Database snapshot

[zullil]

Interface updated. Leaf node score of top 4 moves are shown (Be patient it is a bit slower due to some server requests). Also it is now possible to save game in pgn format.
https://fsmosca.github.io/ChessDB-Online-Book/

I think the behavior that is causing concern for me (and perhaps others) is something like this.

At the start position, the best move is d4, with a leaf-node score of 15. So I play 1. d4, and then the best reply by leaf-node score is 1...d5. This is followed by 2. c4 and then 2...c6. And now 3. e3 is best, with a leaf-node score of 41.

I think some of us are seeking true mini-max. Scores of terminal nodes are assigned by Stockfish search, and then all other nodes are scored recursively.

Thanks for making the nice web interface, and thanks to noobpwnftw for building the database.

[Ovyron]

Thanks for making the nice web interface, and thanks to noobpwnftw for building the database.

Agreed.

So I guess now the question is if using this database to score (and rank) the top moves in the positions it has is better than, say, using MultiPV=4 at Depth 22 from an engine to do it. Again, I independently found MultiPV=4 to be best (3 providing too few info for positions where a 4th move was required, 5 providing too much unnecessary info not worth the slowdown - though I cheated here by just stopping analysis early if it's analyzing losing moves, so let's say this only includes positions where at least 5 different moves aren't losing), so I like how Ferdy also shows leaf node scores for 4 moves (positions where a 5th move is best and missing would be rare.)

Nothing can be gained from a single line, or two, but for curiosity, I'm going to compare the top lines from this against MultiPV=4 d22.

The database gets white first.

Score of d4 after 38 plies: 26
Score of e4 after 48 plies: 48
Score of Nf3 after 48 plies: 13
Score of c4 after 37 plies: 47

Bug found? The database ranks d4 (score 5) over e4 (score 9) in its Book probe results o_O

1.e4

MultiPV:

22/35 0:25 +0.25 1...e6 2.d4 d5 3.Nc3 Bb4 4.exd5 exd5 5.Bd3 Nf6 6.Nge2 Bd6 7.O-O c6 8.Bf4 Bxf4 9.Nxf4 O-O 10.Re1 Bg4 11.f3 Bd7 12.Nfe2 Re8 13.Qd2 c5 14.dxc5 Qc7 (66.330.604) 2652
22/32 0:25 +0.30 1...e5 2.Nf3 d6 3.d4 exd4 4.Nxd4 Nf6 5.Nc3 Be7 6.Bb5+ c6 7.Be2 O-O 8.O-O c5 9.Nf5 Bxf5 10.exf5 Nc6 11.Bf3 Qd7 12.g4 h6 13.Bg2 Rad8 14.Nd5 Nxd5 (66.330.604) 2652
22/35 0:25 +0.30 1...c5 2.Nf3 e6 3.Nc3 Nc6 4.Bb5 Nd4 5.d3 a6 6.Bc4 b5 7.Bb3 Ne7 8.O-O Nec6 9.Nxd4 cxd4 10.Ne2 Bb7 11.c3 dxc3 12.bxc3 Be7 (66.330.604) 2652
22/33 0:25 +0.63 1...d6 2.d4 e5 3.dxe5 Qe7 4.Nf3 dxe5 5.Bc4 Nc6 6.O-O Nf6 7.Nc3 Bg4 8.Be3 Rd8 9.Nd5 Qd6 10.Nxf6+ Qxf6 11.Bd5 Bxf3 12.Qxf3 Qxf3 13.gxf3 Nd4 (66.330.604) 2652

Not looking good already for MultiPV, I doubt it knows how to play the French...

1.e4 e6

Score of d4 after 46 plies: 48
Score of Nf3 after 26 plies: 20
Score of Nc3 after 31 plies: 33
Score of c4 after 27 plies: 0

1.e4 e6 2.d4

22/32 0:35 +0.08 2...d5 3.exd5 exd5 4.Nf3 Nf6 5.Bd3 c5 6.O-O c4 7.Be2 Nc6 8.b3 cxb3 9.axb3 Bd6 10.c4 O-O 11.Nc3 Re8 12.Bg5 Nb4 13.Ne5 Bxe5 14.dxe5 Rxe5 15.Bh4 Bf5 (89.050.767) 2518
22/30 0:35 +0.42 2...a6 3.c3 d5 4.exd5 exd5 5.Bd3 Bd6 6.Nf3 Nf6 7.Be3 O-O 8.O-O Re8 9.h3 h6 10.Nbd2 Bd7 11.c4 Nc6 12.a3 dxc4 13.Nxc4 (89.050.767) 2518
22/33 0:35 +0.43 2...Be7 3.Be2 d5 4.e5 c5 5.c3 Nc6 6.Nf3 f6 7.O-O fxe5 8.dxe5 Qb6 9.Na3 Nh6 10.c4 d4 11.Bxh6 gxh6 12.Nb5 O-O 13.Bd3 (89.050.767) 2518
22/34 0:35 +0.58 2...h6 3.Nf3 (89.050.767) 2518

1.e4 e6 2.d4 d5

Score of Nd2 after 44 plies: 48
Score of a6 after 45 plies: 48
Score of c5 after 45 plies: 48
Score of e5 after 23 plies: 0

Bug found? In this position 3.a6 and 3.c5 aren't legal white moves.

1.e4 e6 2.d4 d5 3.Nd2

22/33 0:18 0.00 3...c5 4.exd5 exd5 5.Ngf3 Nf6 6.Bb5+ Bd7 7.Bxd7+ Nbxd7 8.O-O Be7 9.dxc5 Nxc5 10.Nd4 O-O 11.Nf5 Re8 12.Nf3 Nce4 13.Nxe7+ Rxe7 14.Be3 Nd6 15.Bd4 Re6 16.Be5 Nde4 (49.557.141) 2688
22/35 0:18 +0.07 3...Nf6 4.e5 Nfd7 5.c3 c5 6.Bd3 Nc6 7.Ne2 cxd4 8.cxd4 f6 9.exf6 Nxf6 10.O-O e5 11.dxe5 Nxe5 12.Bb5+ Bd7 13.Re1 Bd6 14.Bxd7+ Qxd7 15.Nc3 O-O 16.Nf1 Nc6 (49.557.141) 2688
22/32 0:18 +0.41 3...Be7 4.Ngf3 Nf6 5.Bd3 c5 6.e5 Nfd7 7.c3 Nc6 8.O-O g5 9.dxc5 g4 10.Nd4 Ndxe5 11.Bc2 Bxc5 12.N2b3 Bd6 13.Re1 Qf6 14.Qe2 O-O 15.a4 (49.557.141) 2688
22/31 0:18 +0.64 3...a6 4.Ngf3 c5 5.dxc5 Bxc5 6.Bd3 Ne7 7.O-O O-O 8.Nb3 dxe4 9.Bxe4 Qxd1 10.Rxd1 Bb6 11.Be3 Bxe3 12.fxe3 f5 13.Bd3 Nd7 14.e4 fxe4 15.Bxe4 Nf6 16.Bd3 (49.557.141) 2688

So this is the Tarrasch. I'd expect that now (soon enough), either MultiPV is going to play a move missing from the db, or the db is going to play moves that are going to make MultiPV regret its choices.

