stockfish's and dragon2.6.1's evaluation bugs

[Uri Blass]

[fen]1nb1kbn1/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1 [/fen]

In this position I expect normal chess evaluation to lead to advantage that is increasing when you search deeper.

This works with old programs like fruit and the evaluation at depth d+8 is always bigger than the evaluation at depth d

d=1 19.73 d=9 19.76
d=2 19.2 d=10 19.78
d=3 19.72 d=11 19.93
d=4 19.49 d=12 20.03
d=5 19.7 d=13 20.02
d=6 19.35 d=14 20.07
d=7 19.78 d=15 20.18
d=8 19.58 d=16 20.27
d=9 19.76 d=17 20.27
d=10 19.78 d=18 20.31
d=11 19.93 d=19 20.28

Dragon Komodo has strange jump at the same depth

Dragon-2.6.1-64bit-avx2:
1 00:00 6 3k +14.55 h2-h4
1 00:00 40 19k +22.29 g2-g4
2 00:00 199 86k +14.61 g2-g4 h7-h5 g4xh5
2 00:00 352 149k +21.99 h2-h4 Ng8-f6

I used 7 cores but at least the 14.55 to 22.29 jump at depth 1 is reproducable and I get big jumps also with 1 core and I wonder what is the reason for it

FEN: 1nb1kbn1/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1

Dragon-2.6.1-64bit-avx2:
1 00:00 1 731 +14.55 h2-h4
1 00:00 2 1k +22.29 g2-g4
2 00:00 31 20k +14.61 g2-g4 h7-h5 g4xh5
2 00:00 43 27k +15.39 h2-h4 g7-g5 h4xg5
3 00:00 126 73k +22.06 h2-h4 Nb8-c6 Nb1-c3
4 00:00 192 104k +14.54 h2-h4 Nb8-c6 Nb1-c3 Nc6-d8
4 00:00 288 146k +14.72 g2-g4 h7-h5 g4xh5 g7-g5
4 00:00 347 165k +14.83 Ng1-f3 Nb8-c6 Nf3-g5 Nc6-d8 Ng5xh7
4 00:00 416 189k +15.19 Nb1-c3 Nb8-c6 Nc3-d5 Nc6-e5 Nd5xc7+ Ke8-d8

For stockfish I get the following strange analysis and for some reason it prefers the inferior e3 and not the superior e4 and I do not know what type of evaluation push it to do it because every normal evaluation is going to prefer e4 and I understood that stockfish even does not use NNUE in this type of position
1.e3 Nf6 2.a3 d5 3.g3 e5 4.d3 Bd6 5.b3 is the main line of stockfish 7 cores at depth 49 and it seems to me a main line of beginner that has no knowledge about chess.

I do not know how much time stockfish needs to find a move like 1.e4 and it may be interesting if somebody with more than 7 scores can give stockfish to analyze for more time to see how much time stockfish needs to find a move like 1.e4 or some increase in the evaluation back to 24 pawns that it had at depth 4(23.30 that stockfish had at depth 49 when I stopped it is almost the minimal evaluation that it got and basically no improvement in the evaluation is a way that I thought in the past to detect fortress because usually in better positions the score simply go up but at least with stockfish's stupid evaluation it seems not to be a good idea).

