Wanted: eval insight for a specific position

[Michael Sherwin]

[D]2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - - 0 11

Stockfish evaluates this position as about even while RomiChess believes that black is better by +1.70. I can understand why RomiChess evaluates the position as such as black has an extra pawn and white has two backward pawns. Also blacks knight looks like it has good squares it can reach. I could guess about white's compensation. But it would be guessing and I would appreciate a more experienced explanation.

[kbhearn]

mainly stockfish is finding white's compensation by search not static eval features at the root. At depth 5 it's showing a 1.1 advantage to black for instance and its not til around depth 26 it starts realising black's no better than equal. To reach that conclusion depends on so many varying features (some lines require white to shatter black's kingside structure, others turn into piece vs pawns, one line white gives up an exchange to win the f7 pawn and play RBB v RRB where those central pawns make themself count.

For common features that would bring the realisation a wee bit sooner, the c6 pawn is due a significant bonus as it's supported and cramping the black king. and white's central pawns instead of wing pawns should be worth a little more as well. A small penalty for an unsupported minor might be of use. But i can totally understand a root eval somewhere in the neighborhood of -1 to -1.5. It's just going to take depth to figure out a position like this where white's compensation is largely tactical and varied.

[bob]

[D]2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - - 0 11

Stockfish evaluates this position as about even while RomiChess believes that black is better by +1.70. I can understand why RomiChess evaluates the position as such as black has an extra pawn and white has two backward pawns. Also blacks knight looks like it has good squares it can reach. I could guess about white's compensation. But it would be guessing and I would appreciate a more experienced explanation.

Statically, Crafty likes white better. After a short search, it thinks things are equal. For your comments, black's knight is HORRIBLE compared to the bishop pair white has. The bishops attack both sides of the board, and the center is open/controlled by white letting the bishops operate on both sides of the board at will, while the black knight is a "slow-mover" at best.

For backward pawns, don't forget that black has two isolated pawns which are just as weak as backward pawns, and one is on an open file with two enemy rooks still on the board. White only has one backward pawn that is accessible for a rook. There are other things as well. White pawn on c6 is a thorn. Black can't get too far away or b6 is problematic turning the white c6 pawn into an advanced passed pawn.

[Ferdy]

[D]2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - - 0 11

Stockfish evaluates this position as about even while RomiChess believes that black is better by +1.70. I can understand why RomiChess evaluates the position as such as black has an extra pawn and white has two backward pawns. Also blacks knight looks like it has good squares it can reach. I could guess about white's compensation. But it would be guessing and I would appreciate a more experienced explanation.

In Deuterium I only consider backward pawn when they are in 2nd and 3rd ranks. So in this case the white pawn in f3, but this will not get a penalty because it can advance to f4 without being captured by a black pawn supposedly located at g5 and/or e5. So for pawn backward penalty white has none. For black it has backward at a7 and c7 that can be penalized.

The squares around black king is a bit weak on light squares, and this creates problem because there is white light bishop that black cannot opposed. This can be easily calculated statically by considering the color of pawn and color of bishops (presence and absence) for both sides.

White has the bishop pair but black has the passer.

White's king is in a good position, if this game goes to ending.

[Michael Sherwin]

Thank you all for the replies. However this position just got more interesting. Statically Crafty believes white is better and Stockfish thinks that black is better(+2.61 at ply 7). Two people clearly think white is better while one believes that search and tactics decides the issue. Both Crafty and Stockfish after search think the position is equal. And since human evaluation tends to favor white and computer evaluation through search tends toward equality then where is the human understanding flawed? What is blacks compensation considering the human perspective? Interesting!

[Ferdy]

Thank you all for the replies. However this position just got more interesting. Statically Crafty believes white is better and Stockfish thinks that black is better(+2.61 at ply 7). Two people clearly think white is better while one believes that search and tactics decides the issue. Both Crafty and Stockfish after search think the position is equal. And since human evaluation tends to favor white and computer evaluation through search tends toward equality then where is the human understanding flawed? What is blacks compensation considering the human perspective? Interesting!