1.e4 e6 2.d4 d5 3.Nd2 c5

Score of Ngf3 after 42 plies: 48
Score of exd5 after 42 plies: 48
Score of dxc5 after 29 plies: -6
Score of Bb5+ after 14 plies: 0

A tie in score. Here the regular score that takes into account how likely is black to blunder is helpful, and Ngf3 (score 11) beats exd5 (score 8)

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3

22/37 0:14 +0.20 4...Nf6 5.exd5 exd5 6.Bb5+ Bd7 7.Qe2+ Be7 8.dxc5 O-O 9.Nb3 Qc7 10.O-O Re8 11.Be3 Bxb5 12.Qxb5 Nbd7 13.a4 h6 14.Rfe1 Rac8 15.Rad1 Bf8 16.c3 Re4 17.g3 Nxc5 (38.912.583) 2728
22/32 0:14 +0.34 4...Nc6 5.exd5 exd5 6.Bb5 Bd6 7.dxc5 Qe7+ 8.Qe2 Qxe2+ 9.Bxe2 Bxc5 10.O-O Nf6 11.Nb3 Bb6 12.Bg5 Ne4 13.Bf4 O-O 14.a4 a5 15.c3 Bd7 16.Nbd4 Nxd4 17.Nxd4 Rac8 (38.912.583) 2728
22/34 0:14 +0.35 4...a6 5.dxc5 Bxc5 6.Bd3 Nf6 7.O-O Nc6 8.Qe2 Qc7 9.c3 O-O 10.h3 dxe4 11.Nxe4 Nxe4 12.Bxe4 Be7 13.Bg5 Bd7 14.Bxe7 Nxe7 15.Rfe1 Kh8 16.Rad1 Rad8 17.Bc2 (38.912.583) 2728
22/33 0:14 +0.35 4...cxd4 5.exd5 Qxd5 6.Bc4 Qd6 7.O-O Nf6 8.Nb3 Nc6 9.Nbxd4 Nxd4 10.Nxd4 Qc7 11.Qe2 Bd7 12.Nb5 Qc6 13.Bg5 Be7 14.Rfd1 O-O 15.Bd3 a6 16.Nc3 Rfd8 17.Ne4 e5 (38.912.583) 2728

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6

Score of exd5 after 27 plies: 12
Score of e5 after 28 plies: 55
Score of Bb5+ after 20 plies: -15
Score of Bd3 after 3 plies: -100

Aha! So here's a position that shows why having access to minmax is cool. exd5 is ranked best in the database with score 15, while e5 is ranked second with score 4. However, if MultiPV happens to play the best black moves that the database already knows, we end in a leaf node with score 12. No matter what black plays, e5 guarantees a leaf node with score 55 (what white has been chasing.) e5 is the critical variation, so I hold it should be ranked first by Book probe results (even if scores couldn't be compared between nodes).

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5

22/34 0:21 -0.03 5...Nfd7 6.c3 Nc6 7.Bd3 g5 8.h3 h5 9.b3 Rg8 10.g4 Rh8 11.O-O a6 12.Bb2 hxg4 13.hxg4 b5 14.a3 c4 15.Bc2 cxb3 16.Nxb3 a5 17.Qe2 a4 18.Nbd2 (55.984.352) 2550
22/35 0:21 +0.59 5...Ne4 6.Bd3 Nxd2 7.Bxd2 cxd4 8.Nxd4 Nc6 9.Nxc6 bxc6 10.O-O Rb8 11.b3 g6 12.Qg4 Bg7 13.Rae1 Qc7 14.Qg3 c5 15.c4 O-O 16.Bf4 d4 17.h4 h5 18.Bg5 Bb7 (55.984.352) 2550
22/32 0:21 +0.85 5...Ng8 6.Bd3 Nc6 7.O-O cxd4 8.Nb3 Nge7 9.Nbxd4 Nxd4 10.Nxd4 Nc6 11.Nxc6 bxc6 12.b3 Rb8 13.c3 c5 14.Re1 h5 15.h3 Bd7 16.Bd2 Bc6 17.c4 d4 (55.984.352) 2550
22/33 0:21 +0.98 5...Ng4 6.Bd3 Nc6 7.h3 c4 8.Bxc4 Ngxe5 9.dxe5 dxc4 10.O-O b5 11.a4 Ba6 12.Ne4 b4 13.Nd6+ Bxd6 14.exd6 O-O 15.Re1 Qd7 16.Bf4 Rad8 17.c3 h6 18.Qc2 (55.984.352) 2550

Getting a bit optimistic here...

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7

Score of c3 after 26 plies: 25
Score of c4 after 4 plies: -29
Score of Bd3 after 16 plies: -46
Score of dxc5 after 6 plies: -22

Oh wait, white's score isn't a guaranteed 55 or 48 anymore...

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3

22/29 0:23 +0.16 6...Nc6 7.Be2 Be7 8.O-O a5 9.a4 O-O 10.Nb3 cxd4 11.cxd4 b6 12.Bb5 Na7 13.Bd3 Nc6 14.Qe2 Nb4 15.Bb5 Ba6 16.Bd2 Bxb5 17.Qxb5 h6 18.Rfc1 (61.449.434) 2656
22/30 0:23 +0.25 6...Be7 7.a3 a5 8.Bd3 Qb6 9.O-O cxd4 10.cxd4 Nc6 11.Qa4 O-O 12.Re1 f6 13.exf6 Nxf6 14.Nb3 Qc7 15.h3 Bd7 16.Bd2 Nb4 17.Bb5 Nc6 18.Rac1 Ne4 (61.449.434) 2656
22/28 0:23 +0.32 6...a5 7.Bb5 Nc6 8.a4 Be7 9.O-O O-O 10.b3 Na7 11.Bd3 Nc6 12.Nb1 cxd4 13.cxd4 Nb4 14.Be2 Nb6 15.Nc3 Bd7 16.Bd2 Rc8 17.Qb1 h6 18.Rc1 Rc6 (61.449.434) 2656
22/28 0:23 +0.40 6...Qb6 7.b3 Nc6 8.Bb2 Be7 9.Bd3 a5 10.a4 cxd4 11.cxd4 Nb4 12.Bb5 Qd8 13.O-O O-O 14.Qe2 b6 15.Rfc1 Ba6 16.Bxa6 Rxa6 17.h3 Ra8 18.Ba3 (61.449.434) 2656

