Some handicap results and conclusions.

[lkaufman]

I don't know why I am posting this, Larry might be angry, but the doubled g3 pawn is so typical of Komodo, that I thought it is worth a post.

here is a game between Komodo 10.1 as white and SF:

[pgn][Event "OWNER-PC, Blitz 1m"]
[Site "Microsoft"]
[Date "2017.06.07"]
[Round "226"]
[White "Komodo 10.1 64-bit"]
[Black "Stockfish 8 64 POPCNT"]
[Result "0-1"]
[ECO "D35"]
[Annotator "0.27;0.29"]
[PlyCount "126"]
[TimeControl "60"]

{Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz 1729 MHz W=15.1 plies; 1
191kN/s; CM8000.ctg B=17.3 plies; 1 234kN/s; CM8000.ctg} 1. d4 {B 0} Nf6 {B 0
} 2. c4 {B 0} e6 {B 0 Both last book move} 3. Nf3 {0.27/16 2} d5 {0.29/19 2} 4.
Nc3 {0.25/17 1} Be7 {0.40/18 2 (c5)} 5. cxd5 {0.39/17 4 (Bf4)} exd5 {0.12/17 1}
6. Bf4 {0.38/18 1 (e3)} O-O {0.26/18 2} 7. e3 {0.34/18 1 (h3)} Nh5 {0.12/18 1
(Bf5)} 8. Be5 {0.48/16 1} f6 {0.16/19 1} 9. Bg3 {0.37/17 1} c6 {0.24/17 0 (g6)}
10. Bd3 {0.42/17 1} g6 {0.14/17 1} 11. O-O {0.26/17 2 (e4)} Nxg3 {-0.18/17 1
(Bg4)} 12. hxg3 {0.38/18 1} f5 {-0.26/16 0 (Bg4)} 13. Rb1 {0.44/15 1} a6 {-0.
11/20 2 (Nd7)} 14. Qb3 {0.47/16 1} a5 {-0.20/18 1 (Nd7)} 15. Rbc1 {0.46/15 1
(Ne5)} Nd7 {-0.25/18 1} 16. Ne5 {0.31/16 2} Nf6 {-0.46/20 3 (Kg7)} 17. Na4 {0.
30/14 1} Bd6 {-0.43/17 0 (Kg7)} 18. Rfd1 {0.20/14 2} Re8 {-0.48/18 1 (Ne4)} 19.
a3 {0.09/16 3 (Nf3)} b5 {-0.72/18 1 (Rb8)} 20. Nc5 {0.28/16 1} Bxe5 {-0.55/19
0 (a4)} 21. dxe5 {0.28/18 1} Rxe5 {0.00/21 3 (a4)} 22. Qc3 {0.23/16 1} Re8 {-0.
27/22 1} 23. b4 {0.24/17 1 (Nb3)} a4 {-0.51/17 1 (Ne4)} 24. Be2 {0.19/14 1} Qe7
{-0.37/15 0 (Nd7)} 25. Rd4 {0.25/15 1 (Qd4)} h5 {-0.87/17 1 (Ra7)} 26. Bd1 {0.
21/16 1 (Rd2)} Rb8 {-1.01/18 1} 27. Qc2 {0.14/17 1 (Be2)} Rb6 {-0.98/19 1 (Nd7)
} 28. Bf3 {0.14/19 1 (Be2)} Ne4 {-1.04/19 1} 29. Qa2 {0.13/20 1 (Rdd1)} Qf7 {
-1.08/20 3 (Nxc5)} 30. Be2 {0.13/21 1 (Qc2)} Nd6 {-1.08/21 1 (Nxc5)} 31. Bf3 {
0.13/19 1 (Qb2)} Be6 {-1.18/20 0 (Ne4)} 32. Qe2 {0.13/15 1 (Be2)} Re7 {-1.32/
18 1 (Ne4)} 33. Qd1 {0.13/15 1 (Rc2)} Qf6 {-1.39/18 1 (Re8)} 34. Qd2 {0.13/18
1 (Be2)} Bf7 {-1.39/20 1 (Re8)} 35. Rd1 {0.12/15 1 (Be2)} g5 {-1.33/22 1 (Be6)}
36. Be2 {0.05/15 1} Bg6 {-1.32/22 3 (Ne4)} 37. Rc1 {0.00/15 1} Rb8 {-1.52/19 1
(Ne4)} 38. Na6 {0.00/17 1 (Qb2)} Rc8 {-1.58/21 0 (Rb6)} 39. Nc5 {-0.37/16 2
(Qc3)} Rg7 {-1.53/19 1 (Rce8)} 40. Bd3 {0.00/17 1 (Qc3)} Rh7 {-1.70/16 1 (Re7)}
41. Qd1 {-0.21/16 1 (Be2)} g4 {-1.92/16 0 (Rf8)} 42. Rf4 {-0.27/15 0} Qe7 {-1.
45/17 1 (Rd8)} 43. Qc2 {-0.65/14 0 (Bb1)} Rf8 {-1.91/16 1 (h4)} 44. Qc3 {-0.60/
14 1} h4 {-1.87/17 1} 45. gxh4 {-0.75/15 1 (Re1)} Qxh4 {-2.73/16 0 (Rxh4)} 46.
Kf1 {-0.50/15 0} d4 {-2.92/15 0 (Qe7)} 47. Qd2 {-2.15/14 1 (Qe1)} dxe3 {-3.65/
16 0} 48. fxe3 {-2.08/15 0} Re8 {-3.62/17 2} 49. Rd1 {-2.23/13 1 (Ke2)} Qg5 {
-4.81/15 0 (Qh1+)} 50. g3 {-2.55/12 0 (Bb1)} Rh3 {-4.23/16 1} 51. Bc2 {-3.10/
13 0 (Be2)} Nc4 {-5.89/16 0} 52. Rxc4 {-3.28/14 0} bxc4 {-5.77/14 0} 53. Bxa4 {
-3.41/14 0} Qh6 {-6.33/14 0 (Rh1+)} 54. Bxc6 {-3.24/12 0} Rxg3 {-6.38/12 0} 55.
Re1 {-3.89/12 1 (Bg2)} Qh3+ {-7.35/14 0 (Bf7)} 56. Bg2 {-4.23/11 0} Rf3+ {-6.
69/13 0} 57. Kg1 {-4.68/11 0} Qg3 {-8.29/16 1} 58. Rf1 {-4.83/12 0 (Ra1)} Rexe3
{-9.46/14 0 (Rxf1+)} 59. Kh1 {-5.79/11 0 (Qd1)} Rxf1+ {-10.59/14 0} 60. Bxf1 {
-5.79/5 0} Qf4 {-10.84/14 0 (Re1)} 61. Qd1 {-6.93/12 0} Bf7 {-13.08/15 0} 62.
b5 {-7.84/12 0} Kg7 {-20.71/15 0 (Rxa3)} 63. Bg2 {-10.83/11 0} Qf2 {-24.45/13 0
} 0-1