Stockfish_22021019_x64_avx2:
NNUE evaluation using nn-6877cd24400e.nnue enabled
1/1 00:00 182 182k +23.69 e2-e3
2/2 00:00 1k 584k +23.53 d2-d3 c7-c6
3/3 00:00 2k 1,223k +23.73 Nb1-c3 c7-c6 Ng1-f3
4/4 00:00 6k 2,804k +24.00 Nb1-c3 a7-a6 e2-e3 Nb8-c6 Ng1-f3
5/5 00:00 9k 4,622k +23.69 Nb1-c3 a7-a6 e2-e3 Nb8-c6 Ng1-f3
6/6 00:00 23k 7,505k +23.76 Ng1-f3 Ng8-f6 Nb1-c3 Nb8-c6 Nc3-b5 Ke8-d8 g2-g3
7/7 00:00 83k 11,911k +23.42 Nb1-a3 Nb8-c6 Na3-b5 Ke8-d8 Ng1-f3 Ng8-f6 g2-g3 a7-a6
8/9 00:00 105k 13,136k +23.53 Nb1-a3 a7-a6 e2-e3 e7-e6 Na3-c4 Nb8-c6 Ng1-e2 Ng8-f6
9/11 00:00 136k 12,354k +23.42 Nb1-a3 a7-a6 Na3-c4 Ng8-f6 d2-d3 b7-b5 Nc4-d2 Nb8-c6 Ng1-f3
10/14 00:00 352k 12,150k +23.30 Nb1-a3 a7-a6 Ng1-f3 e7-e6 Na3-c4 Ng8-f6 Nf3-g5 d7-d5 Nc4-e3 Nb8-c6
11/12 00:00 570k 12,118k +23.30 Ng1-f3 e7-e6 a2-a3 Nb8-c6 d2-d4 Ng8-f6 Nb1-d2 Bf8-d6 e2-e4 Ke8-f8 Bf1-d3
12/15 00:00 905k 12,752k +23.23 Ng1-f3 Ng8-f6 Nb1-a3 Nb8-c6 Na3-b5 Ke8-d8 Nf3-g5 a7-a6 Nb5-c3 Kd8-e8 e2-e4 e7-e6
13/16 00:00 1,319k 13,460k +23.30 Ng1-f3 Ng8-f6 d2-d3 Nb8-c6 h2-h3 e7-e6 Nb1-d2 d7-d5 a2-a3 Bf8-d6 e2-e3 Ke8-f8 d3-d4
14/18 00:00 2,625k 14,665k +23.30 Nb1-c3 Nb8-c6 d2-d4 e7-e6 a2-a3 Ng8-f6 Ng1-f3 Nf6-d5 Nc3-e4 f7-f5 Ne4-d2 Nd5-f6 c2-c3 Nf6-e4
15/15 00:00 2,772k 14,669k +23.30 Nb1-c3 Nb8-c6 Ng1-f3 e7-e6 a2-a3 Ng8-f6 e2-e3 Bf8-d6 d2-d4 Ke8-f8 Bf1-d3 Nf6-d5 Nc3-e4 f7-f5 Ne4-d2
16/17 00:00 3,264k 14,705k +23.23 Nb1-c3 Nb8-c6 Ng1-f3 e7-e6 a2-a3 Ng8-f6 e2-e3 Nf6-d5 Nc3-e2 Nd5-f6 c2-c4 Bf8-d6 d2-d4 Ke8-f8 Ne2-c3 Kf8-g8 Bf1-d3
17/26 00:00 4,428k 14,615k +23.23 Nb1-c3 e7-e6 Nc3-b5 a7-a6 Nb5xc7+ Ke8-d8 Nc7-a8 b7-b5 e2-e4 Ng8-h6 Qd1-f3 Bc8-b7 Na8-b6 Kd8-c7 a2-a4 Kc7xb6 a4xb5 a6xb5 Qf3-e3+ Bf8-c5 d2-d4 Bc5-b4+ c2-c3
18/33 00:00 8,938k 14,922k +23.30 Nb1-c3 e7-e6 a2-a3 Ng8-f6 e2-e3 Nb8-c6 d2-d4 Nf6-d5 Nc3-b1 Nd5-f6 Nb1-d2 d7-d5 f2-f3 Bf8-d6 Ke1-f2 Ke8-f8 Bf1-d3 Nc6-e7
19/24 00:00 9,838k 14,975k +23.30 Nb1-c3 Nb8-c6 d2-d4 e7-e6 a2-a3 Ng8-f6 e2-e3 Nf6-d5 Nc3-b1 Nd5-f6 Nb1-d2 d7-d5 f2-f3 Bf8-d6 Ke1-f2 Ke8-f8 Bf1-d3 e6-e5 Ng1-e2
20/27 00:01 20,245k 15,466k +23.34 e2-e3 Ng8-f6 Nb1-c3 e7-e6 a2-a3 Nb8-c6 d2-d4 d7-d5 f2-f3 Bf8-d6 Ke1-f2 e6-e5 Ng1-e2 Ke8-f8 Kf2-g1 Kf8-g8 Ra1-a2 b7-b6 Qd1-d3 e5xd4 e3xd4 Nc6-e7
21/26 00:01 24,087k 15,450k +23.30 e2-e3 Ng8-f6 Ng1-e2 e7-e6 f2-f3 Nb8-c6 Ke1-f2 b7-b6 Kf2-g1 a7-a5 b2-b3 Bc8-a6 d2-d3 Bf8-d6 c2-c4 Ke8-f8 d3-d4 Kf8-g8 e3-e4
22/23 00:01 25,381k 15,401k +23.30 e2-e3 Ng8-f6 Ng1-e2 e7-e6 f2-f3 Nb8-c6 Ke1-f2 b7-b6 Kf2-g1 a7-a5 b2-b3 Bc8-a6 d2-d3 Bf8-d6 c2-c4 Bd6-b4 d3-d4 Ke8-f8 a2-a3 Bb4-d6 Nb1-c3 Kf8-g8