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6

Score of Bd3 after 24 plies: 25
Score of Be2 after 15 plies: 0
Score of Rb1 after 9 plies: 0
Score of Nb3 after 12 plies: -13

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3

22/35 0:31 -0.04 7...Qb6 8.O-O cxd4 9.cxd4 a5 10.Qa4 Be7 11.Bb5 O-O 12.Re1 f6 13.exf6 Nxf6 14.Nb3 Ne4 15.Be3 Bd7 16.Rac1 Qd8 17.Kh1 Bd6 18.Kg1 h6 19.Nc5 Bxc5 20.dxc5 Qf6 (73.321.421) 2292
22/35 0:31 +0.35 7...a5 8.O-O Be7 9.a4 O-O 10.Nb1 cxd4 11.cxd4 Nb4 12.Nc3 Nxd3 13.Qxd3 b6 14.Nb5 Ba6 15.Be3 Rc8 16.Rac1 Re8 17.h3 Nf8 18.Rxc8 Qxc8 19.Rc1 Qd7 (73.321.421) 2292
22/33 0:31 +0.38 7...Be7 8.O-O a5 9.a4 O-O 10.Nb1 cxd4 11.cxd4 Nb4 12.Nc3 Nxd3 13.Qxd3 b6 14.Nb5 Ba6 15.Be3 Rc8 16.Rac1 Rxc1 17.Rxc1 Re8 18.h3 Bxb5 19.Qxb5 Nf8 20.Qc6 Ng6 (73.321.421) 2292
22/35 0:31 +0.38 7...h5 8.h3 Be7 9.b3 Qc7 10.Bb2 g5 11.Rc1 b6 12.h4 gxh4 13.Nxh4 Bb7 14.g3 c4 15.Be2 O-O-O 16.f4 Rdg8 17.Kf2 f6 18.bxc4 dxc4 19.Bxc4 fxe5 20.Bxe6 exf4 (73.321.421) 2292

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6

Score of O-O after 22 plies: 25
Score of a4 after 14 plies: -31
Score of a3 after 24 plies: -12
Score of Qe2 after 8 plies: -29

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O

22/33 0:13 +0.19 8...cxd4 9.cxd4 a5 10.Re1 a4 11.h3 Be7 12.a3 Nxd4 13.Nxd4 Qxd4 14.Nf3 Qb6 15.Be3 Qxb2 16.Bd4 Qb3 17.Qd2 Nb6 18.Reb1 Nc4 19.Bxc4 Qxc4 20.Rc1 Qb5 21.g3 (37.971.691) 2760
22/35 0:13 +0.27 8...Be7 9.Re1 cxd4 10.cxd4 Nxd4 11.Nxd4 Qxd4 12.Nf3 Qb6 13.a4 Qc7 14.a5 Nc5 15.Bc2 O-O 16.Bd2 Bd7 17.Rc1 Qb8 18.Be3 Na6 19.Qd3 g6 20.Bh6 Rd8 21.Qd2 Nc5 (37.971.691) 2760
22/30 0:13 +0.36 8...g6 9.dxc5 Nxc5 10.Nb3 Nxd3 11.Qxd3 Bg7 12.Re1 O-O 13.Bf4 h6 14.Be3 Qc7 15.Bc5 Rd8 16.Qe3 b6 17.Ba3 Ba6 18.Bd6 Qb7 19.Nbd4 Rdc8 20.Qd2 (37.971.691) 2760
22/30 0:13 +0.62 8...a5 9.dxc5 Nxc5 10.Bc2 Be7 11.Nb3 Ne4 12.c4 dxc4 13.Bxe4 cxb3 14.Be3 Qb5 15.Qxb3 Qxb3 16.axb3 O-O 17.Rfc1 Nb4 18.Rd1 Rd8 19.Bb6 Rxd1+ 20.Rxd1 g6 (37.971.691) 2760

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4

Score of cxd4 after 20 plies: 25
Score of Re1 after 1 plies: -56
Score of Qe2 after 1 plies: -106
Score of c4 after 1 plies: -109

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4

22/34 0:13 +0.10 9...Nxd4 10.Nxd4 (38.686.676) 2787
22/37 0:13 +0.13 9...a5 10.Re1 a4 11.a3 Nxd4 12.Nxd4 Qxd4 13.Nf3 Qb6 14.Be3 Bc5 15.Rb1 h6 16.Bc2 Bxe3 17.Rxe3 Nc5 18.Nd4 O-O 19.Rg3 Ne4 20.Bxe4 dxe4 21.Qg4 g6 22.Qxe4 Rd8 (38.686.676) 2787
22/33 0:13 +0.72 9...Be7 10.Nb3 a5 11.Be3 Nb4 12.Bb1 Nc6 13.Qc2 a4 14.Nc1 a3 15.b3 Nf8 16.h4 Bd7 17.Nd3 h5 18.Rc1 Ng6 19.g3 Rd8 20.Qe2 O-O 21.Nc5 Qc7 22.Bxg6 fxg6 (38.686.676) 2787
22/34 0:13 +0.78 9...Nb4 10.Bb1 Nc6 11.Nb3 a5 12.a4 Be7 13.Bd2 Qd8 14.Bd3 O-O 15.Bb5 Nb4 16.Bxb4 Bxb4 17.Rc1 h6 18.Qc2 b6 19.Qe2 Bb7 20.Rc2 Rc8 21.Rxc8 Bxc8 22.Rc1 (38.686.676) 2787

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4

Score of Nxd4 after 30 plies: 15
Score of h3 after 1 plies: -62
Score of a3 after 1 plies: -90
Score of Qa4 after 1 plies: -104

Huh, oh, white's score already drifting down...

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4

22/36 0:16 +0.51 10...Qxd4 11.Nf3 Qb6 12.a3 Nc5 13.Bc2 Nd7 14.Rb1 a5 15.Be3 Bc5 16.Re1 Qd8 17.Qd2 b6 18.Bg5 Be7 19.Qf4 h6 20.Bxe7 Qxe7 (47.345.001) 2889
22/33 0:16 +2.80 10...Nxe5 11.N2f3 Nxd3 12.Qxd3 Be7 13.Be3 Qa6 14.Nb5 O-O 15.Qb3 Qa5 16.Rfc1 Bd7 17.Rc7 Bxb5 18.Rxe7 Bc4 19.Qd1 Qb4 20.Rc7 Qxb2 21.Bd4 Qe2 22.Ne5 Qxd1+ 23.Rxd1 (47.345.001) 2889
22/33 0:16 +4.43 10...Be7 11.N2f3 Nc5 12.Be3 Nxd3 13.Qxd3 Qa6 14.Qxa6 bxa6 15.Nc6 f6 16.Rac1 Bd7 17.Rc3 Bd8 18.Nxd8 Kxd8 19.Bd4 Ke7 20.Rc7 Rhc8 21.exf6+ gxf6 22.Bxf6+ Ke8 23.Rfc1 Rxc7 (47.345.001) 2889
22/32 0:16 +4.62 10...Nc5 11.Bc2 Bd7 12.Rb1 a5 13.Qh5 Nd3 14.Bxd3 Qxd4 15.Qe2 Be7 16.Nb3 Qh4 17.Be3 O-O 18.f4 g6 19.Nd4 Rfc8 20.Rbc1 Rxc1 21.Rxc1 Rc8 22.Rxc8+ Bxc8 23.Bb5 Qh5 (47.345.001) 2889

Whoa! Looks like db will win this experiment. MultiPV would now rather play some other move instead of 1...e6.