[/pgn]

[d]rnbq1rk1/pp2b2p/2p3p1/3p1p2/3P4/2NBPNP1/PP3PP1/R2Q1RK1 w - - 0 13

SF 20cps black advantage, Komodo 40cps white edge.

quite probably, white is already lost.

see how SF fixes the g3 doubled shelter weakness, and then h5-h4 uses the weakness to open lines for decisive attack.

I guess tuning engines against only their predecessors is a bit shaky concept, as Komodo playing against its predecessor might never utilise effectively the g3 weakness, no matter the statistically significant number of games.

but then, the doubled g3 shelter pawn is maybe Komodo's pet flaw.

On my laptop, current Komodo takes only a few seconds to show a zero score. The search depth of 15 you show is extremely low, even for a bullet game; do you have a very old or even 32 bit computer? My laptop gets to 20 ply in a couple seconds in this position. Aside from that, the more interesting question in this game is the pawn sacrifice on e5 played by Komodo; although it gets fair comp, I think most GMs would not offer that pawn. Maybe another example of Komodo's generosity. As for the g2/g3 pawns, you are correct that Komodo values them more for defense than do other engines, I think correctly so. they do a great job of keeping enemy pieces out, True, they provide a "hook", but you can't have everything. Perhaps we should find a way to lower the value of the second pawn a bit.

the position is simply lost, so score is more like 60-70cps black advantage.

whatever my pc, both SF and Komodo are using it.

g2/g3 doubled pair is simply no good, when the pawns are within the king shelter.

with b2/b3 combination, this is just fine.

problem is related exclusively to king safety.

do you distinguish between such pawns, when they are within the shelter and otherwise, or just have a general psqt for pawns/doubled pawns?

Well, shelter pawns have their own rules, but I don't think we have any extra penalty specifically for doubled pawns in the shelter. I've never thought of that as a valid general principle (I'm not talking about doubled isolated shelter pawns of course). Do you think White would be much better off in this position if we move the g3 pawn to h3? It still provides a hook for an attack. Doubled pawns are generally ok for defense, just not useful for offense.

[lkaufman]

another one.

this probably has to do with depth and move ordering/general search.

