draw masking a win

[jdart]

Here is an interesting position I have been looking at recently:

[d]6k1/2pb1p1p/p2p1qp1/3P1n2/8/1QP2BP1/1P2rP1P/R4NK1 b - - 0 1

Here Arasan (White) correctly predicted the sacrifice Nxg3, but it gave it a draw score. The problem is, once Black has sacrificed, he is down material, and then the engine thinks that Black should try very hard to draw. And a draw is possible, but in fact Black can win. The main line is as follows:

[pgn][Event "?"] [Site "?"] [Date "????.??.??"] [Round "?"] [White "?"] [Black "?"] [Result "*"] [SetUp "1"] [FEN "6k1/2pb1p1p/p2p1qp1/3P1n2/8/1QP2BP1/1P2rP1P/R4NK1 b - - 0 1"] [PlyCount "9"] {[#]} 1... Nxg3 2. Nxg3 Qxf3 3. Nxe2 Bh3 4. Nf4 Qg4+ 5. Kh1 Qxf4 (5... Qf3+ 6. Kg1 Qg4+ 7. Kh1 Qf3+ 8. Kg1) *[/pgn]

As indicated though Black has a drawing line, too via 5.. Qf3+. Arasan does find the winning sequence but it takes some time and a very large depth. Even at the branching point here:

[d]6k1/2p2p1p/p2p2p1/3P4/5Nq1/1QP4b/1P3P1P/R6K b - -

it is depth 32 before it plays ... Qxf4. Not sure why yet.

--Jon

[bob]

As a note, there are other drawing lines. Takes Crafty about 4 secs on my MacBook to discover that the capture on g3 is better than a draw...

Here's my output:

26 2.54 0.00 1. ... Nxg3 2. Nxg3 Qxf3 3. Nxe2 Bh3
4. Nf4 Qg4+ 5. Kh1 Qxf4 6. c4 Bg4 7. Kg2
Be2 8. Rxa6 Bxc4 9. Ra8+ Kg7 10. Qc3+ f6
11. f3 Qg5+ 12. Kf2 Qf4 13. Kg1 Qg5+
14. Kf2 Qf4 15. Kg1
26-> 2.71 0.00 1. ... Nxg3 2. Nxg3 Qxf3 3. Nxe2 Bh3
4. Nf4 Qg4+ 5. Kh1 Qxf4 6. c4 Bg4 7. Kg2
Be2 8. Rxa6 Bxc4 9. Ra8+ Kg7 10. Qc3+ f6
11. f3 Qg5+ 12. Kf2 Qf4 13. Kg1 Qg5+
14. Kf2 Qf4 15. Kg1
27 4.77 -1.22 1. ... Nxg3 2. Nxg3 Qxf3 3. Nxe2 Bh3
4. Nf4 Qg4+ 5. Kh1 Qxf4 6. c4 Qe4+ 7. f3
Qe2 8. Rg1 Bf1 9. h4 Qe1 10. Rxf1 Qxf1+
11. Kh2 Kg7 12. Qc3+ f6 13. b4 Qf2+
14. Kh3 h5 15. c5 Qf1+ 16. Kh2 Qd1
17. cxd6 Qe2+ 18. Kg3 cxd6
27-> 5.05 -1.22 1. ... Nxg3 2. Nxg3 Qxf3 3. Nxe2 Bh3
4. Nf4 Qg4+ 5. Kh1 Qxf4 6. c4 Qe4+ 7. f3
Qe2 8. Rg1 Bf1 9. h4 Qe1 10. Rxf1 Qxf1+
11. Kh2 Kg7 12. Qc3+ f6 13. b4 Qf2+
14. Kh3 h5 15. c5 Qf1+ 16. Kh2 Qd1

[hgm]

This is quite common on engines that heavily prune and reduce: they tend to settle for a small gain at low depth, even if there is a much larger gain at higher depth. Because the small gain will get to be PV first, an after that the alternative moves hardly get any depth anymore. So that you now have to search the small gain to twice or triple the depth that was already higher to begin with to get the larger gain within the horizon.

Bad thing is that when you are winning, the advantage usually grows with depth, so by the time you reach the depth that sees the bigger gain, the score in the line with the small gain has already increase so much by additional gains that it is still larger than the large game (which, when searched to the same depth, would have had even larger additional gains). So it will never switch to the stronger move.

I see this behavior very strongly in the derivative of Stockfish that plays mini-Shogi. It often cannot find the move that wins fastest at all, unless you force it (by analyzing after it, or using ' searchmoves' ).

[xr_a_y]

Minic gets the winning sequence quite fast but cannot convert to a win immediately, score is -2.25 (same with stockfish ...)