Static analysis and dynamic analysis thru search are two different things. As we know the search is now dominated by the engines. Given long search by engine and if it returns a draw or close to draw score, then the human way of strategic planning could be of help here.

[Michael Sherwin]

Thank you all for the replies. However this position just got more interesting. Statically Crafty believes white is better and Stockfish thinks that black is better(+2.61 at ply 7). Two people clearly think white is better while one believes that search and tactics decides the issue. Both Crafty and Stockfish after search think the position is equal. And since human evaluation tends to favor white and computer evaluation through search tends toward equality then where is the human understanding flawed? What is blacks compensation considering the human perspective? Interesting!

Static analysis and dynamic analysis thru search are two different things. As we know the search is now dominated by the engines. Given long search by engine and if it returns a draw or close to draw score, then the human way of strategic planning could be of help here.

I was told a couple of times here at talkchess that I am not a mind reader but I'm going to take a stab at it anyway. Are you saying that if there are several best moves that are basically equal after search that a Humanized static evaluation should be used to determine which move should be played?

[Ferdy]

Thank you all for the replies. However this position just got more interesting. Statically Crafty believes white is better and Stockfish thinks that black is better(+2.61 at ply 7). Two people clearly think white is better while one believes that search and tactics decides the issue. Both Crafty and Stockfish after search think the position is equal. And since human evaluation tends to favor white and computer evaluation through search tends toward equality then where is the human understanding flawed? What is blacks compensation considering the human perspective? Interesting!

Static analysis and dynamic analysis thru search are two different things. As we know the search is now dominated by the engines. Given long search by engine and if it returns a draw or close to draw score, then the human way of strategic planning could be of help here.

I was told a couple of times here at talkchess that I am not a mind reader but I'm going to take a stab at it anyway. Are you saying that if there are several best moves that are basically equal after search that a Humanized static evaluation should be used to determine which move should be played?

What I mean is when the engine score returns a draw score where draw score is zero, or close to draw score say +/- 15 cp, after a long search say 10 minutes, then human strategy could be of help in such position. It is not recommended to use human strategy to judge the position if given position is full of tactics or one that have subtle manueverings/transpositions or something complicated, where long search is need to resolve those tactics.

[zullil]

[D]2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - - 0 11

Stockfish evaluates this position as about even while RomiChess believes that black is better by +1.70. I can understand why RomiChess evaluates the position as such as black has an extra pawn and white has two backward pawns. Also blacks knight looks like it has good squares it can reach. I could guess about white's compensation. But it would be guessing and I would appreciate a more experienced explanation.

From the latest SF, the static evaluation and the MultiPV=3 evaluations at iteration 40 (without endgame tables). Will research with endgame tables later.

Code: Select all

Fen: 2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - -
Key: 041AB64F77D5D165
Checkers:
eval  
      Eval term |    White    |    Black    |    Total
                |   MG    EG  |   MG    EG  |   MG    EG
----------------+-------------+-------------+-------------
       Material |   ---   --- |   ---   --- | -0.66 -0.78
      Imbalance |   ---   --- |   ---   --- |  0.48  0.48
          Pawns |   ---   --- |   ---   --- |  0.69  0.57
        Knights |  0.00  0.00 |  0.03  0.01 | -0.03 -0.01
         Bishop | -0.09 -0.23 | -0.09 -0.14 |  0.00 -0.09
          Rooks |  0.07  0.04 |  0.07  0.04 |  0.00  0.00
         Queens |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
       Mobility |  0.36  0.74 |  0.33  0.84 |  0.03 -0.10
    King safety | -0.10 -0.12 | -0.70 -0.06 |  0.60 -0.06
        Threats |  0.02  0.12 |  0.14  0.36 | -0.12 -0.24
   Passed pawns |  0.00  0.00 |  0.11  0.14 | -0.11 -0.14
          Space |  0.16  0.00 |  0.03  0.00 |  0.12  0.00
----------------+-------------+-------------+-------------
          Total |   ---   --- |   ---   --- |  0.81 -0.51

Total Evaluation: -0.08 (white side)