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4

Score of Nf3 after 28 plies: 15
Score of Nb3 after 5 plies: -85
Score of Bb5 after 1 plies: -187
Score of Nb1 after 1 plies: -198

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3

22/35 0:12 +0.30 11...Qb6 12.a3 Nc5 13.Bc2 Nd7 14.Rb1 a5 15.Re1 Bc5 16.Be3 Qd8 17.Bxc5 Nxc5 18.Qd4 Bd7 19.Rbc1 Ne4 20.Bxe4 dxe4 21.Qxe4 Bc6 22.Qg4 O-O 23.Nd4 Qd5 24.Nxc6 bxc6 (33.869.142) 2776
22/39 0:12 +0.61 11...Qg4 12.h3 Qh5 13.Bf4 Be7 14.Kh2 f6 15.exf6 Nxf6 16.Bb5+ Bd7 17.Qe2 Qf5 18.Bg3 O-O 19.Bxd7 Nxd7 20.Nd4 Qe4 21.Qxe4 dxe4 22.Nxe6 Rf7 23.Nc7 Rc8 24.Rfe1 Bf6 (33.869.142) 2776
22/36 0:12 +0.87 11...Qb4 12.a3 Qa5 13.Bd2 Qd8 14.Qa4 Be7 15.Qg4 Kf8 16.b4 Nb6 17.Rfc1 h5 18.Qf4 Bd7 19.h4 Kg8 20.Be3 Be8 21.Ng5 Bxg5 22.hxg5 g6 23.g3 Rc8 24.Rxc8 Qxc8 (33.869.142) 2776
22/28 0:12 +1.07 11...Qc5 12.Bd2 (33.869.142) 2776

This is just way too jumpy.

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6

Score of a3 after 26 plies: 15
Score of Re1 after 14 plies: 31
Score of b3 after 4 plies: 13
Score of Qa4 after 12 plies: 30

Here's another one. 12.a3 with score 21, 12.Re1 with score 15.

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6 12.Re1

22/31 0:20 +0.12 12...Be7 13.a3 Nc5 14.Bc2 a5 15.Bg5 Qd8 16.Be3 Bd7 17.Qd4 a4 18.Qg4 Kf8 19.Qf4 Rc8 20.Rac1 h5 21.Bxc5 Rxc5 22.Bxa4 Rc8 23.Bxd7 (55.387.087) 2682
22/31 0:20 +0.13 12...Bc5 13.Re2 a5 14.Qa4 O-O 15.Qh4 f5 16.exf6 Nxf6 17.Bg5 Ne4 18.Bxe4 dxe4 19.Qxe4 Bd7 20.Be3 Bxe3 21.Rxe3 Qb4 22.Ng5 Qxe4 23.Rxe4 Bc6 24.Re3 Rae8 25.Nxe6 (55.387.087) 2682
22/34 0:20 +0.19 12...h6 13.a3 Nc5 14.Bc2 a5 15.Be3 Qc7 16.Qd4 a4 17.Rac1 Bd7 18.Qg4 Rc8 19.Bxa4 Bxa4 20.Bxc5 Bxc5 21.Qxg7 Rf8 22.b4 Bxf2+ 23.Kxf2 Qxc1 24.Rxc1 Rxc1 25.h4 Rc2+ (55.387.087) 2682
22/34 0:20 +0.27 12...Bb4 13.Re2 Be7 14.Rc2 Qd8 15.Bf4 Nc5 16.Bf1 Bd7 17.Rac1 b6 18.Nd4 O-O 19.b4 Ba4 20.bxc5 bxc5 21.Nb3 c4 22.Qg4 Qb8 23.Nd4 Bxc2 24.Rxc2 Kh8 25.Nc6 Qe8 (55.387.087) 2682

MultiPV is happy again.

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6 12.Re1 Be7

Score of a3 after 12 plies: 31
Score of Rb1 after 4 plies: 21
Score of Qc2 after 1 plies: 0
Score of Qa4 after 21 plies: 0

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6 12.Re1 Be7 13.a3

22/32 0:20 +0.16 13...Qd8 14.Be3 Nc5 15.Bc2 a5 16.Qd4 Bd7 17.Qg4 Kf8 18.h4 h6 19.Rac1 a4 20.g3 Kg8 21.Red1 Nb3 22.Bxb3 axb3 23.Rd3 Ra4 24.Rd4 Qe8 (56.783.761) 2710
22/31 0:20 +0.19 13...Nc5 14.Bc2 a5 15.Qd4 Bd7 16.Rb1 a4 17.Qg4 Bf8 18.Be3 h6 19.Nd4 Qc7 20.Qf4 g5 21.Qg3 Bg7 22.Bd2 Qb6 23.Bc3 Ne4 24.Bxe4 dxe4 25.Rxe4 O-O 26.f4 (56.783.761) 2710
22/38 0:20 +0.30 13...O-O 14.Be3 Qd8 15.Qa4 f5 16.exf6 Nxf6 17.Bd4 Bd7 18.Qd1 Ng4 19.h3 Nh6 20.Rc1 Bd6 21.Qe2 Bf4 22.Rc3 Nf5 23.Be5 Bxe5 24.Nxe5 Rc8 25.Nxd7 Qxd7 26.Rxc8 Rxc8 (56.783.761) 2710
22/36 0:20 +0.45 13...a5 14.Qa4 Qd8 15.Be3 O-O 16.Rad1 f5 17.exf6 Nxf6 18.Qd4 Bd6 19.Bd2 Bd7 20.Rc1 Rc8 21.Rxc8 Bxc8 22.Qh4 h6 23.Bc3 Bd7 24.Ne5 Be8 25.Ng6 (56.783.761) 2710

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6 12.Re1 Be7 13.a3 Qd8

Score of Nd4 after 1 plies: 60

Only one move in the db, but score jumps up!