[pgn][Event "OWNER-PC, Blitz 1m"]
[Site "Microsoft"]
[Date "2017.06.12"]
[Round "3"]
[White "Stockfish 8 64 POPCNT"]
[Black "Komodo 10.1 64-bit"]
[Result "1-0"]
[ECO "C01"]
[Annotator "0.34;0.09"]
[PlyCount "99"]
[TimeControl "60"]

{Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz 1729 MHz W=17.7 plies; 1
189kN/s; Empty.ctg B=16.9 plies; 1 296kN/s; Empty.ctg} 1. e4 {0.34/19 6} e6 {
0.09/16 2} 2. d4 {0.32/19 1} d5 {0.20/16 1} 3. Nc3 {0.05/20 3} Bb4 {0.17/17 2}
4. exd5 {0.20/19 1} exd5 {0.20/17 1} 5. Bd3 {0.12/19 1} Nc6 {0.21/17 1 (Nf6)}
6. Nf3 {0.24/19 2} Nf6 {0.25/18 3} 7. O-O {0.27/16 0 (a3)} a6 {0.16/17 1 (0-0)}
8. Re1+ {0.43/18 1 (a3)} Be7 {0.32/16 1} 9. a3 {0.35/20 2 (Bf4)} O-O {0.26/16 0
} 10. h3 {0.31/17 0} Bd6 {0.21/18 1 (h6)} 11. Bg5 {0.43/18 1} Be6 {0.24/17 1}
12. Ne2 {0.46/17 0 (Qd2)} h6 {0.07/14 0} 13. Bh4 {0.78/19 4 (Bf4)} g5 {0.02/15
1 (Rb8)} 14. Nxg5 {1.20/17 1 (Bg3)} hxg5 {0.00/16 1} 15. Bxg5 {0.66/16 0} Re8 {
0.06/17 2} 16. c3 {0.87/18 1} Bd7 {0.17/16 2} 17. Ng3 {0.83/21 1} Rxe1+ {0.32/
16 1} 18. Qxe1 {0.91/20 0} Qf8 {0.33/17 1 (Qe7)} 19. Qe3 {0.41/19 2 (Bxf6)} Ne8
{0.33/16 1} 20. Nh5 {0.67/19 1 (Qf3)} f5 {0.28/16 1} 21. Bf4 {0.71/19 3 (Qf3)}
Qf7 {0.21/16 1 (Be7)} 22. Qg3+ {0.68/17 1} Ng7 {0.31/18 1} 23. Be2 {1.03/19 2
(Bxd6)} Re8 {0.00/18 1} 24. Bxd6 {0.80/22 2 (Bf3)} Rxe2 {-0.18/16 0} 25. Nf4 {
0.97/19 0} Rxb2 {0.39/16 2 (Rc2)} 26. Bxc7 {1.47/16 1} Rb3 {0.32/16 1 (Be8)}
27. Re1 {1.76/16 0} Nxd4 {0.51/17 1 (a5)} 28. Be5 {2.42/20 2} Nc6 {0.38/17 0}
29. Qg5 {2.20/18 0 (Nxd5)} Rxa3 {0.42/15 1} 30. Qh6 {2.54/20 1 (Bf6)} Nxe5 {0.
37/16 1 (Ra1)} 31. Rxe5 {3.37/16 0} Ne8 {0.43/18 1 (Bc6)} 32. Re3 {4.11/16 0
(Nxd5)} Qg7 {3.13/14 1} 33. Rg3 {4.53/15 0} Ra1+ {3.63/16 1} 34. Kh2 {4.53/1 0}
Qxg3+ {3.97/18 2} 35. fxg3 {4.88/16 0} Re1 {4.01/18 1} 36. Nxd5 {4.96/16 0} a5
{4.11/18 2 (Be6)} 37. Qg6+ {5.01/20 1} Kf8 {4.09/18 0} 38. Qh7 {5.10/19 0} Ng7
{4.36/19 0} 39. Qh8+ {5.36/19 0} Kf7 {4.36/5 0} 40. Qd8 {5.55/19 0} Bc6 {4.30/
22 1} 41. Qf6+ {5.58/17 0} Kg8 {4.20/23 0 (Ke8)} 42. Ne7+ {5.90/16 0} Rxe7 {4.
13/24 1} 43. Qxe7 {6.35/18 0} a4 {4.59/23 1} 44. h4 {6.44/16 0 (Kg1)} Kh7 {4.
58/22 1} 45. Kh3 {6.91/18 1 (Qf7)} Kg8 {4.84/17 1 (Kh6)} 46. g4 {7.88/14 0
(Qb4)} fxg4+ {6.26/15 1} 47. Kxg4 {8.36/15 0} a3 {6.59/16 1 (Be8)} 48. Qxa3 {
8.88/15 0} Kf7 {6.67/16 1} 49. h5 {9.44/14 0 (Qa2+)} Ne8 {6.85/13 0 (Ke6)} 50.
Kg5 {9.87/14 0} 1-0

[/pgn]

[d]r2q1rk1/1pp2p2/p1nbbn2/3p2B1/3P4/P2B3P/1PP1NPP1/R2QR1K1 b - - 0 15

SF 70cps white edge, Komodo 0.0.

this is quite indicative.

most SF wins are like this.