23/27 00:01 29,027k 15,464k +23.34 e2-e3 Ng8-f6 Ng1-e2 e7-e6 f2-f3 Nb8-c6 Ke1-f2 b7-b6 Kf2-g1 a7-a5 a2-a3 Bc8-a6 d2-d3 Bf8-e7 c2-c4 Ke8-f8 Ne2-c3 Kf8-g8 d3-d4 Ba6-b7 Nb1-d2 Be7-d6 Nc3-e2 Bb7-a6
24/28 00:02 31,578k 15,457k +23.34 e2-e3 Ng8-f6 Ng1-e2 e7-e6 f2-f3 Nb8-c6 Ke1-f2 b7-b6 Kf2-g1 a7-a5 a2-a3 Bc8-a6 d2-d3 Bf8-e7 c2-c4 Ke8-f8 Nb1-c3 Kf8-g8 b2-b3 d7-d5 Bc1-b2 h7-h6 Ra1-c1 d5xc4 b3xc4 Nc6-e5 f3-f4 Ne5-c6
25/31 00:03 59,299k 15,708k +23.46 e2-e3 Ng8-f6 g2-g3 Nb8-c6 Bf1-h3 e7-e5 Ng1-e2 Ke8-d8 Bh3-g2 d7-d6 h2-h3 Bc8-e6 d2-d3 Kd8-c8 O-O g7-g6 a2-a3 Bf8-h6 Rf1-e1 Kc8-b8 c2-c4 d6-d5 Nb1-d2 d5xc4 d3xc4 Be6-d7
26/33 00:04 72,752k 15,629k +23.34 e2-e3 Ng8-f6 g2-g3 d7-d5 Bf1-g2 Nb8-c6 a2-a3 e7-e5 h2-h3 Bf8-d6 Ng1-e2 e5-e4 d2-d4 b7-b6 Nb1-d2 Ke8-f8 b2-b4 Bc8-a6 Ne2-c3 Ba6-c4 g3-g4 h7-h6 h3-h4 Kf8-g8 g4-g5
27/32 00:05 81,861k 15,581k +23.30 e2-e3 Ng8-f6 g2-g3 Nb8-c6 Bf1-h3 e7-e5 d2-d3 Ke8-d8 Ng1-e2 d7-d5 Bh3-g2 Bf8-d6 Ke1-f1 e5-e4 h2-h3 Nc6-e5 d3xe4 d5xe4 Nb1-c3 Bc8-d7 Qd1-d4 Ne5-c6 Qd4-a4 Kd8-c8 Kf1-g1 Kc8-b8 Nc3xe4 Nf6xe4 Qa4xe4
28/39 00:05 93,248k 15,599k +23.38 e2-e3 Ng8-f6 g2-g3 Nb8-c6 Bf1-h3 e7-e5 d2-d3 Ke8-d8 Bh3-g2 d7-d5 h2-h3 Bf8-d6 g3-g4 d5-d4 Ng1-e2 d4xe3 f2xe3 Bc8-e6 O-O h7-h6 Nb1-c3 a7-a6 a2-a3 Kd8-c8 Rf1-e1 Kc8-b8 Ne2-g3 Bd6-c5 b2-b4
29/37 00:07 109,206k 15,517k +23.38 e2-e3 Ng8-f6 g2-g3 Nb8-c6 Bf1-h3 e7-e5 d2-d3 Ke8-d8 Bh3-g2 d7-d5 h2-h3 Bf8-d6 Nb1-d2 d5-d4 e3xd4 Nc6xd4 c2-c3 Nd4-c6 Qd1-b3 Kd8-e7 Ng1-f3 h7-h6 Ke1-f1 Nf6-d7 Qb3-c2 Ke7-f8 a2-a3 f7-f6
30/36 00:08 128,822k 15,476k +23.46 e2-e3 Ng8-f6 g2-g3 Nb8-c6 Bf1-g2 e7-e5 a2-a3 e5-e4 d2-d4 d7-d5 h2-h3 Bf8-d6 Ng1-e2 b7-b6 Nb1-d2 Ke8-f8 b2-b3 Nc6-a5 c2-c4 d5xc4 b3xc4 c7-c5 O-O Kf8-g8 Qd1-c2 Na5-c6 Qc2-c3 Bc8-b7 g3-g4 c5xd4 e3xd4
31/38 00:14 221,414k 15,312k +23.30 e2-e3 Ng8-f6 g2-g3 Nb8-c6 Bf1-h3 e7-e5 d2-d3 Ke8-d8 Bh3-g2 d7-d5 h2-h3 Bf8-d6 Nb1-d2 d5-d4 Ng1-e2 d4xe3 f2xe3 Bc8-e6 c2-c3 Bd6-c5 Nd2-f3 Bc5-b6 e3-e4 a7-a5 Ke1-f1 Kd8-c8 a2-a3 Nf6-d7 b2-b4 f7-f6 Bc1-b2 Nc6-e7 b4xa5
32/46 01:12 1,071,944k 14,870k +23.38 e2-e3 Ng8-f6 g2-g4 e7-e6 Bf1-g2 Nb8-c6 f2-f3 d7-d5 Ng1-e2 Bf8-d6 O-O Ke8-f8 d2-d3 h7-h5 g4-g5 Nf6-d7 Nb1-d2 Kf8-g8 c2-c3 Nc6-e7 f3-f4 g7-g6 Nd2-f3 Ne7-f5 Qd1-c2 Nf5-e7 a2-a3 c7-c6 c3-c4 Ne7-f5
33/40 01:18 1,156,480k 14,817k +23.38 e2-e3 Ng8-f6 g2-g4 e7-e6 Bf1-g2 Nb8-c6 f2-f3 d7-d5 Ng1-e2 Bf8-d6 O-O Ke8-f8 d2-d3 h7-h5 g4-g5 Nf6-d7 Nb1-d2 g7-g6 f3-f4 Kf8-g8 a2-a3 Nc6-e7 Nd2-f3 Ne7-f5 b2-b4 c7-c6 c2-c3 Bd6-c7
