weren't you playing at a disadvantage of one ply?

yes, but that is less than 100 Elo. so you should actually win about 1.5 games out of every 4...

[CRoberson]

Not really, I cheated.

This comesfrom a discussion that Bob, Zach and I had in October.
I had Telepath play without pondering and I always played the
move from its PV. I believe theory suggests that I can win lose or
draw doing this. I don't know what the odds are on this but it sounds
like an interesting experiment to code.

I have played 3 games in this manner (all as white). I won 2 and
lost 1. Here is one of the games.

Here's my thinking. You are copying the PV from ply 2 on for your move. So you are searching exactly one ply shallower than Telepath each and every move. This is worth somewhere in the 60-100 Elo range depending on the program and the depth. Which means that you should win a game reasonably often. If you were using a 2-ply less deep search, which would be around 200 Elo optimistically, then you should still win one of every four games approximately. Just being down less than 100 Elo means something like 1.5-2.5 for a 4 game match, fairly close. Probably closer than one would first expect.

That was exactly my reasoning which is why I ran the experiment.

The bad thing is how this idea could be misused.
When paired with a program (for example) 400 pts above your
program, you could switch to this algorithm to increase your odds.

Makes you wonder if we should keep the kibitz PV rule in tournaments.
I suppose changing it to must kibitz even numbered
moves would work. Of course, that is assuming we think we are
seeing the problem.

[Dirt]

I had Telepath play without pondering and I always played the move from its PV.

I think you were quite lucky. Weren't you playing at a disadvantage of one ply?

yes, but that is less than 100 Elo. so you should actually win about 1.5 games out of every 4...

I don't think you're including the magnification of the Elo difference you get by playing one program against a modified version of itself.

[CRoberson]

I had Telepath play without pondering and I always played the move from its PV.

I think you were quite lucky. Weren't you playing at a disadvantage of one ply?

yes, but that is less than 100 Elo. so you should actually win about 1.5 games out of every 4...

I don't think you're including the magnification of the Elo difference you get by playing one program against a modified version of itself.

What published research are you referring to?

[Dirt]

What published research are you referring to?

I'm just relying on what engine authors have said about testing on this board. If you have information refuting this, published or not, I'd be interested in seeing it.

[bob]

I had Telepath play without pondering and I always played the move from its PV.

I think you were quite lucky. Weren't you playing at a disadvantage of one ply?

yes, but that is less than 100 Elo. so you should actually win about 1.5 games out of every 4...

I don't think you're including the magnification of the Elo difference you get by playing one program against a modified version of itself.

I was actually taking that into account. We have had many such experiments done and reported in the literature, Ken Thompson and Belle being the first I remember...

I've run quite a few of these kinds of experiments. Today, one ply is about a factor of 2x in speed. A factor of 2x is right at 100Elo, and I just posted some results in the SMP efficiency thread that supports this pretty well...

[bob]

Not really, I cheated.

This comesfrom a discussion that Bob, Zach and I had in October.
I had Telepath play without pondering and I always played the
move from its PV. I believe theory suggests that I can win lose or
draw doing this. I don't know what the odds are on this but it sounds
like an interesting experiment to code.

I have played 3 games in this manner (all as white). I won 2 and
lost 1. Here is one of the games.

Here's my thinking. You are copying the PV from ply 2 on for your move. So you are searching exactly one ply shallower than Telepath each and every move. This is worth somewhere in the 60-100 Elo range depending on the program and the depth. Which means that you should win a game reasonably often. If you were using a 2-ply less deep search, which would be around 200 Elo optimistically, then you should still win one of every four games approximately. Just being down less than 100 Elo means something like 1.5-2.5 for a 4 game match, fairly close. Probably closer than one would first expect.

That was exactly my reasoning which is why I ran the experiment.

The bad thing is how this idea could be misused.
When paired with a program (for example) 400 pts above your
program, you could switch to this algorithm to increase your odds.

Makes you wonder if we should keep the kibitz PV rule in tournaments.
I suppose changing it to must kibitz even numbered
moves would work. Of course, that is assuming we think we are
seeing the problem.

Its an issue. We could use whisper, but an observer could feed things to the program, or the program could create a second thread, log on to ICC under a different handle, and observe its own game.