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6 12.Re1 Be7 13.a3 Qd8 14.Nd4

22/32 0:15 0.00 14...O-O 15.b4 Re8 16.Bd2 Nb6 17.Qh5 g6 18.Qh6 Bd7 19.Nf3 Bf8 20.Qh3 Be7 (39.931.713) 2617
22/26 0:15 +0.03 14...Nc5 15.Bc2 a5 16.Qg4 Kf8 17.Bd2 a4 18.Re3 h5 19.Qe2 Bd7 20.Rf3 Kg8 21.Qe3 Ne4 22.Qf4 Be8 23.Bxe4 dxe4 24.Qxe4 Qb6 (39.931.713) 2617
22/40 0:15 +0.26 14...Bc5 15.Qg4 Bxd4 16.Qxd4 O-O 17.b4 f6 18.exf6 Qxf6 19.Bb2 Qxd4 20.Bxd4 Rf4 21.Bb2 Nb6 22.Be5 Rf7 23.Rac1 Bd7 24.f3 Nc4 25.Bxc4 dxc4 26.Rxc4 a5 27.Rc2 axb4 (39.931.713) 2617
22/29 0:15 +0.32 14...Kf8 15.Qg4 Nc5 16.Bc2 a5 17.Be3 Bd7 18.b4 axb4 19.axb4 Na4 20.Qf4 Nb6 21.Nf3 Rc8 22.Rec1 Nc4 23.Ra7 Nxe3 24.Qxe3 Bxb4 (39.931.713) 2617

Database is out of moves

So this is our leaf node:

1.e4 e6 2.d4 d5 3.Nd2 c5 4.Ngf3 Nf6 5.e5 Nfd7 6.c3 Nc6 7.Bd3 Qb6 8.O-O cxd4 9.cxd4 Nxd4 10.Nxd4 Qxd4 11.Nf3 Qb6 12.Re1 Be7 13.a3 Qd8 14.Nd4 O-O

[d]r1bq1rk1/pp1nbppp/4p3/3pP3/3N4/P2B4/1P3PPP/R1BQR1K1 w - -

Now, we need an impartial judge to tell us how we did. Let's say... SinglePV Depth 36?

33/48 1:30 0.00 15.Bf4 Nc5 16.Bxh7+ Kxh7 17.Qh5+ Kg8 18.Re3 f6 19.Rh3 Ne4 20.Qh8+ Kf7 21.Qh5+ (257.529.705) 2846
34/15 1:46 0.00 15.Bf4 Nc5 16.Bxh7+ Kxh7 17.Qh5+ Kg8 18.Re3 f6 19.Rh3 Ne4 20.Rc1 fxe5 21.Qh7+ Kf7 22.Qh5+ (306.272.138) 2867
35/47 2:14 0.00 15.Bf4 Nc5 16.Bxh7+ Kxh7 17.Qh5+ Kg8 18.Re3 f6 19.Rh3 Ne4 20.Qh8+ Kf7 21.Qh5+ (389.478.447) 2885
36/16 2:24 0.00 15.Bf4 Nc5 16.Bxh7+ Kxh7 17.Qh5+ Kg8 18.Re3 f6 19.Rh3 Ne4 20.Rc1 fxe5 21.Qh7+ Kf7 22.Qh5+ (417.272.506) 2891

The database successfully manages to reach a highly scored leaf node. What I'd like to know is from where the 60 comes from (apparently 54.53% winrate from this position).

Now, instead of repeating this with colors reversed, I want to know how top scoring "Book probe results" moves (top ranked) do. 5.exd5 was the first disagreement.

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5

22/34 0:17 +0.11 5...exd5 6.Bb5+ Bd7 7.Bxd7+ Nbxd7 8.O-O Be7 9.dxc5 Nxc5 10.Nb3 Nce4 11.Nbd4 O-O 12.Nf5 Re8 13.Nxe7+ Rxe7 14.Be3 Nd6 15.c3 Rc8 16.Qc2 b6 17.Rfe1 Qd7 18.Bg5 Rxe1+ (48.280.469) 2790
22/36 0:17 +0.18 5...Nxd5 6.Bc4 Nc6 7.O-O cxd4 8.Nb3 Be7 9.Nbxd4 Nxd4 10.Qxd4 O-O 11.Rd1 Nb6 12.Qxd8 Rxd8 13.Rxd8+ Bxd8 14.Bb3 Bd7 15.Be3 Bc7 (48.280.469) 2790
22/32 0:17 +0.34 5...Qxd5 6.Bc4 Qd6 7.dxc5 Qxc5 8.O-O Be7 9.Qe2 O-O 10.Re1 Nbd7 11.Bd3 Qc7 12.Ne4 b6 13.Nxf6+ Nxf6 14.Ne5 Bb7 15.Bf4 Bd6 16.Rad1 Rac8 17.c4 Kh8 18.a3 h6 (48.280.469) 2790
22/34 0:17 +0.55 5...cxd4 6.dxe6 Bxe6 7.Nb3 Nc6 8.Nbxd4 Bc5 9.Be3 Nxd4 10.Nxd4 Bd7 11.Bc4 O-O 12.O-O Qc7 13.h3 Rad8 14.Qe2 Bxd4 15.Bxd4 Bg4 16.Qd3 Bf5 17.Qxf5 Rxd4 18.Bb3 Rh4 (48.280.469) 2790

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5

Bb5+ 20 51.51
c3 0 50.00
Be2 0 50.00
dxc5 -7 49.47

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5

22/31 0:11 +0.19 6...Bd7 7.Bxd7+ Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1 Re8 11.dxc5 Nxc5 12.Be3 a5 13.c3 a4 14.Bd4 Ne6 15.Be5 Bd6 16.h3 Nf4 17.Bxf6 Rxe1 18.Bxd8 Rxd1 19.Rxd1 Rxd8 (33.126.598) 2907
22/35 0:11 +0.54 6...Nc6 7.O-O Be6 8.Qe2 Qb6 9.dxc5 Bxc5 10.Nb3 O-O 11.Nxc5 Qxc5 12.c3 d4 13.cxd4 Qd6 14.Be3 a6 15.Bd3 Rfe8 16.h3 Bd5 17.Rfc1 Qd7 18.Re1 Rad8 19.a3 h6 (33.126.598) 2907
22/38 0:11 +0.69 6...Nbd7 7.O-O a6 8.Bxd7+ Bxd7 9.Re1+ Be7 10.dxc5 O-O 11.Nb3 Re8 12.Ne5 Rc8 13.Bg5 Bxc5 14.Nxc5 Rxc5 15.c3 Be6 16.Qd4 Rc8 17.Rad1 h6 18.Bxf6 gxf6 19.Nd3 Rc4 (33.126.598) 2907
22/30 0:11 +0.75 6...Nfd7 7.dxc5 a6 8.Bxd7+ Nxd7 9.b4 a5 10.c3 Qe7+ 11.Qe2 Qxe2+ 12.Kxe2 Be7 13.Bb2 O-O 14.Rhe1 Bf6 15.a3 Ne5 16.Nxe5 Bxe5 17.Nf3 Bg4 18.Kf1 Bf6 19.Nd4 Bd7 (33.126.598) 2907