34/42 01:22 1,223,565k 14,789k +23.30 e2-e3 Ng8-f6 g2-g4 e7-e6 Bf1-g2 Nb8-c6 f2-f3 d7-d5 Ng1-e2 Bf8-d6 O-O Ke8-f8 d2-d3 Kf8-g8 g4-g5 Nf6-d7 a2-a3 b7-b6 Nb1-c3 Bc8-a6 f3-f4 Nc6-e7 b2-b4 Ne7-f5 Qd1-e1 h7-h6 h2-h4 h6xg5 h4xg5 c7-c5 b4-b5 Ba6-b7 Qe1-f2 Bd6-b8 e3-e4
35/43 01:31 1,348,760k 14,768k +23.38 e2-e3 Ng8-f6 g2-g4 e7-e6 Bf1-g2 Nb8-c6 f2-f3 d7-d5 Ng1-e2 Bf8-d6 O-O Ke8-f8 d2-d3 Kf8-g8 c2-c3 h7-h6 h2-h3 h6-h5 d3-d4 h5xg4 h3xg4 Nc6-a5 Qd1-d3 Nf6-d7 b2-b3 c7-c5 Nb1-d2 c5xd4 c3xd4 Bd6-b4 Nd2-b1 Nd7-f6 a2-a3 Bb4-e7 Nb1-d2
36/38 01:31 1,353,335k 14,767k +23.38 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 Nb8-c6 d2-d4 e7-e6 Ng1-e2 b7-b6 c2-c3 Bf8-e7 Qd1-d3 Ke8-f8 Nb1-d2 g7-g6 Bf1-g2 Nc6-b8 b2-b4 Kf8-g8 b4-b5 h7-h5 h2-h3 c7-c5 O-O Nb8-d7 a2-a4 Be7-d6 a4-a5 b6xa5 Ra1xa5 c5xd4 c3xd4 Bd6-b4 Ra5xa7 Bb4xd2 Qd3xd2
37/39 01:38 1,453,242k 14,754k +23.34 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 Nb8-c6 d2-d4 e7-e6 Ng1-e2 b7-b6 c2-c3 Bf8-e7 Qd1-d3 Ke8-f8 Nb1-d2 g7-g6 Bf1-g2 Nc6-b8 b2-b4 Kf8-g8 b4-b5 h7-h5 h2-h3 c7-c5 O-O Nb8-d7 a2-a4 Be7-d6 a4-a5 c5xd4 a5xb6 Nd7xb6 Ne2xd4 e6-e5 Nd4-c6 h5xg4 h3xg4 Nb6-d7
38/47 14:31 12,611,370k 14,466k +23.23 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 e7-e6 Ng1-e2 Bf8-d6 Bf1-g2 Ke8-f8 O-O Nb8-c6 a2-a3 b7-b6 d2-d3 h7-h5 g4-g5 Nf6-d7 f3-f4 Nc6-e7 b2-b4 g7-g6 Nb1-c3 Bc8-a6 Qd1-e1 Kf8-g8 Qe1-f2 c7-c5 h2-h4 Ne7-f5 Bc1-d2 Bd6-c7 b4-b5 Ba6-b7 e3-e4 d5xe4 d3xe4 Nf5-d4
39/48 14:41 12,743,637k 14,463k +23.23 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 e7-e6 Ng1-e2 Bf8-d6 Bf1-g2 Ke8-f8 O-O Nb8-c6 a2-a3 b7-b6 d2-d3 h7-h5 g4-g5 Nf6-d7 f3-f4 Nc6-e7 b2-b4 g7-g6 Nb1-c3 Bc8-a6 Qd1-e1 Ne7-f5 Qe1-f2 c7-c5 Rf1-e1 Kf8-g8 Bc1-d2 h5-h4 b4-b5 Ba6-b7 a3-a4 Bd6-c7 Kg1-h1 a7-a6 b5xa6
40/45 14:46 12,816,992k 14,462k +23.23 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 e7-e6 Ng1-e2 Bf8-d6 Bf1-g2 Ke8-f8 O-O Nb8-c6 a2-a3 b7-b6 d2-d3 h7-h5 g4-g5 Nf6-d7 f3-f4 Nc6-e7 b2-b4 g7-g6 Nb1-d2 e6-e5 Nd2-f3 Kf8-g8 c2-c3 e5xf4 e3xf4 Nd7-f8 Qd1-a4 Bc8-g4 Qa4xa7 Bg4xf3 Bg2xf3 Kg8-g7 h2-h4 Nf8-e6 a3-a4 Kg7-g8 a4-a5 b6xa5
41/51 14:55 12,941,685k 14,459k +23.23 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 e7-e6 Ng1-e2 Bf8-d6 Bf1-g2 Ke8-f8 O-O Nb8-c6 a2-a3 b7-b6 d2-d3 h7-h5 g4-g5 Nf6-g8 f3-f4 Ng8-e7 b2-b4 e6-e5 f4xe5 Nc6xe5 h2-h3 g7-g6 Nb1-c3 Bc8-d7 Ne2-f4 a7-a5 Bc1-b2 Kf8-g8 b4xa5 b6xa5 e3-e4 Bd6-c5+ Kg1-h1 d5xe4 Nc3xe4