Code: Select all

18 29 60 627342 31 1045 83	f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 g8h8 b3c3 h8g8 b2b4 h3f5 b4b5 a6b5 
.....
38 225 8815 123414331 69 1399 17946	f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 h2h4 e2f2 g1f1 f2f1 h1h2 f1f2 h2h3 a6a5 b3c3 f7f5 c3f6 f2f3 h3h2 f3f2 h2h1 f2f1 h1h2 f1e2 h2g1 e2e1 g1g2 e1e5 f6e5 d6e5 c4c5 g8f7 g2f3 f7e7 f3f2 h7h6 d5d6 c7d6 c5c6 d6d5 c6c7 e7d7 

Because after that, c4 is quite annoying ... it get the rook the tempo needed to come in defense.

Chances are engines will go the the endgame with the two queens that seems quite sharp and full of delaying checks...

[jdart]

The real problem here that Arasan shouldn't have played into the line allowing the sac. The sac looked like a draw at the depths it was reaching. I have seen this before in other games but this is a clearer example than most.

--Jon

[hgm]

You mean there was another move than the sac with a positive score, and yet it played something it evaluated as a draw? That can only happen with a very bad bug. If it earlier had a PV where the sac was evaluated as winning, but when it actually got to it the win had disappeared... Well, that can happen. For instance by hash grafting. This can catch over-the-horizon gains, and later lose them because the move ordering then sorts them first (as PV), so that there is nothing to graft yet.

[zullil]

Here is an interesting position I have been looking at recently:

[d]6k1/2pb1p1p/p2p1qp1/3P1n2/8/1QP2BP1/1P2rP1P/R4NK1 b - - 0 1

...
--Jon

Seem it takes 157 ms for Lc0 to settle on the winning line (as opposed to simply the first move of the line). Note: Lc0 scores here represent winning percentage multiplied by 100.

Code: Select all

$ ./lc0
       _
|   _ | |
|_ |_ |_| v0.22.0 built Aug  5 2019

setoption name Backend value cudnn-fp16
setoption name NNCacheSize value 1000000
setoption name ScoreType value win_percentage
ucinewgame