Here in the diagram position current Komodo on my laptop takes less than a second to see a solid White advantage, so again the problem is some combination of the old Komodo version, your hardware, and the bullet tc. Because of your slow hardware, I think bullet games are too fast to be useful.

same PC, same TC for both SF and Komodo.

the bullet games that are too fast to be useful are the same ones, you are using for testing, or at least have been using for testing extensively.

of course, we all know your latest dev is flawless, pity most people don't have access to it.

Actually current dev is only a couple elo stronger than latest release, but I think you are using a year-old version. Ultra-fast games are useful for the statistics they generate, but I never actually look at the games, as most bad moves would change with another quarter second or so. When Komodo plays a bad move that doesn't change after some reasonable thinking time, that interests me.

[Lyudmil Tsvetkov]

another one.

this probably has to do with depth and move ordering/general search.

[pgn][Event "OWNER-PC, Blitz 1m"]
[Site "Microsoft"]
[Date "2017.06.12"]
[Round "3"]
[White "Stockfish 8 64 POPCNT"]
[Black "Komodo 10.1 64-bit"]
[Result "1-0"]
[ECO "C01"]
[Annotator "0.34;0.09"]
[PlyCount "99"]
[TimeControl "60"]

{Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz 1729 MHz W=17.7 plies; 1
189kN/s; Empty.ctg B=16.9 plies; 1 296kN/s; Empty.ctg} 1. e4 {0.34/19 6} e6 {
0.09/16 2} 2. d4 {0.32/19 1} d5 {0.20/16 1} 3. Nc3 {0.05/20 3} Bb4 {0.17/17 2}
4. exd5 {0.20/19 1} exd5 {0.20/17 1} 5. Bd3 {0.12/19 1} Nc6 {0.21/17 1 (Nf6)}
6. Nf3 {0.24/19 2} Nf6 {0.25/18 3} 7. O-O {0.27/16 0 (a3)} a6 {0.16/17 1 (0-0)}
8. Re1+ {0.43/18 1 (a3)} Be7 {0.32/16 1} 9. a3 {0.35/20 2 (Bf4)} O-O {0.26/16 0
} 10. h3 {0.31/17 0} Bd6 {0.21/18 1 (h6)} 11. Bg5 {0.43/18 1} Be6 {0.24/17 1}
12. Ne2 {0.46/17 0 (Qd2)} h6 {0.07/14 0} 13. Bh4 {0.78/19 4 (Bf4)} g5 {0.02/15
1 (Rb8)} 14. Nxg5 {1.20/17 1 (Bg3)} hxg5 {0.00/16 1} 15. Bxg5 {0.66/16 0} Re8 {
0.06/17 2} 16. c3 {0.87/18 1} Bd7 {0.17/16 2} 17. Ng3 {0.83/21 1} Rxe1+ {0.32/
16 1} 18. Qxe1 {0.91/20 0} Qf8 {0.33/17 1 (Qe7)} 19. Qe3 {0.41/19 2 (Bxf6)} Ne8
{0.33/16 1} 20. Nh5 {0.67/19 1 (Qf3)} f5 {0.28/16 1} 21. Bf4 {0.71/19 3 (Qf3)}
Qf7 {0.21/16 1 (Be7)} 22. Qg3+ {0.68/17 1} Ng7 {0.31/18 1} 23. Be2 {1.03/19 2
(Bxd6)} Re8 {0.00/18 1} 24. Bxd6 {0.80/22 2 (Bf3)} Rxe2 {-0.18/16 0} 25. Nf4 {
0.97/19 0} Rxb2 {0.39/16 2 (Rc2)} 26. Bxc7 {1.47/16 1} Rb3 {0.32/16 1 (Be8)}
27. Re1 {1.76/16 0} Nxd4 {0.51/17 1 (a5)} 28. Be5 {2.42/20 2} Nc6 {0.38/17 0}
29. Qg5 {2.20/18 0 (Nxd5)} Rxa3 {0.42/15 1} 30. Qh6 {2.54/20 1 (Bf6)} Nxe5 {0.
37/16 1 (Ra1)} 31. Rxe5 {3.37/16 0} Ne8 {0.43/18 1 (Bc6)} 32. Re3 {4.11/16 0
(Nxd5)} Qg7 {3.13/14 1} 33. Rg3 {4.53/15 0} Ra1+ {3.63/16 1} 34. Kh2 {4.53/1 0}
Qxg3+ {3.97/18 2} 35. fxg3 {4.88/16 0} Re1 {4.01/18 1} 36. Nxd5 {4.96/16 0} a5
{4.11/18 2 (Be6)} 37. Qg6+ {5.01/20 1} Kf8 {4.09/18 0} 38. Qh7 {5.10/19 0} Ng7
{4.36/19 0} 39. Qh8+ {5.36/19 0} Kf7 {4.36/5 0} 40. Qd8 {5.55/19 0} Bc6 {4.30/
22 1} 41. Qf6+ {5.58/17 0} Kg8 {4.20/23 0 (Ke8)} 42. Ne7+ {5.90/16 0} Rxe7 {4.
13/24 1} 43. Qxe7 {6.35/18 0} a4 {4.59/23 1} 44. h4 {6.44/16 0 (Kg1)} Kh7 {4.
58/22 1} 45. Kh3 {6.91/18 1 (Qf7)} Kg8 {4.84/17 1 (Kh6)} 46. g4 {7.88/14 0
(Qb4)} fxg4+ {6.26/15 1} 47. Kxg4 {8.36/15 0} a3 {6.59/16 1 (Be8)} 48. Qxa3 {
8.88/15 0} Kf7 {6.67/16 1} 49. h5 {9.44/14 0 (Qa2+)} Ne8 {6.85/13 0 (Ke6)} 50.
Kg5 {9.87/14 0} 1-0