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7

Bxd7+ 17 51.29
a4 15 51.14
Qe2+ 5 50.38
Be2 0 50.00

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7

22/36 0:12 +0.25 7...Nbxd7 8.O-O Be7 9.dxc5 Nxc5 10.Nb3 Nce4 11.Nbd4 O-O 12.Nf5 Re8 13.Nxe7+ Rxe7 14.Be3 Qc7 15.Bd4 Rae8 16.c3 h6 17.Re1 b6 18.h3 Nd7 19.Qb3 Ndc5 20.Bxc5 Qxc5 (36.041.755) 2946
22/40 0:12 +0.36 7...Qxd7 8.O-O cxd4 9.Ne5 Qc8 10.Ndf3 Bc5 11.Nd3 O-O 12.Nxc5 Qxc5 13.Nxd4 Nc6 14.Be3 Qb4 15.Rb1 Rfe8 16.c3 Qd6 17.h3 Qd7 18.Ne2 Re4 19.Ng3 Ree8 20.Re1 (36.041.755) 2946
22/30 0:12 +0.52 7...Nfxd7 8.O-O Be7 9.dxc5 Nxc5 10.Nb3 Ne6 11.Be3 Nc6 12.c3 O-O 13.Re1 Re8 14.Qc2 Qc7 15.Rad1 Rad8 16.g3 Bf6 17.Qf5 Qd7 18.h4 b6 19.Qg4 Kh8 20.Kg2 g6 (36.041.755) 2946
22/29 0:12 +2.15 7...Kxd7 8.O-O cxd4 9.Nxd4 Nc6 10.N2f3 Qb6 11.c3 Re8 12.Nf5 Kc8 13.Bg5 Re6 14.Bxf6 Rxf6 15.Ne3 Re6 16.Ng5 Rxe3 17.Qg4+ Kb8 18.fxe3 (36.041.755) 2946

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7

O-O 19 51.44
dxc5 5 50.38
h3 4 50.30
Qe2+ -7 49.47

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O

22/25 0:15 +0.11 8...Be7 9.dxc5 Nxc5 10.Nb3 Nce4 11.Nbd4 O-O 12.Nf5 Bc5 13.Be3 Bxe3 14.Nxe3 Re8 15.h3 Qa5 16.c3 Rad8 17.Re1 Qc7 18.Qb3 a6 19.Rad1 Nc5 20.Qc2 Nce4 21.Nf5 (43.609.980) 2902
22/29 0:15 +0.24 8...Rc8 9.dxc5 Nxc5 10.Re1+ Be7 11.Nf1 O-O 12.c3 Re8 13.Be3 Bd6 14.Qc2 b5 15.Rad1 Ne6 16.Ng3 b4 17.Nf5 bxc3 18.bxc3 Bb8 19.Bd4 a6 20.Ne3 Nxd4 21.Rxd4 (43.609.980) 2902
22/35 0:15 +0.26 8...a5 9.Re1+ Be7 10.Nf1 O-O 11.c3 Bd6 12.Bg5 c4 13.Ne3 Qc7 14.Ne5 Bxe5 15.dxe5 Qxe5 16.Bxf6 Nxf6 17.Nxc4 Qc7 18.Ne3 Rad8 19.Qd4 Rfe8 20.Rad1 b5 21.Qd3 b4 (43.609.980) 2902
22/36 0:15 +1.08 8...h6 9.Re1+ Be7 10.Qe2 Rc8 11.dxc5 Nxc5 12.Nd4 Ne6 13.N2b3 Qd7 14.c3 O-O 15.Nxe6 fxe6 16.Qxe6+ Qxe6 17.Rxe6 Kf7 18.Re1 b5 19.Be3 a6 20.h3 b4 21.cxb4 Bxb4 (43.609.980) 2902

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7

dxc5 16 51.21
Re1 16 51.21
c3 11 50.83
Ne5 9 50.68

Ohh, I didn't expect this, a tie in scores! So now, leaf nodes tie break. dxc5 has 21 score, Re1 has 29 score (along with illegal 9.a5 and 9.a6.)

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1

22/30 0:11 +0.12 9...O-O 10.Nf1 Re8 11.c3 Qb6 12.Qc2 Rac8 13.dxc5 Nxc5 14.Bg5 Ne6 15.Bxf6 Bxf6 16.Ne3 Rcd8 17.Qf5 Qxb2 18.Nxd5 Rxd5 19.Qxd5 Rd8 20.Qf5 Qxc3 21.Rec1 Qd3 22.Qxd3 Rxd3 (34.233.759) 2920
22/30 0:11 +0.33 9...Rc8 10.dxc5 Nxc5 11.Qe2 Ne6 12.Nb3 O-O 13.c3 a6 14.Nbd4 Nxd4 15.Nxd4 Bc5 16.Be3 Qd7 17.h3 Rfe8 18.Qf3 Re5 19.Rad1 Rce8 20.g4 Qa4 21.b3 (34.233.759) 2920
22/35 0:11 +0.67 9...a5 10.Qe2 Ra6 11.Qb5 O-O 12.Qxb7 Rb6 13.Qa7 cxd4 14.a3 d3 15.Qxa5 dxc2 16.Qa4 Re6 17.Qxc2 Bd6 18.Rxe6 fxe6 19.b4 e5 20.Qf5 Qb8 21.Qh3 Kh8 22.Bb2 Rc8 (34.233.759) 2920
22/39 0:11 +0.81 9...h6 10.Qe2 Rc8 11.b3 Rc6 12.Bb2 Re6 13.Qd3 Rxe1+ 14.Rxe1 O-O 15.Qb5 b6 16.Qa6 Bd6 17.Qxa7 Re8 18.Rxe8+ Qxe8 19.Qa6 Bf4 20.Qb5 Bxd2 21.dxc5 Ne4 22.c6 Nf8 (34.233.759) 2920

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O

Nf1 14 51.06
dxc5 8 50.61
a4 0 50.00
c3 0 50.00

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1

22/30 0:14 0.00 10...Re8 11.c3 a5 12.Bf4 cxd4 13.Nxd4 Bc5 14.Rxe8+ Qxe8 15.Ne3 Ne5 16.Bxe5 Qxe5 17.Qd3 Qc7 18.Rd1 g6 19.Qb5 a4 20.g3 Kg7 21.Kg2 a3 22.b4 (43.604.054) 2960
22/31 0:14 +0.08 10...a5 11.c3 Re8 12.Bf4 cxd4 13.Nxd4 Bc5 14.Rxe8+ Qxe8 15.Ne3 Ne5 16.Bxe5 Qxe5 17.Qd3 Qc7 18.a4 g6 19.Rd1 Re8 20.g3 Re4 21.Kg2 Re8 22.Qb5 (43.604.054) 2960
22/33 0:14 +0.27 10...Rc8 11.c3 Re8 12.dxc5 Nxc5 13.Bg5 Nce4 14.Be3 a6 15.Bd4 b5 16.Qd3 Nc5 17.Bxc5 Bxc5 18.Rxe8+ Qxe8 19.Re1 Qd7 20.h3 b4 21.Ne5 Qe6 (43.604.054) 2960
22/35 0:14 +0.35 10...cxd4 11.Nxd4 Re8 12.a4 Nc5 13.Ng3 Qd7 14.Ndf5 Bf8 15.Rxe8 Rxe8 16.Be3 Nfe4 17.Qg4 g6 18.Nh6+ Bxh6 19.Qxd7 Nxd7 20.Bxh6 Nxg3 21.fxg3 a6 22.Kf2 (43.604.054) 2960