Found pb network file: ./network42850
Creating backend [cudnn-fp16]...
GPU: GeForce RTX 2080 Ti
GPU memory: 10.7534 Gb
GPU clock frequency: 1635 MHz
GPU compute capability: 7.5
CUDA Runtime version: 10.1.0
Cudnn version: 7.6.2
Latest version of CUDA supported by the driver: 10.1.0
position fen 6k1/2pb1p1p/p2p1qp1/3P1n2/8/1QP2BP1/1P2rP1P/R4NK1 b - - 0 1
go nodes 1000000
info depth 1 seldepth 2 time 34 nodes 5 score cp 6716 hashfull 0 nps 147 tbhits 0 pv f5d4 c3d4
info depth 2 seldepth 3 time 43 nodes 11 score cp 6877 hashfull 0 nps 255 tbhits 0 pv f5g3 f1g3 f6f3
info depth 2 seldepth 4 time 51 nodes 17 score cp 7296 hashfull 0 nps 333 tbhits 0 pv f5g3 f1g3 f6f3 g3e2
info depth 3 seldepth 4 time 54 nodes 21 score cp 7532 hashfull 0 nps 388 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 f3h1
info depth 3 seldepth 5 time 58 nodes 23 score cp 7427 hashfull 0 nps 396 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 f3e2
info depth 3 seldepth 6 time 64 nodes 26 score cp 7271 hashfull 0 nps 406 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4
info depth 3 seldepth 7 time 68 nodes 33 score cp 6920 hashfull 0 nps 485 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4
info depth 4 seldepth 8 time 76 nodes 47 score cp 7006 hashfull 0 nps 618 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1
info depth 4 seldepth 9 time 85 nodes 64 score cp 7090 hashfull 0 nps 752 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3
info depth 5 seldepth 10 time 93 nodes 76 score cp 6794 hashfull 0 nps 817 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4
info depth 5 seldepth 11 time 97 nodes 94 score cp 6524 hashfull 0 nps 969 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4
info depth 6 seldepth 12 time 101 nodes 112 score cp 6179 hashfull 0 nps 1108 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3
info depth 6 seldepth 13 time 111 nodes 143 score cp 5873 hashfull 0 nps 1288 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 f3e2 c3c4 e2g4 g1f1 g4e4 f1g1 e4g4 g1h1
info depth 6 seldepth 14 time 119 nodes 177 score cp 5805 hashfull 0 nps 1487 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 f3e2 c3c4 e2g4 g1f1 g4e4 f1g1 e4g4 g1h1 g4e4
info depth 8 seldepth 18 time 123 nodes 230 score cp 5760 hashfull 0 nps 1869 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 8 seldepth 19 time 127 nodes 290 score cp 5802 hashfull 0 nps 2283 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 8 seldepth 20 time 138 nodes 364 score cp 5779 hashfull 0 nps 2637 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 9 seldepth 20 time 151 nodes 432 score cp 5955 hashfull 0 nps 2860 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 10 seldepth 20 time 157 nodes 507 score cp 6248 hashfull 0 nps 3229 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 a6a5
info depth 11 seldepth 20 time 173 nodes 745 score cp 6684 hashfull 0 nps 4306 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 h3f1 h2h3 f1g2
info depth 11 seldepth 21 time 181 nodes 827 score cp 6725 hashfull 1 nps 4569 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 e2f2 b2b4 h3f1
info depth 11 seldepth 22 time 185 nodes 884 score cp 6764 hashfull 1 nps 4778 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 e2f2 b2b4 h3f1 f6h8
info depth 12 seldepth 22 time 191 nodes 1034 score cp 6905 hashfull 1 nps 5413 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 e2f2 b2b4 h3f1 h2h3
info depth 12 seldepth 23 time 203 nodes 1286 score cp 7027 hashfull 1 nps 6334 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 e2f2 b2b4 h3f1 h2h3
info depth 13 seldepth 25 time 216 nodes 1521 score cp 6974 hashfull 1 nps 7041 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 h5h4 c3f6 e2f2 b2b4 h3f1 h2h3
info depth 13 seldepth 26 time 223 nodes 1652 score cp 6954 hashfull 1 nps 7408 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 e2f2 c3f6 h3g4 g1g3 f2f1
info depth 14 seldepth 26 time 243 nodes 2167 score cp 6930 hashfull 1 nps 8917 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 e2f2 g1g3 h3f1 c3f6 f1e2 g3g6
info depth 14 seldepth 27 time 273 nodes 2839 score cp 6948 hashfull 2 nps 10399 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6c3 h3f1 h2h3 f1h3 c4c5
info depth 14 seldepth 29 time 300 nodes 3223 score cp 7015 hashfull 2 nps 10743 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6c3 h3f1 h2h3 f1h3 c3f6
info depth 15 seldepth 29 time 304 nodes 3459 score cp 7017 hashfull 2 nps 11378 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6c3 h3f1 h2h3 f1h3 c3f6
info depth 16 seldepth 29 time 373 nodes 5111 score cp 6926 hashfull 3 nps 13702 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6f4 h3f1 h2h3 f1h3 f4f6 h3f1 g1g4
info depth 16 seldepth 30 time 392 nodes 5637 score cp 6934 hashfull 4 nps 14380 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6f4 h3f1 h2h3 f1h3 f4f6 h3f1 g1g4
info depth 16 seldepth 31 time 463 nodes 7229 score cp 7028 hashfull 5 nps 15613 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c4c5 d6c5 c3f6 e2f2 f6f4 h3f1 h2h3 f1e2 g1g2 e2f3 f4g5
info depth 17 seldepth 32 time 473 nodes 7499 score cp 7047 hashfull 5 nps 15854 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c4c5 d6c5 c3f6 e2f2 f6f4 h3f1 h2h3 f1e2 g1g2 e2f3 f4g5
info depth 17 seldepth 33 time 590 nodes 10833 score cp 7050 hashfull 7 nps 18361 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6f4 h3f1 f4f6 f1d3 h2h3 d3f1 g1g4 f1e2 g4f4 f2f1
info depth 18 seldepth 33 time 607 nodes 11455 score cp 7025 hashfull 7 nps 18871 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h7h5 b3c3 a6a5 b2b3 h5h4 c3f6 e2f2 f6f4 h3f1 f4f6 f1d3 h2h3 d3f1 g1g4 f1e2 g4f4 f2f1
info depth 18 seldepth 34 time 912 nodes 19939 score cp 7405 hashfull 11 nps 21862 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 h7h5 f3f4 f2f1 g3g1 e2f3 c3f3 f1f3
info depth 18 seldepth 35 time 1029 nodes 24786 score cp 7619 hashfull 13 nps 24087 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 b3b8 g8g7 b8b3 e2f2 b3c3 g7g8 g1g3 f1e2 b2b3 h7h5 f3f4 f2f1 g3g1 e2f3 c3f3 f1f3
info depth 19 seldepth 35 time 1080 nodes 26536 score cp 7691 hashfull 14 nps 24570 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 b3b8 g8g7 b8b3 e2f2 b3c3 g7g8 g1g3 f1e2 b2b3 h7h5 f3f4 f2f1 g3g1 e2f3 c3f3 f1f3
info depth 19 seldepth 36 time 1129 nodes 28614 score cp 7769 hashfull 15 nps 25344 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 h7h5 f3f4 f2f1 g3g1 e2f3 c3f3 f1f3
info depth 19 seldepth 37 time 1139 nodes 29127 score cp 7791 hashfull 15 nps 25572 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 h7h5 f3f4 f2f1 g3g1 e2f3 c3f3 f1f3
info depth 19 seldepth 38 time 1186 nodes 31350 score cp 7865 hashfull 16 nps 26433 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 h7h5 f3f4 f2f1 g3g1 e2f3 c3f3 f1f3
info depth 19 seldepth 39 time 1772 nodes 59807 score cp 8371 hashfull 27 nps 33751 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 b3b8 g8g7 b8b3 e2f2 b3c3 g7g8 g1g3 f1e2 b2b3 f2f1 g3g1 e2f3 c3f3 f1f3 g1g2 f3d1 g2g1 d1b3 g1c1 a6a5 h1g1 a5a4
info depth 19 seldepth 40 time 1972 nodes 71070 score cp 8479 hashfull 30 nps 36039 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 b3b8 g8g7 b8b3 e2f2 b3c3 g7g8 g1g3 f1e2 c3f6 f2f1 g3g1 e2f3 f6f3 f1f3 g1g2 f3f1 g2g1 f1c4 g1a1 c4d5 h1g1 f7f5 h2h4
info depth 20 seldepth 40 time 2712 nodes 115431 score cp 8709 hashfull 45 nps 42563 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 b3b8 g8g7 b8b3 e2f2 b3c3 g7g8 g1g3 f1e2 b2b3 f2f1 g3g1 e2f3 c3f3 f1f3 g1g2 f3b3 g2g6 h7g6 h1g2 b3c4
info depth 20 seldepth 41 time 3976 nodes 198511 score cp 8922 hashfull 67 nps 49927 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 f2f1 g3g1 e2f3 c3f3 f1f3 g1g2 f3b3 g2g6 h7g6 h1g2 b3c4
info depth 20 seldepth 42 time 5574 nodes 303273 score cp 9046 hashfull 96 nps 54408 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8c8 f1c4 c8f5 f2b2 f5e4 b2b3 e4d4 g7g8 h2h4 c4d5 d4f6 h7h5 h1h2 b3c2
info depth 21 seldepth 42 time 7333 nodes 440255 score cp 9131 hashfull 126 nps 60037 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 f2f1 g3g1 e2f3 c3f3 f1f3 g1g2 f3b3 g2g6 h7g6 h1g2 b3c4
info depth 21 seldepth 42 time 8086 nodes 503611 score cp 9158 hashfull 139 nps 62281 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 g1g3 e2f2 b3b8 g8g7 b8b3 f1e2 b3c3 g7g8 b2b3 f2f1 g3g1 e2f3 c3f3 f1f3 g1g2 f3b3 g2g6 h7g6 h1g2 b3c4
bestmove f5g3 ponder f1g3