[/pgn]

[d]r2q1rk1/1pp2p2/p1nbbn2/3p2B1/3P4/P2B3P/1PP1NPP1/R2QR1K1 b - - 0 15

SF 70cps white edge, Komodo 0.0.

this is quite indicative.

most SF wins are like this.

Here in the diagram position current Komodo on my laptop takes less than a second to see a solid White advantage, so again the problem is some combination of the old Komodo version, your hardware, and the bullet tc. Because of your slow hardware, I think bullet games are too fast to be useful.

same PC, same TC for both SF and Komodo.

the bullet games that are too fast to be useful are the same ones, you are using for testing, or at least have been using for testing extensively.

of course, we all know your latest dev is flawless, pity most people don't have access to it.

Actually current dev is only a couple elo stronger than latest release, but I think you are using a year-old version. Ultra-fast games are useful for the statistics they generate, but I never actually look at the games, as most bad moves would change with another quarter second or so. When Komodo plays a bad move that doesn't change after some reasonable thinking time, that interests me.

I guess similarities between a year-old version and latest dev are much bigger than distinctions.

for the past year, Komodo has improved its play on features worth 50 elo or so, but the features which are still unperfected are worth some 5000 elo or so.

so, it should be easier to see what is imperfect than what has improved.

that is why I don't necessarily need every single version of every single engine.

if you are looking for moves that don't change with much bigger thinking time, than you need only LTC games, but I am afraid there are not very many like this.

btw., believe it or not, engines mostly play identically at different TCs.
for example, you might find hundreds of games in TCEC and other LTC tournaments, where Komodo has doubled g2/g3 pawns.

seemingly, nothing changes with time, maybe what concerns tactical decisions, but not what concerns principled evaluation choices.

[Nordlandia]

Komodo drew latest SF Dev in Berthelot vs Flear - 1988 position.

i7-5960X | 8-Core
4096 Mb hash per engine
Probing 5-men syzygy | 6-men adjudication
"TCEC draw rule" active.

[pgn][Event "?"]
[Site "Cutechess 1.0.0 | i7-5960X 4.1GHz"]
[Date "2017.08.19"]
[Round "?"]
[White "Stockfish_17081811_x64_bmi2"]
[Black "komodo-11.2.2-64bit"]
[Result "1/2-1/2"]
[FEN "5k2/ppp2ppp/1b6/8/6b1/2P5/PP4PP/RN2K3 w - - 0 1"]
[PlyCount "151"]
[SetUp "1"]
[TimeControl "9000+30"]