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1 Re8

dxc5 16 51.21
Rb1 12 50.91
a4 7 50.53
c3 5 50.38

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1 Re8 11.dxc5

22/31 0:13 +0.23 11...Nxc5 12.Be3 Ne6 13.c3 Qd7 14.Qd2 a6 15.Ng3 Bc5 16.Nf5 Bxe3 17.Qxe3 Qc7 18.Rad1 Rad8 19.Qd2 g6 20.N5d4 Kg7 21.h3 Ne4 22.Nxe6+ fxe6 (40.603.271) 2925
22/27 0:13 +0.23 11...Rc8 12.c3 Nxc5 13.Be3 Bd6 14.a4 Qd7 15.Bd4 Nfe4 16.Ne3 Bf4 17.g3 Bh6 18.Bxc5 Rxc5 19.Nd4 Bxe3 20.Rxe3 a6 21.Re2 Rcc8 (40.603.271) 2925
22/34 0:13 +0.33 11...a5 12.a4 Nxc5 13.Be3 Qd7 14.c3 Nfe4 15.Ng3 Rad8 16.Nh5 Ne6 17.Qb3 Bc5 18.Ng3 Bxe3 19.Rxe3 N6c5 20.Qb5 Qc7 21.b4 Ne6 22.Qxa5 Qxa5 23.bxa5 Nxg3 24.hxg3 (40.603.271) 2925
22/32 0:13 +0.38 11...Bxc5 12.Rxe8+ Qxe8 13.Be3 Qe4 14.c3 Re8 15.Qb3 h6 16.Bd4 Bxd4 17.cxd4 Qe6 18.Rc1 Qb6 19.Qxb6 Nxb6 20.Rc7 Re2 21.N3d2 Ne4 22.Nxe4 dxe4 23.Rxb7 Rxb2 (40.603.271) 2925

A different type of tie-break happens naturally here, as Nxc5 had score 0.20 and Rc8 had score 0.23 at depth 21, Nxc5 is sorted first.

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1 Re8 11.dxc5 Nxc5

Bf4 24 51.82
Re2 20 51.51
Bg5 18 51.36
c3 17 51.29

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1 Re8 11.dxc5 Nxc5 12.Bf4

22/31 0:18 0.00 12...Qb6 13.Rb1 Ne6 14.Be3 Qa6 15.Ra1 Bd6 16.Ng3 Bxg3 17.hxg3 Qb5 18.Rb1 Qc4 19.Ra1 (56.263.395) 3075
22/34 0:18 +0.22 12...Nce4 13.Be3 Ng4 14.Qd3 Nxe3 15.Nxe3 Bc5 16.Qb3 Qd7 17.Rad1 d4 18.Nd5 Rad8 19.c4 d3 20.Rxe4 Rxe4 21.Qxd3 Qe6 22.Nf6+ gxf6 23.Qxd8+ Kg7 24.Qd5 Qxd5 25.Rxd5 Rxc4 (56.263.395) 3075
22/31 0:18 +0.32 12...Ne6 13.Be3 (56.263.395) 3075
22/29 0:18 +0.43 12...a5 13.Be5 Nce4 14.Bd4 Rc8 15.Ne3 Bc5 16.c3 Qd7 17.Bxc5 Rxc5 18.a3 a4 19.Nd4 Rcc8 20.Qe2 b6 21.Ndf5 Nd6 22.Nxd6 Qxd6 23.Rad1 (56.263.395) 3075

1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6.Bb5 Bd7 7.Bxd7 Nbxd7 8.O-O Be7 9.Re1 O-O 10.Nf1 Re8 11.dxc5 Nxc5 12.Bf4 Qb6

Database is out of moves

So this is our leaf node:

[d]r3r1k1/pp2bppp/1q3n2/2np4/5B2/5N2/PPP2PPP/R2QRNK1 w - -

And the judge...

34/23 1:59 +0.09-- 13.Be5 Nce4 (373.550.041) 3125
34/48 2:13 +0.18++ 13.Be5 (418.023.809) 3126
34/48 2:46 0.00-- 13.Be5 Rad8 (521.927.774) 3128
34/48 3:11 +0.16++ 13.Be5 (599.908.907) 3132
34/48 3:28 +0.20 13.Be5 Nce4 14.Bd4 Bc5 15.Ne3 Rad8 16.h3 a5 17.Rb1 Bxd4 18.Nxd4 Nc5 19.c3 Ne6 20.b4 h6 21.a4 Nxd4 22.Qxd4 Qc7 23.Red1 Ne4 24.Rd3 Nf6 25.Qc5 Qxc5 26.bxc5 (652.333.609) 3128
35/48 3:51 +0.10-- 13.Be5 Nce4 (719.841.229) 3107
35/48 3:53 +0.01-- 13.Be5 Nce4 (723.283.099) 3104
35/48 4:03 +0.10++ 13.Be5 (754.595.608) 3092
35/48 4:25 +0.18 13.Be5 Rad8 14.Ne3 Nce4 15.Bd4 Bc5 16.h3 h6 17.Rb1 Bxd4 18.Nxd4 Nd6 19.Qd3 g6 20.c3 Kg7 21.Re2 Qa5 22.Ra1 h5 23.a3 Qc7 24.Rae1 Nde4 25.Qc2 Nd6 26.Rd1 (823.224.503) 3096
36/45 4:43 +0.24 13.Be5 Rad8 14.Ne3 Nce4 15.Bd4 Bc5 16.h3 h6 17.Rb1 Bxd4 18.Nxd4 Qa6 19.a3 g6 20.c3 Kg7 21.Qe2 Qb6 22.Qc2 Nc5 23.Re2 Nce4 24.Rbe1 h5 25.Qc1 a6 26.Qc2 (877.926.924) 3098

Despite lower score at the leaf node, and a lower winrate, it seems that following the best Score performed better in this one.