Code: Select all

info depth 40 seldepth 66 multipv 1 score cp 0 nodes 36943348719 nps 25008308 hashfull 999 tbhits 0 time 1477243 pv g7g5 h4h1 h5h4 d1d3 a4c3 d3d2 b2a3 e2c4 a3b4 e1f1 c3d5 d2a2 c8b8 f1g2 b4c3 c4d5 d8d5 h1c1 h4h3 g2h1 c3b4 c1a1 b4c3 a1c1

info depth 40 seldepth 66 multipv 2 score cp -31 nodes 36943348719 nps 25008308 hashfull 999 tbhits 0 time 1477243 pv b2c3 e1f1 g7g5 h4h1 a4b6 f3f4 g5g4 e2d3 b6d5 d3f5 c8b8 d1d3 c3a5 f2g3 d5b4 d3d1 b4d5 f1f2 a5b4 g3h4 d8g8 h4g5 f7f6 e5f6 g8f8 f5c2 d5f6 d4d5 b4c5 f2g3 c5d6 g3g2 h5h4 h1h4 h8h4 g5h4 f6h5 f4f5 h5f4 g2f1 f8h8 h4f2 h8h1 f2g1 f4h3 f1g2 h1g1 d1g1 h3g1 g2g1 d6e7

info depth 40 seldepth 66 multipv 3 score cp -67 nodes 36943348719 nps 25008308 hashfull 999 tbhits 0 time 1477243 pv f5f4 h4f4 a4c3 d1d3 f7f6 d
4d5 d8d5 f4b4 c3e2 d3d5 b2c3 e1e2 c3b4 e5f6 g7f6 f2h4 b4e7 f3f4 c8b8 d5d7 h8e8 e2f3 a7a6 b5a6 b8a7 d7c7 a7a6 c7b7 e7c5 h4f6 e8e3 f3g2 e3e6 f6h8 e6h6 h8a1 h6c6 b7h7 c6g6 g2f3 g6g1 h7h5 g1f1 f3e2 f1f4 h5c5 f4a4 a1f6 a4f4 f6c3 f4a4 e2d3 a4a3 d3c4 a6b6

[zullil]

[D]2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - - 0 11

Stockfish evaluates this position as about even while RomiChess believes that black is better by +1.70. I can understand why RomiChess evaluates the position as such as black has an extra pawn and white has two backward pawns. Also blacks knight looks like it has good squares it can reach. I could guess about white's compensation. But it would be guessing and I would appreciate a more experienced explanation.

From the latest SF, the static evaluation and the MultiPV=3 evaluations at iteration 40 (without endgame tables). Will research with endgame tables later.

Code: Select all

Fen: 2kr3r/p1p2pp1/2P5/1P2Pp1p/n2P3R/5P2/1b2BB2/3RK3 b - -
Key: 041AB64F77D5D165
Checkers:
eval  
      Eval term |    White    |    Black    |    Total
                |   MG    EG  |   MG    EG  |   MG    EG
----------------+-------------+-------------+-------------
       Material |   ---   --- |   ---   --- | -0.66 -0.78
      Imbalance |   ---   --- |   ---   --- |  0.48  0.48
          Pawns |   ---   --- |   ---   --- |  0.69  0.57
        Knights |  0.00  0.00 |  0.03  0.01 | -0.03 -0.01
         Bishop | -0.09 -0.23 | -0.09 -0.14 |  0.00 -0.09
          Rooks |  0.07  0.04 |  0.07  0.04 |  0.00  0.00
         Queens |  0.00  0.00 |  0.00  0.00 |  0.00  0.00
       Mobility |  0.36  0.74 |  0.33  0.84 |  0.03 -0.10
    King safety | -0.10 -0.12 | -0.70 -0.06 |  0.60 -0.06
        Threats |  0.02  0.12 |  0.14  0.36 | -0.12 -0.24
   Passed pawns |  0.00  0.00 |  0.11  0.14 | -0.11 -0.14
          Space |  0.16  0.00 |  0.03  0.00 |  0.12  0.00
----------------+-------------+-------------+-------------
          Total |   ---   --- |   ---   --- |  0.81 -0.51

Total Evaluation: -0.08 (white side)