1. Nd2 {+0.90/42 234s} Bc5 {-0.62/32 581s} 2. Nf3 {+0.85/36 330s}
Bd6 {-0.58/34 349s} 3. Kf2 {+0.85/38 174s} f6 {-0.60/33 222s}
4. Ke3 {+0.79/40 306s} Ke7 {-0.60/32 434s} 5. g3 {+0.93/34 152s}
Be6 {-0.58/34 213s} 6. Kd3 {+0.77/38 307s} Bf5+ {-0.59/36 272s}
7. Ke3 {+0.82/42 826s} Be6 {-0.67/34 347s} 8. Kd3 {+0.79/43 440s}
Bf5+ {-0.58/31 95s} 9. Kd2 {+0.79/42 69s} c5 {-0.57/31 95s}
10. Re1+ {+0.80/40 188s} Kd7 {-0.56/33 154s} 11. Ke3 {+0.83/39 26s}
Be6 {-0.59/33 230s} 12. Nd2 {+0.85/40 165s} Be5 {-0.61/34 224s}
13. a4 {+0.84/41 182s} b6 {-0.67/34 242s} 14. Ke4 {+0.85/42 408s}
Kc6 {-0.62/35 160s} 15. Nf1 {+0.89/42 260s} Bb3 {-0.65/36 188s}
16. Ra1 {+0.81/42 26s} Bc2+ {-0.70/38 625s} 17. Ke3 {+0.87/43 515s}
a5 {-0.78/37 404s} 18. Kf3 {+0.93/41 210s} Bd3 {-0.75/35 137s}
19. g4 {+1.11/38 89s} g5 {-0.95/36 208s} 20. h3 {+1.09/47 147s}
Bg6 {-1.06/37 131s} 21. Ne3 {+0.95/44 31s} h5 {-1.03/41 292s}
22. Re1 {+0.95/50 123s} b5 {-0.72/42 82s} 23. axb5+ {+0.95/49 39s}
Kxb5 {-0.73/42 59s} 24. Rd1 {+1.03/49 99s} a4 {-0.75/43 125s}
25. Rd8 {+1.11/51 267s} hxg4+ {-0.76/46 195s} 26. hxg4 {+1.11/53 28s}
Bb1 {-0.76/44 92s} 27. Ra8 {+1.11/56 134s} Ba2 {-0.76/47 111s}
28. Ke2 {+1.11/58 150s} Bb3 {-0.77/45 181s} 29. Kd3 {+1.11/52 37s}
Bf7 {-0.78/43 85s} 30. Ke2 {+1.11/57 117s} Bb3 {-0.77/46 208s}
31. Rf8 {+1.11/57 121s} Kb6 {-0.77/42 63s} 32. Rc8 {+1.11/58 139s}
Kb5 {-0.77/44 109s} 33. Kd3 {+1.11/57 142s} Bf7 {-0.77/42 63s}
34. Rd8 {+1.11/59 124s} Bb3 {-0.78/43 77s} 35. Ke4 {+1.11/60 113s}
Be6 {-0.78/47 154s} 36. Re8 {+1.11/62 30s} Ba2 {-0.77/48 219s}
37. Ra8 {+1.11/62 130s} Bb3 {-0.77/44 110s} 38. Rc8 {+1.11/62 133s}
Ba2 {-0.77/43 149s} 39. Kf3 {+1.11/61 134s} Bc4 {-0.77/44 150s}
40. Re8 {+1.11/62 128s} Bd3 {-0.77/45 134s} 41. Ra8 {+1.11/63 155s}
Bb1 {-0.77/43 51s} 42. Re8 {+1.11/60 150s} Ba2 {-0.77/43 80s}
43. Ra8 {+1.11/63 123s} Bb3 {-0.77/46 142s} 44. Rd8 {+1.11/61 50s}
Be6 {-0.61/41 179s} 45. Re8 {+1.11/62 102s} Bd7 {-0.73/41 52s}
46. Re7 {+1.11/61 97s} Bc8 {-0.74/41 55s} 47. Rf7 {+1.11/56 138s}
Be6 {-0.74/45 96s} 48. Re7 {+1.11/64 80s} Bb3 {-0.75/43 73s}
49. Ke2 {+1.11/63 16s} Bf4 {-0.75/44 100s} 50. Re8 {+1.11/59 79s}
Bd6 {-0.76/42 99s} 51. Kf3 {+1.11/62 111s} Be5 {-0.76/44 86s}
52. Rd8 {+1.11/64 208s} Be6 {-0.48/39 51s} 53. Re8 {+1.11/59 17s}
Bf7 {-0.77/45 134s} 54. Rf8 {+1.11/59 150s} Bc4 {-0.77/44 80s}
55. Re8 {+1.11/57 104s} Bd3 {-0.37/44 260s} 56. Ra8 {+1.11/52 196s}
Bb1 {-0.76/41 70s} 57. Ke2 {+1.11/55 114s} Ba2 {-0.74/38 47s}
58. Re8 {+1.11/52 81s} Bb3 {-0.74/40 75s} 59. Ra8 {+1.11/52 97s}
Kb6 {-0.56/39 111s} 60. Rc8 {+1.06/52 143s} Be6 {-0.55/38 35s}
61. Re8 {+0.96/53 154s} Bf7 {-0.56/37 36s} 62. Re7 {+0.84/49 90s}
Ba2 {-0.55/34 68s} 63. Kd3 {+0.86/48 178s} Kb5 {-0.53/33 40s}
64. Kd2 {+0.84/49 92s} Bf4 {-0.53/31 39s} 65. Kd3 {+0.77/50 161s}
Be5 {-0.46/36 124s} 66. Rg7 {+0.70/50 71s} Bb1+ {-0.44/35 105s}
67. Ke2 {+0.69/53 52s} Ba2 {-0.18/33 28s} 68. Re7 {+0.61/51 81s}
Bd6 {0.00/34 43s} 69. Rb7+ {+0.96/45 48s} Ka6 {0.00/43 32s}
70. Rd7 {+0.55/45 41s} Be5 {0.00/45 26s} 71. Kd3 {+0.55/51 63s}
Kb5 {0.00/44 22s} 72. Re7 {+0.55/52 198s} Bb1+ {0.00/47 38s}
73. Kd2 {0.00/53 63s} Ka6 {0.00/48 30s} 74. Re8 {0.00/53 32s} Kb7 {0.00/46 27s}
75. Re7+ {0.00/58 31s} Ka6 {0.00/52 23s}
76. Nc2 {0.00/65 31s, Draw by fifty moves rule} 1/2-1/2