42/52 18:47 16,249,154k 14,407k +23.30 e2-e3 Ng8-f6 g2-g4 d7-d5 f2-f3 e7-e6 Ng1-e2 Bf8-d6 Bf1-g2 Ke8-f8 O-O Nb8-c6 a2-a3 b7-b6 d2-d3 h7-h5 g4-g5 Nf6-g8 f3-f4 Ng8-e7 b2-b4 e6-e5 f4xe5 Nc6xe5 h2-h3 g7-g6 Nb1-c3 Kf8-g8 Qd1-e1 Bc8-e6 b4-b5 c7-c5 Qe1-f2 Ne5-d7 d3-d4 Be6-f5 h3-h4 f7-f6 Bc1-b2 Bd6-c7 Nc3xd5 Ne7xd5
43/58 26:36 22,938,352k 14,369k +23.30 e2-e3 Ng8-f6 a2-a3 d7-d5 Ng1-e2 Nb8-c6 h2-h3 e7-e5 d2-d3 Bf8-d6 Nb1-d2 Ke8-f8 g2-g3 e5-e4 d3-d4 b7-b6 b2-b3 Bc8-a6 c2-c4 Nc6-a5 Qd1-c2 d5xc4 b3xc4 c7-c5 Bc1-b2 c5xd4 e3xd4 Kf8-g8 O-O-O Bd6-f8 Qc2-a4 Ba6-c8 g3-g4 h7-h6 Qa4-c2 Bf8-d6 Kc1-b1 Bc8-b7 Ne2-c3 Na5-c6 Nc3xe4 Nf6xe4 Nd2xe4
44/52 26:46 23,084,443k 14,370k +23.30 e2-e3 Ng8-f6 a2-a3 d7-d5 Ng1-e2 Nb8-c6 h2-h3 e7-e5 d2-d3 Bf8-d6 Nb1-d2 Ke8-f8 g2-g3 b7-b6 Bf1-g2 Bc8-b7 b2-b4 a7-a5 b4-b5 Nc6-e7 c2-c3 Nf6-d7 Qd1-c2 Kf8-g8 O-O g7-g6 g3-g4 Nd7-c5 d3-d4 Nc5-e6 f2-f4 e5xf4 e3xf4 h7-h5 g4xh5 g6xh5 Nd2-f3 Ne6-g7 Qc2-d3 Ne7-f5 a3-a4 Bd6-e7 c3-c4 d5xc4
45/58 27:18 23,508,914k 14,349k +23.30 e2-e3 Ng8-f6 a2-a3 d7-d5 Ng1-e2 Nb8-c6 h2-h3 e7-e5 d2-d3 Bf8-d6 Nb1-d2 Ke8-f8 g2-g3 b7-b6 Bf1-g2 Bc8-b7 O-O Kf8-g8 b2-b4 a7-a5 b4-b5 Nc6-d8 Bc1-b2 Nd8-e6 Qd1-b1 Ne6-c5 a3-a4 Nc5-e6 h3-h4 Bd6-b4 Qb1-d1 e5-e4 Nd2-b1 e4xd3 c2xd3 Bb4-d6 Bb2xf6 g7xf6 Nb1-c3 c7-c6 b5xc6 Bb7xc6 Nc3xd5
46/62 1:00:08 51,194,025k 14,186k +23.23 e2-e3 Ng8-f6 a2-a3 d7-d5 Ng1-e2 Nb8-c6 h2-h3 e7-e5 d2-d3 Bf8-d6 Nb1-d2 Ke8-f8 g2-g3 b7-b6 Bf1-g2 a7-a5 c2-c4 Nc6-e7 c4xd5 Ne7xd5 Qd1-b3 Nd5-e7 O-O Bc8-e6 Qb3-b5 Be6-d5 e3-e4 Bd5-b7 Nd2-f3 Ne7-c6 Qb5-c4 h7-h6 Bc1-d2 Nf6-d7 b2-b4 a5xb4 a3xb4 Nc6-d4 Ne2xd4 e5xd4 Qc4xd4 f7-f6 Ra1-a7 Bb7-c8
47/68 1:18:17 66,421,843k 14,141k +23.30 e2-e3 Ng8-f6 Ng1-e2 d7-d5 h2-h3 Nb8-c6 d2-d3 e7-e5 a2-a3 Bf8-d6 Nb1-d2 Ke8-f8 g2-g3 b7-b6 Bf1-g2 a7-a5 b2-b3 h7-h6 Bc1-b2 Kf8-g8 O-O Nc6-d8 c2-c4 d5xc4 b3xc4 Bc8-b7 f2-f3 Nf6-d7 Ra1-b1 Bd6-c5 Kg1-f2 Nd8-e6 Qd1-a4 Ne6-f8 Qa4-c2 Nf8-e6 h3-h4 Bc5-d6 Rb1-e1 Nd7-f6 g3-g4 h6-h5 g4-g5 Nf6-d7
48/55 1:18:37 66,699,120k 14,139k +23.30 e2-e3 Ng8-f6 Ng1-e2 d7-d5 h2-h3 Nb8-c6 d2-d3 e7-e5 a2-a3 Bf8-d6 Nb1-d2 Ke8-f8 g2-g3 b7-b6 Bf1-g2 a7-a5 b2-b3 h7-h6 Bc1-b2 Kf8-g8 c2-c4 Bc8-e6 Qd1-c2 Nc6-e7 O-O-O d5xc4 d3xc4 Be6-d7 f2-f3 Ne7-f5 Qc2-c3 Bd6-c5 Nd2-f1 Bc5-d6 b3-b4 a5xb4 a3xb4 Bd7-a4 Rd1-e1 Nf5-e7 h3-h4 Ne7-c6 c4-c5 b6xc5 b4xc5 Bd6-e7 e3-e4
49/64 2:38:23 133,639,033k 14,062k +23.30 e2-e3 Ng8-f6 a2-a3 d7-d5 g2-g3 e7-e5 d2-d3 Bf8-d6 b2-b3 Ke8-f8 Bc1-b2 Nb8-c6 h2-h3 a7-a5 Nb1-d2 b7-b6 Bf1-g2 Bc8-b7 Ng1-e2 Kf8-g8 O-O Nc6-d8 b3-b4 Nd8-c6 b4-b5 Nc6-e7 Qd1-e1 e5-e4 Ra1-c1 Ne7-f5 Nd2-b3 Bd6-e7 Qe1-c3 Be7-d6 a3-a4 Nf5-e7 d3xe4 d5xe4 Qc3-e1 Nf6-d5 Bg2xe4 Nd5xe3 Be4xb7 Ne3xf1 Kg1xf1