[Joost Buijs]

Nightmare exhibits the same behavior like Arasan does, it finds Nxg3 with a draw score very fast, after this it takes a long time before it sees a positive score. It depends a bit upon randomness from YBW, sometimes it sees a positive score after a few seconds and sometimes it takes almost 30 seconds.

Code: Select all

info depth 13 score cp 0 time 256 nodes 2718730 nps 10620039 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4
info depth 14 score cp 0 time 280 nodes 2923472 nps 10440971 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1
info depth 15 score cp 0 time 332 nodes 3399004 nps 10237963 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3
info depth 16 score cp 0 time 365 nodes 3807216 nps 10430728 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 17 score cp 0 time 551 nodes 6388493 nps 11594361 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4
info depth 18 score cp 0 time 646 nodes 7490822 nps 11595699 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1
info depth 19 score cp 0 time 847 nodes 9956991 nps 11755597 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3
info depth 20 score cp 0 time 1038 nodes 12298690 nps 11848448 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 21 score cp 0 time 1308 nodes 15514785 nps 11861456 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4
info depth 22 score cp 0 time 1754 nodes 20966916 nps 11953771 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1
info depth 23 score cp 0 time 2681 nodes 32835328 nps 12247418 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3
info depth 24 score cp 0 time 3988 nodes 50648206 nps 12700151 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1
info depth 25 score cp 0 time 5736 nodes 76381639 nps 13316185 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4 g1h1 g4f3 h1g1 f3g4
info depth 26 score cp 238 time 26740 nodes 381887546 nps 14281508 tbhits 0 pv f5g3 f1g3 f6f3 g3e2 d7h3 e2f4 f3g4 g1h1 g4f4 c3c4 f4e4 f2f3 e4e2 a1g1 h3f1 h2h4 e2f2 g1f1 f2f1 h1h2 h7h5 b3b8 g8h7 b8c7 f1f2 h2h3 f2f3 h3h2

This is an interesting position to play a bit more with, and to find out what is actually happening.

[jdart]

You mean there was another move than the sac with a positive score, and yet it played something it evaluated as a draw?

Arasan is White here. I meant, it played a move allowing the sac.

--Jon