[/pgn]



Lichess analysis - https://lichess.org/gNpkA3bY

[Lyudmil Tsvetkov]

Well, shelter pawns have their own rules, but I don't think we have any extra penalty specifically for doubled pawns in the shelter. I've never thought of that as a valid general principle (I'm not talking about doubled isolated shelter pawns of course). Do you think White would be much better off in this position if we move the g3 pawn to h3? It still provides a hook for an attack. Doubled pawns are generally ok for defense, just not useful for offense.

doubled shelter pawn should be a very valid chess knowledge principle, at least according to my understanding.

SF also does not have specific doubled shelter penalty, but it seems to handle such situations for some reason better, maybe more appropriately tuned specific values, or more relevant specific search.

of course, h3 pawn is much better than g3 pawn in the shelter.

- structure is not compromised
- g4 square is defended
- g3 doubled pawn can be fixed by an enemy pawn by blocking it, making both doubled pawns weak
- most importantly, concerning hooks, thanks for teaching me that term, I did not know what it mean until now, there is a major distinction between 'hooking' a normal and a doubled pawn. a normal hooked pawn can advance and pass, preventing opening of attacking lines, while a doubled hooked pawn can only capture or being captured in the vast majority of cases, so in this way it is an excellent opponent target for opening vital lines. for example, with white pawns on f2,g2,g3, after black's h5-h4 break, white can not play g3-g4 to avoid opening lines usually, as the pawn structure is clumsy, while, with white pawns on f2,g2,h3, on black's g5-g4 break, white has the vital option of passing with h3-h4, thus avoiding opening lines near the king. later, h4 pawn can be defended by the g3 pawn, while a doubled g4 pawn can rarely be defended by an f3 pawn. that seems to make the large portion of the distinction.

[Lyudmil Tsvetkov]

but of course, backward shelter pawn, twice backward shelter pawn, etc., are much more valid chess principles with way higher influence on game development.

[lkaufman]

Well, shelter pawns have their own rules, but I don't think we have any extra penalty specifically for doubled pawns in the shelter. I've never thought of that as a valid general principle (I'm not talking about doubled isolated shelter pawns of course). Do you think White would be much better off in this position if we move the g3 pawn to h3? It still provides a hook for an attack. Doubled pawns are generally ok for defense, just not useful for offense.

doubled shelter pawn should be a very valid chess knowledge principle, at least according to my understanding.

SF also does not have specific doubled shelter penalty, but it seems to handle such situations for some reason better, maybe more appropriately tuned specific values, or more relevant specific search.

of course, h3 pawn is much better than g3 pawn in the shelter.

- structure is not compromised
- g4 square is defended
- g3 doubled pawn can be fixed by an enemy pawn by blocking it, making both doubled pawns weak
- most importantly, concerning hooks, thanks for teaching me that term, I did not know what it mean until now, there is a major distinction between 'hooking' a normal and a doubled pawn. a normal hooked pawn can advance and pass, preventing opening of attacking lines, while a doubled hooked pawn can only capture or being captured in the vast majority of cases, so in this way it is an excellent opponent target for opening vital lines. for example, with white pawns on f2,g2,g3, after black's h5-h4 break, white can not play g3-g4 to avoid opening lines usually, as the pawn structure is clumsy, while, with white pawns on f2,g2,h3, on black's g5-g4 break, white has the vital option of passing with h3-h4, thus avoiding opening lines near the king. later, h4 pawn can be defended by the g3 pawn, while a doubled g4 pawn can rarely be defended by an f3 pawn. that seems to make the large portion of the distinction.

Well, it's worth a test. Doubled pawns also guard key squares, and can also sometimes advance when "hooked", but you might be right. Maybe all weak pawn penalties should be increased by some percentage in the shelter.

[Nordlandia]

Komodo understand this position better than stockfish.