[RubiChess]

Simplest explanation would be: Neither Komodo nor Stockfish team wastes training ressources on positions with material imbalance > 20 pawns and all pawns in start position.
Both engines will win the game with white. Chess is about win/draw/loss, not about evaluation being a continuous function of depth. So I wouldn't call it a bug.

Regards, Andreas

[Uri Blass]

They will certainly win but from a user point of view I consider it as a bug because when I analyze a winning position I want to see
better moves.

In the case of stockfish I understood that they even do not use nnue so the behavioue seem to be strange because even weak engines that do not use nnue usually do not show this type of behaviour and you do not need a special evaluation in order to have an improvement when you search deeper.

simply encouraging trading pieces when you are up should be enough but for some reason stockfish does not want to trade pieces when it is up by a big material

I complained about evaluation of 23.69 at depth 1 and 23.30 at depth 49 but the following is only 22.53 at depth 28 and 22.30 at depth 1 with 1 core for some reason.

[fen]1nb1kb2/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQ - 0 1[/fen]

Stockfish_22021019_x64_avx2:
NNUE evaluation using nn-6877cd24400e.nnue enabled
1/1 00:00 19 19k +22.53 e2-e3
2/2 00:00 145 145k +22.38 d2-d3 a7-a6
3/3 00:00 306 306k +22.12 d2-d3 e7-e6 Nb1-d2
4/4 00:00 898 449k +22.61 Nb1-c3 e7-e6 g2-g3 g7-g6
5/5 00:00 2k 797k +22.58 Nb1-c3 a7-a6 Nc3-d5 c7-c5 e2-e3 Nb8-c6
6/6 00:00 4k 1,236k +22.46 Nb1-c3 Nb8-c6 Nc3-b5 Ke8-d8 g2-g3 g7-g6 e2-e3
7/7 00:00 8k 1,611k +22.50 Nb1-c3 Nb8-c6 Nc3-b5 Ke8-d8 c2-c3 e7-e6 f2-f3
8/10 00:00 15k 1,879k +22.42 Nb1-c3 Nb8-c6 e2-e3 e7-e6 Nc3-e4 Bf8-d6 g2-g3 g7-g6 Ne4-f6+ Ke8-f8
9/9 00:00 21k 1,881k +22.43 Nb1-c3 a7-a6 e2-e3 Nb8-c6 Bf1-d3 h7-h6 Bd3-e4 Nc6-e5 O-O
10/13 00:00 51k 1,901k +22.30 Nb1-c3 Nb8-c6 e2-e3 e7-e6 Qd1-h5 h7-h6 Bf1-e2 Bf8-b4 Nc3-e4 Ke8-f8 c2-c3
11/14 00:00 90k 2,054k +22.30 Nb1-c3 Nb8-c6 e2-e3 e7-e6 Bf1-e2 Bf8-b4 Nc3-a4 Ke8-f8 c2-c3 Bb4-d6 O-O
12/15 00:00 134k 2,162k +22.30 Nb1-c3 Nb8-c6 e2-e3 e7-e6 Bf1-d3 Bf8-b4 Nc3-a4 h7-h6 a2-a3 Bb4-d6 O-O Ke8-f8 f2-f4 Kf8-g8
13/18 00:00 225k 2,231k +22.30 Nb1-c3 Nb8-c6 e2-e3 g7-g6 Bf1-c4 e7-e6 O-O Bf8-d6 Bc4-b3 Ke8-f8 Nc3-b5 Kf8-g7
14/19 00:00 489k 2,373k +22.27 d2-d3 Nb8-c6 g2-g3 e7-e6 Nb1-d2 Bf8-d6 Bf1-g2 Ke8-f8 O-O Kf8-g8 e2-e3 Nc6-e7 Nd2-c4 a7-a6
15/17 00:00 549k 2,409k +22.27 d2-d3 Nb8-c6 g2-g3 e7-e6 Nb1-d2 Bf8-d6 Bf1-g2 Ke8-f8 a2-a3 Bd6-e5 O-O g7-g6 Nd2-c4 d7-d5 Nc4-d2
16/20 00:00 686k 2,440k +22.27 d2-d3 Nb8-c6 g2-g3 e7-e6 Nb1-d2 g7-g6 Bf1-g2 Bf8-h6 f2-f4 Ke8-f8 O-O f7-f5 c2-c4 Kf8-g8 e2-e3 Kg8-g7
17/19 00:00 903k 2,501k +22.27 d2-d3 Nb8-c6 g2-g3 e7-e6 Nb1-d2 g7-g6 Bf1-g2 Bf8-h6 f2-f4 Ke8-f8 e2-e3 Nc6-e7 O-O Ne7-f5 Nd2-f3 Kf8-g7 a2-a3 Nf5-d6
18/21 00:00 1,557k 2,579k +22.27 d2-d3 Nb8-c6 g2-g3 g7-g6 Nb1-d2 d7-d6 Bf1-g2 Bf8-g7 O-O Ke8-f8 e2-e3 Bg7-f6 h2-h4 Kf8-g8 a2-a3 Bc8-e6 d3-d4 Bf6-g7
19/20 00:00 2,434k 2,544k +22.23 d2-d3 Nb8-c6 g2-g3 g7-g6 Nb1-d2 d7-d6 Bf1-g2 Bf8-g7 O-O Ke8-f8 e2-e3 Bc8-f5 h2-h4 e7-e5 a2-a3 Kf8-g8 Rf1-e1 Bf5-e6 h4-h5
20/23 00:01 2,726k 2,538k +22.23 d2-d3 Nb8-c6 g2-g3 g7-g6 Nb1-d2 d7-d6 Bf1-g2 Bf8-g7 O-O Ke8-f8 e2-e3 Kf8-g8 d3-d4 h7-h5 h2-h4 Bc8-g4 f2-f3 Bg4-f5 Rf1-e1 a7-a6
21/25 00:01 4,472k 2,554k +22.38 g2-g3 g7-g6 Bf1-g2 Nb8-c6 O-O Bf8-g7 c2-c3 Ke8-f8 d2-d4 d7-d6 Nb1-d2 a7-a6 e2-e3 e7-e5 Qd1-b3 Bg7-h6 f2-f4 Kf8-g8 d4xe5 d6xe5
22/26 00:02 6,035k 2,561k +22.35 g2-g3 g7-g6 Bf1-g2 Nb8-c6 O-O Bf8-g7 c2-c3 Ke8-f8 d2-d4 a7-a6 Nb1-d2 e7-e5 e2-e3 d7-d6 d4-d5 Nc6-e7 Qd1-a4 c7-c6 d5xc6 b7xc6 Qa4-c4 h7-h5
23/27 00:03 8,325k 2,566k +22.30 g2-g3 g7-g6 Bf1-g2 e7-e6 O-O Nb8-c6 d2-d3 Bf8-g7 Nb1-d2 Ke8-f8 c2-c3 Kf8-g8 Qd1-b3 a7-a6 d3-d4 b7-b6 e2-e4 Bc8-b7 Qb3-c4 a6-a5 Qc4-e2 h7-h5 Qe2-c4
24/29 00:04 11,683k 2,489k +22.35 g2-g3 g7-g6 Bf1-g2 e7-e6 O-O Nb8-c6 d2-d4 b7-b6 c2-c3 Bc8-b7 e2-e4 Nc6-a5 Qd1-d3 Bf8-d6 b2-b4 Na5-c6 Nb1-d2 f7-f5 Nd2-c4 f5xe4 Bg2xe4 Ke8-f8
25/29 00:06 14,586k 2,363k +22.30 g2-g3 g7-g6 Bf1-g2 e7-e6 O-O Nb8-c6 d2-d4 b7-b6 c2-c3 Bc8-b7 Qd1-d3 Bf8-d6 Nb1-d2 Ke8-f8 e2-e4 Kf8-g8 a2-a4 a7-a5 f2-f4 Nc6-e7 b2-b3 h7-h5 h2-h4
26/31 00:10 24,791k 2,378k +22.30 g2-g3 e7-e6 Bf1-g2 Bf8-d6 O-O b7-b6 d2-d3 Ke8-f8 Nb1-d2 Nb8-c6 c2-c3 Bc8-b7 Qd1-a4 a7-a6 f2-f4 Kf8-g8 e2-e4 g7-g6 Rf1-f2 b6-b5 Qa4-d1 a6-a5 e4-e5 Bd6-e7 d3-d4 b5-b4 Nd2-e4
27/35 00:12 28,709k 2,364k +22.30 g2-g3 e7-e6 Bf1-g2 Bf8-d6 O-O b7-b6 d2-d3 Ke8-f8 Nb1-d2 Nb8-c6 c2-c3 Bc8-b7 Qd1-a4 a7-a6 f2-f4 Kf8-g8 e2-e4 g7-g6 Rf1-f2 b6-b5 Qa4-d1 a6-a5 e4-e5 Bd6-e7 d3-d4 b5-b4 Nd2-e4 Kg8-g7
28/42 00:17 40,188k 2,342k +22.30 g2-g3 e7-e6 Bf1-g2 Bf8-d6 O-O b7-b6 d2-d3 Ke8-f8 Nb1-d2 Nb8-c6 c2-c3 Bc8-b7 Qd1-a4 a7-a6 h2-h4 g7-g6 Qa4-d1 Kf8-g8 Rf1-e1 Bd6-e7 e2-e4 a6-a5 d3-d4 Bb7-a6 Re1-e3 Kg8-g7 e4-e5 f7-f5 e5xf6/ep+ Be7xf6