Code: Select all

info depth 40 seldepth 66 multipv 1 score cp 0 nodes 36943348719 nps 25008308 hashfull 999 tbhits 0 time 1477243 pv g7g5 h4h1 h5h4 d1d3 a4c3 d3d2 b2a3 e2c4 a3b4 e1f1 c3d5 d2a2 c8b8 f1g2 b4c3 c4d5 d8d5 h1c1 h4h3 g2h1 c3b4 c1a1 b4c3 a1c1

info depth 40 seldepth 66 multipv 2 score cp -31 nodes 36943348719 nps 25008308 hashfull 999 tbhits 0 time 1477243 pv b2c3 e1f1 g7g5 h4h1 a4b6 f3f4 g5g4 e2d3 b6d5 d3f5 c8b8 d1d3 c3a5 f2g3 d5b4 d3d1 b4d5 f1f2 a5b4 g3h4 d8g8 h4g5 f7f6 e5f6 g8f8 f5c2 d5f6 d4d5 b4c5 f2g3 c5d6 g3g2 h5h4 h1h4 h8h4 g5h4 f6h5 f4f5 h5f4 g2f1 f8h8 h4f2 h8h1 f2g1 f4h3 f1g2 h1g1 d1g1 h3g1 g2g1 d6e7

info depth 40 seldepth 66 multipv 3 score cp -67 nodes 36943348719 nps 25008308 hashfull 999 tbhits 0 time 1477243 pv f5f4 h4f4 a4c3 d1d3 f7f6 d
4d5 d8d5 f4b4 c3e2 d3d5 b2c3 e1e2 c3b4 e5f6 g7f6 f2h4 b4e7 f3f4 c8b8 d5d7 h8e8 e2f3 a7a6 b5a6 b8a7 d7c7 a7a6 c7b7 e7c5 h4f6 e8e3 f3g2 e3e6 f6h8 e6h6 h8a1 h6c6 b7h7 c6g6 g2f3 g6g1 h7h5 g1f1 f3e2 f1f4 h5c5 f4a4 a1f6 a4f4 f6c3 f4a4 e2d3 a4a3 d3c4 a6b6

With endgame tables:

Code: Select all

info depth 40 seldepth 64 multipv 1 score cp 0 nodes 28314775816 nps 24149350 hashfull 999 tbhits 687586 time 1172486 pv g7g5 h4h1 h5h4 d1d3 d8d5 f3f4 g5f4 h1h4 h8h4 f2h4 d5d4 h4f6 d4d3 e2d3 a4b6 e1e2 b6d5 f6g7 c8d8 e2f3 d8e7 d3f5 f7f6 f3e4 d5c3 e4f3 c3d5

info depth 40 seldepth 64 multipv 2 score cp -41 nodes 28314775816 nps 24149350 hashfull 999 tbhits 687586 time 1172486 pv b2c3 e1f1 g7g5 h4h1 a4b6 e2d3 f5f4 d3c2 d8d5 d1d3 c3b2 h1g1 b6c4 g1g5 c4e3 f2e3 f4e3 c2b3 d5d4 d3d4 b2d4 b3f7 h5h4 f3f4 c8d8 g5h5 h8h5 f7h5 h4h3 h5g4 e3e2 f1e2 h3h2 g4f3 d8e7 f3d5 e7d8 d5e4 d8e7 e2d3 d4b2 e4g2 e7e6 d3e4 b2c3 f4f5 e6e7 g2f3 c3b2

info depth 40 seldepth 64 multipv 3 score cp -43 nodes 28314775816 nps 24149350 hashfull 999 tbhits 687586 time 1172486 pv a4c3 d1d2 g7g5 h4h1 b2a3 e2c4 a3b4 e1f1 c3d5 d2a2 c8b8 a2a4 b4c3 c4d5 d8d5 f1e2 d5b5 e2d3 c3b2 a4a2 b5b6 d3c2 h8h6 a2b2 h6c6 c2b1 c6c3 d4d5 b6b2 b1b2 c3d3 b2c2 d3f3 f2c5 f3f4 c5e7 f4e4 e7f6 g5g4 h1h5 f5f4 h5g5 g4g3 g5g7 b8b7 c2d3 e4e3 d3d4