[d]5r2/bp1b3k/p1n1p1np/3p4/5p2/2P5/PP3PPP/R2Q1RK1 b - - 0 24

[Nordlandia]

[pgn][Event "?"]
[Site "Cutechess 1.0.0 | i7-5960X 4.1GHz"]
[Date "2017.08.19"]
[Round "?"]
[White "Stockfish_17081914_x64_bmi2"]
[Black "komodo-11.2.2-64bit"]
[Result "0-1"]
[FEN "5r2/bp1b3k/p1n1p1np/3p4/5p2/2P5/PP3PPP/R2Q1RK1 b - - 0 1"]
[PlyCount "44"]
[SetUp "1"]
[Termination "adjudication"]
[TimeControl "1800+30"]

1... Nce5 {+1.13/28 54s} 2. a4 {-0.34/32 76s} Rf5 {+1.38/31 51s}
3. b4 {-0.45/34 123s} Nc4 {+1.56/31 82s} 4. b5 {-0.34/34 27s} a5 {+1.68/30 31s}
5. Qc2 {-0.90/35 89s} Kg7 {+1.70/33 96s} 6. Rad1 {-0.92/38 244s}
Nge5 {+1.77/34 74s} 7. Rd4 {-0.69/35 38s} Bc5 {+1.96/33 45s}
8. Re1 {-0.74/39 54s} b6 {+2.10/33 35s} 9. Qd1 {-1.69/38 247s} f3 {+1.95/36 37s}
10. g4 {-2.08/41 284s} Rf6 {+2.05/36 100s} 11. Rxe5 {-2.34/38 205s}
Nxe5 {+2.18/34 31s} 12. h3 {-1.82/40 109s} Be8 {+2.41/35 73s}
13. Qe1 {-2.82/35 9.8s} Ng6 {+2.33/36 77s} 14. Qd2 {-3.09/42 200s}
Bf7 {+2.91/36 111s} 15. Kf1 {-3.63/42 170s} e5 {+4.25/35 89s}
16. Rxd5 {-3.25/33 33s} Bxd5 {+4.35/37 64s} 17. Qxd5 {-3.13/32 13s}
Rf4 {+4.37/35 26s} 18. c4 {-4.21/44 214s} Rd4 {+6.08/32 35s}
19. Qxf3 {-4.35/39 9.2s} Rf4 {+6.65/32 29s} 20. Qb7+ {-7.56/36 113s}
Rf7 {+7.35/33 38s} 21. Qd5 {-10.02/39 111s} Rxf2+ {+8.42/34 55s}
22. Ke1 {-9.68/38 30s} Rf6 {+9.71/33 81s}
23. Qb7+ {-7.70/30 13s, Black wins by adjudication} 0-1

[/pgn]

[Lyudmil Tsvetkov]

Well, shelter pawns have their own rules, but I don't think we have any extra penalty specifically for doubled pawns in the shelter. I've never thought of that as a valid general principle (I'm not talking about doubled isolated shelter pawns of course). Do you think White would be much better off in this position if we move the g3 pawn to h3? It still provides a hook for an attack. Doubled pawns are generally ok for defense, just not useful for offense.

doubled shelter pawn should be a very valid chess knowledge principle, at least according to my understanding.

SF also does not have specific doubled shelter penalty, but it seems to handle such situations for some reason better, maybe more appropriately tuned specific values, or more relevant specific search.

of course, h3 pawn is much better than g3 pawn in the shelter.

- structure is not compromised
- g4 square is defended
- g3 doubled pawn can be fixed by an enemy pawn by blocking it, making both doubled pawns weak
- most importantly, concerning hooks, thanks for teaching me that term, I did not know what it mean until now, there is a major distinction between 'hooking' a normal and a doubled pawn. a normal hooked pawn can advance and pass, preventing opening of attacking lines, while a doubled hooked pawn can only capture or being captured in the vast majority of cases, so in this way it is an excellent opponent target for opening vital lines. for example, with white pawns on f2,g2,g3, after black's h5-h4 break, white can not play g3-g4 to avoid opening lines usually, as the pawn structure is clumsy, while, with white pawns on f2,g2,h3, on black's g5-g4 break, white has the vital option of passing with h3-h4, thus avoiding opening lines near the king. later, h4 pawn can be defended by the g3 pawn, while a doubled g4 pawn can rarely be defended by an f3 pawn. that seems to make the large portion of the distinction.

Well, it's worth a test. Doubled pawns also guard key squares, and can also sometimes advance when "hooked", but you might be right. Maybe all weak pawn penalties should be increased by some percentage in the shelter.

I wish you best of luck with any possible experiment.

still, I fail to grasp how are you going to tune this.
you have to tune at the same time, pawn psqt, doubled pawns(possibly also doubled pawns psqt, if you have one), shelter weakness, specific weak pawn shelter weakness, if you try it, etc.
so, you should tune at least 4 evaluation features at the same time, not to mention search, and I fully fail to understand how is it possible to successfully tune those.

that is one of the reasons I am a bit reluctant to become an engine developer.