the following is only 21.00 pawns at depth 1.
I do not understand the logic behind an evaluation that encourage you not to trade pieces when you have a big advantage espacielly when it is something that humans wrote and not something that is nnue.
Even if stockfish had no code that is against simplification but only code to evaluate the position by piece square table it could be able to have an identical number for trading pieces and hopefully an improvement in the evaluation because of other factors.

[fen]2b1kb2/pppppppp/8/8/8/8/PPPPPPPP/R1BQKB1R w KQ - 0 1[/fen]


Stockfish_22021019_x64_avx2:
NNUE evaluation using nn-6877cd24400e.nnue enabled
1/1 00:00 18 18k +21.00 e2-e3
2/2 00:00 77 39k +20.38 e2-e3 a7-a6
3/3 00:00 319 160k +20.43 d2-d3 a7-a6 Ke1-d2
4/4 00:00 957 479k +20.76 g2-g3 Ke8-d8 a2-a3 e7-e6
5/5 00:00 2k 724k +20.84 f2-f3 c7-c6 Ke1-f2 e7-e6 a2-a3
6/6 00:00 5k 1,272k +20.76 d2-d3 g7-g6 f2-f3 e7-e6 Ke1-f2 Bf8-c5+ e2-e3
7/8 00:00 10k 1,998k +20.69 d2-d3 e7-e6 Bc1-e3 b7-b6 Be3-d4 Bf8-b4+ c2-c3
8/8 00:00 12k 1,962k +20.69 d2-d3 e7-e6 Bc1-e3 b7-b6 c2-c3 Bf8-c5 d3-d4 Bc5-d6
9/14 00:00 30k 2,138k +20.69 f2-f3 e7-e6 Ke1-f2 Bf8-d6 d2-d4 Ke8-f8 c2-c3 Kf8-g8 e2-e3
10/12 00:00 45k 1,968k +20.69 f2-f3 e7-e6 Ke1-f2 Bf8-d6 c2-c3 Bd6-c5+ d2-d4 Bc5-b6 Qd1-a4 Ke8-f8 h2-h3
11/13 00:00 76k 2,102k +20.72 e2-e3 g7-g6 Bf1-e2 Bf8-g7 c2-c3 a7-a6 h2-h3 Ke8-f8 d2-d4
12/16 00:00 125k 2,189k +20.66 e2-e3 g7-g6 Bf1-e2 a7-a6 O-O d7-d6 h2-h4 Bf8-g7 Rf1-e1 Ke8-f8 g2-g3 Bc8-e6
13/16 00:00 188k 2,191k +20.69 e2-e3 g7-g6 Bf1-e2 a7-a6 O-O d7-d6 h2-h4 Bf8-g7 g2-g3 Bc8-e6 a2-a4 Ke8-f8 Kg1-g2 Bg7-e5 d2-d4 Be6-d5+ f2-f3
14/17 00:00 270k 2,284k +20.82 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 c7-c6 Bf1-d3 d7-d6 Bd3-e2 Bc8-e6 Qa4-h4 h7-h5 Qh4-b4 b7-b5 O-O
15/17 00:00 342k 2,325k +20.76 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 Bf8-g7 Bf1-e2 Ke8-f8 O-O d7-d6 Qa4-b4 Kf8-g8 h2-h4 h7-h5 d2-d4
16/19 00:00 377k 2,330k +20.76 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 Bf8-g7 Bf1-e2 Ke8-f8 O-O d7-d6 Qa4-c4 c7-c6 Qc4-b4 Kf8-g8 d2-d4 h7-h5
17/19 00:00 460k 2,346k +20.76 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 e7-e6 Qa4-e4 Bf8-e7 O-O Be7-d6 Qe4-d4 e6-e5 Qd4-d5 Ke8-f8 g2-g3
18/23 00:00 657k 2,379k +20.74 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 e7-e6 Qa4-e4 Bf8-e7 O-O Ke8-f8 g2-g3 Be7-d6 Bd3-e2 Bd6-c5 Kg1-g2 Bc5-b6 d2-d4 Kf8-g7
19/21 00:00 774k 2,396k +20.74 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 e7-e6 Qa4-e4 Bf8-e7 Qe4-c4 Be7-d6 Qc4-d4 Bd6-e7 Qd4-h8+ Be7-f8 O-O d7-d6 Qh8-h7
20/23 00:00 920k 2,402k +20.74 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 e7-e6 O-O Bf8-c5 g2-g3 Bc5-b6 Bd3-e2 Ke8-f8 Qa4-h4 Kf8-e8 a2-a4 d7-d6 d2-d4 Bc8-d7
21/25 00:00 1,486k 2,420k +20.76 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 e7-e6 O-O Bf8-d6 Qa4-d4 b7-b6 Qd4-h8+ Bd6-f8 g2-g3 Bc8-b7 a2-a4 a6-a5 h2-h4 Bb7-d5 b2-b4 a5xb4 c3xb4
22/27 00:00 1,739k 2,429k +20.76 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 Bf8-g7 O-O Bg7-f6 Qa4-a5 Ke8-d8 Qa5-d5 e7-e6 Qd5-c4 d7-d6 Bd3-e2 Bf6-e5 d2-d4 Be5-g7 g2-g3 Bc8-d7 Kg1-g2 Bd7-b5
23/29 00:01 4,212k 2,459k +20.79 e2-e3 g7-g6 c2-c3 a7-a6 Qd1-a4 h7-h5 Bf1-d3 Bf8-g7 O-O Bg7-f6 Qa4-a5 Ke8-d8 Qa5-d5 e7-e6 Qd5-c4 Bf6-e5 Qc4-h4+ Kd8-e8 g2-g3 d7-d5 Qh4-g5 Be5-d6 Kg1-g2 Bc8-d7 a2-a4 Bd6-e7 Qg5-e5
24/31 00:02 5,172k 2,455k +20.82 e2-e3 g7-g6 c2-c3 a7-a6 Bf1-d3 h7-h5 O-O Bf8-g7 Qd1-a4 Bg7-f6 Qa4-c4 c7-c6 Qc4-c5 d7-d6 Qc5-a5 Ke8-d7 g2-g3 Bf6-e5 f2-f4 Be5-f6 f4-f5 g6-g5 Qa5-b4 Bf6-e5 Qb4-e4
25/30 00:02 5,729k 2,456k +20.82 e2-e3 g7-g6 c2-c3 a7-a6 Bf1-d3 h7-h5 O-O Bf8-g7 Qd1-a4 Bg7-f6 Qa4-c4 c7-c6 Qc4-b4 d7-d6 a2-a4 Ke8-d8 a4-a5 Kd8-d7 g2-g3 Kd7-c7 Qb4-b6+ Kc7-d7 f2-f4 e7-e6 b2-b4 d6-d5
26/34 00:02 7,083k 2,446k +20.82 e2-e3 g7-g6 c2-c3 a7-a6 Bf1-d3 h7-h5 O-O Bf8-g7 Qd1-a4 e7-e6 g2-g3 Bg7-e5 f2-f4 Be5-f6 Qa4-c4 Bf6-d8 Qc4-d4 d7-d6 Qd4-h8+ Ke8-e7 Qh8-g7 Ke7-e8 Rf1-f2 Bc8-d7 a2-a4 c7-c5 b2-b4 c5xb4 c3xb4

[Sopel]

Why is e4 better than e3? Can you prove it?

[Raphexon]

What's the issue?

SF thinks it's +23 (massively winning), and after a few seconds of thinking it still thinks it's +23... (massively winning)

[Uri Blass]

Why is e4 better than e3? Can you prove it?

I know that a program that plays e4 (Wasp4.5) mated faster in my tests.

[Uri Blass]

What's the issue?

SF thinks it's +23 (massively winning), and after a few seconds of thinking it still thinks it's +23... (massively winning)

The issue is that the evaluation should go up when you are winning and stockfish seems to have no plan.


I believe that I found the bad patch
A lazy evaluation by itself is not a bad idea but it should be done only when the score is sufficiently large above beta and not every time when it is large.

If the evaluation of the root position is 0.45 then I really do not care if the evaluation of a leaf position is +15 pawns or +20 pawns but if the evaluation fo the root position is +16 I think that I should care so I believe the way this patch is implemented was simply wrong.

Author: mstembera
Date: Fri Jul 17 20:24:02 2020 +0200
Timestamp: 1595010242

Multiple lazy stages.

An extension of the lazy eval idea: when the score is sufficiently large
we now skip more granular parts of the eval.

Inspired by an original patch by Moez Jellouli
https://tests.stockfishchess.org/tests/ ... 3532894529
Credit to him!

STC https://tests.stockfishchess.org/tests/ ... 3532894924
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 13504 W: 2684 L: 2472 D: 8348 Elo +5.45
Ptnml(0-2): 229, 1496, 3111, 1666, 250

LTC https://tests.stockfishchess.org/tests/ ... 353289495b
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 31312 W: 3926 L: 3677 D: 23709 Elo +2.76
Ptnml(0-2): 185, 2773, 9509, 2986, 203


Edit:Not everything is about elo and when it is not important if you mate in 50 moves or in 60 moves in winning positions for rating then it is still important for anlaysis to have something that suggest me winning moves that lead to mate in less moves and I believe it can be done without changing the elo of the program simply by being lazy (because evaluation is expensive) only when you are practically sure that the evaluation is above beta and not being lazy every time when the evaluation is very high.

Note that It is surprising that the lazy stupid evaluation lead to encouraging not trading material because I do not expect this behaviour even from only PST evaluation.

[connor_mcmonigle]

Why is e4 better than e3? Can you prove it?

I know that a program that plays e4 (Wasp4.5) mated faster in my tests.

That doesn't prove that e3 is better than e4 even if e4 leads to a shorter conversion (which it also doesn't prove). Engines only care about the game result, not how long it takes to reach said result. The length of a conversion has zero impact on Elo. Labeling this behavior "evaluation bugs" is quite silly in light of this.

[Cornfed]

Are not you guys trying to use a tool (chess engine) designed to best play a game of chess, based these days largely on NNUE for what is really a non-chess playing situation with these 'odds games'?

I mean, strictly speaking evaluation may not be spot on...but 'who really cares' as the NNUE engine is not designed for those kinds of positions but to play a best possible game of normal chess. I'm just not sure it is fair shout "evaluation bug" all the time. Then again, I suppose one could use anything to sit on and call it a 'chair'...but that kind of does disservice to the good ol' traditional concept of a chair...

Just wondering, that's all...just woke up and my brain tends to the philosophical before it gets cluttered with normal day things.

*** lOL! As I was typing this, a similar comment just got posted.

[FormazChar]

Why would you expect PSQT and material eval to play well in high handicap match? These eval factors hate sacrifices and will want keep the unbalance.