Conclusion: After doing this experiment I end up with the feeling that, ranking moves in a chess position, in general, is kind of subjective. If 5.exd5 and 5.e5 are both draws with perfect play, who can conclude that one should be ranked over the other? Who is to say that minmax is better than this scoring, or that playing the move that guarantees the best possible winrate leaf node is better than playing the move with best winrate every time? I wonder if someone comes up with a way to answer these questions in the future, to give scores and rank moves in a very strong way, and then MultiPV=4 at depth 22 ends with some +1.00 score at the leaf node...

[Uri Blass]

I do not understand why multipv=4 and depth 22 is the best.

For multi-pv I think that it is dependent on the position and for example it is not logical to have multipv=4 in positions when there are only 2 options that do not lose a lot of material when it is logical to use more than 4 options in case that more than 4 options get almost the same score.

For depth I think that depth 22 is too small and it is better to have more quality even at the price of analyzing smaller number of positions.

I also do not understand why the same depth for all positions(I think that it is better to analyze earlier positions to higher depth and maybe you will find that the best move at depth 40 is not one of the top 4 moves at depth 22).

[noobpwnftw]

I do not understand why multipv=4 and depth 22 is the best.

For multi-pv I think that it is dependent on the position and for example it is not logical to have multipv=4 in positions when there are only 2 options that do not lose a lot of material when it is logical to use more than 4 options in case that more than 4 options get almost the same score.

For depth I think that depth 22 is too small and it is better to have more quality even at the price of analyzing smaller number of positions.

I also do not understand why the same depth for all positions(I think that it is better to analyze earlier positions to higher depth and maybe you will find that the best move at depth 40 is not one of the top 4 moves at depth 22).

There is never a multiPV=4 parameter, rather:

Code: Select all

multiPV = rootMoves.size();

I agree that depth 22 may not be enough for some positions, but the chances that it will not discover a correct move under the above conditions is very slight.

[Ovyron]

I do not understand why multipv=4 and depth 22 is the best.

MultiPV=4 is best "in the general case", because figuring out what is the best MultiPV to use on a given position takes more time than analyzing more positions (in the long run if after a while you'd have analyzed 110 positions instead of 100, those extra 10 positions will be better than just 100 with the ones benefiting from a more correct MultiPV), but quality is considerably worse with MultiPV=3 and time taken is considerably worse with MultiPV=5, so it's some kind of "sweet spot."

A similar case can be made for Depth 22, for "a large number of positions", as 21 produces too low quality and 23 is too slow. Numbers gets fuzzy, but I'm talking of about 50000 positions analyzed with the best possible quality in the least possible time. As positions go down and you find you don't care much about time, you increase depth.

For "a small number of positions", like when interacting with a chess engine to figure out what to play on a corr chess game (finding a move better than unassisted engine, possibly being helped out by a secondary engine) or when trying to find what are the best moves in a critical opening line, I recommend pumping up the depth to 28, where 27 is low quality and 29 isn't worth your time, as depth 28 is enough to beat some engine slave chunking out unassisted depth 40 moves. Interestingly, back in Stockfish 9 times this depth 28 used to be depth 32, so software is improving and I expect to require less depth in the future (or to be able to increase to MultiPV=5 without it being too slow.)

I also do not understand why the same depth for all positions(I think that it is better to analyze earlier positions to higher depth and maybe you will find that the best move at depth 40 is not one of the top 4 moves at depth 22).

Right, depth 22 would be recommended to give a score to leaf nodes, or to play many games in reasonable time to get statistics and use some formula to give a score according to those like noob's database is doing.

If you actually want to know what moves to play, what alternative moves to consider, or what moves should be extended, then you need higher depth.

I'd be great to find actual game situations where MultiPV=4 Depth 28 misses the best move (does not appear in the top 4) while Depth 40 finds it. The best I've seen is best move being top 8 ranked in MultiPV=8 but it was better by a few centipawns, not like "this was the only move that wins/saves the game and it didn't appear in the top 4 in MultiPV."

[zullil]

Thanks for making the nice web interface, and thanks to noobpwnftw for building the database.

Agreed.

Who is to say that minmax is better than this scoring, ...

Not sure what "better" might mean, but minimax scoring at least has the comforting (to me) property that scores remain constant along any PV determined by the scores.

[noobpwnftw]

Thanks for making the nice web interface, and thanks to noobpwnftw for building the database.

Agreed.

Who is to say that minmax is better than this scoring, ...

Not sure what "better" might mean, but minimax scoring at least has the comforting (to me) property that scores remain constant along any PV determined by the scores.

Provided that the search is also done consistently according to PV(new best move has to reach at least the same depth as the one being refuted for its score to make sense), which unfortunately is not the case due to the scale of my database.

[noobpwnftw]

A full database snapshot which contains about 5.6 billion unique positions as of 2019-10-16 is available at:
ftp://ftp.chessdb.cn/pub/chessdb/data-s ... 191016.tar

[Master Om]

analyzed by Stockfish with no less than 22 plies at terminal node

Interesting, my private database uses depth 22 as well, looks like we found it to be optimal (depth 21 having considerably less quality, depth 23 being consirerably more slow) independently?

I tried to probe from startpos with the following result.

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

Code: Select all

    Move   Score  Rank       Note  winrate%
0   e2e4  15 (8)     2  ! (20-04)     50.61
1   d2d4  15 (4)     2  ! (20-03)     50.30
2   g1f3  15 (2)     2  ! (20-04)     50.15
3   g2g3  10 (2)     2  ! (20-07)     50.15
4   c2c4  10 (2)     2  ! (20-04)     50.15
5   d2d3       0     1  * (20-12)     50.00
6   c2c3       0     1  * (20-08)     50.00
7   e2e3       0     1  * (20-10)     50.00
8   b2b3       0     1  * (20-10)     50.00
9   b1c3       0     1  * (20-04)     50.00
10  a2a3       0     1  * (20-09)     50.00
11  h2h3      -1     1  * (20-09)     49.92
12  f2f4      -4     0  ? (20-14)     49.70
13  a2a4      -5     0  ? (20-11)     49.62
14  b2b4      -6     0  ? (20-11)     49.55
15  g1h3     -41     0  ? (20-05)     46.90
16  b1a3     -51     0  ? (20-01)     46.14
17  h2h4     -57     0  ? (20-01)     45.69
18  f2f3     -82     0  ? (20-01)     43.82
19  g2g4    -103     0  ? (20-01)     42.26

Surprising to see scores that high. Mine has everything at 0.00 except for 1.d4 which is 0.03 (all white tries have been refuted to a 0.00 score otherwise).

...

Oh, three billion means your database is 1000 times larger than mine

I'd wish for a way to check it online (see https://www.365chess.com/opening.php for an example)

Depth 22 seems to be a good balance between quality and speed.

I have applied penalties to a 0.00 score in back-propagation, maybe that caused it.

For a nice GUI like those I someone would look up the data from my API so that no reinventing wheels is needed.

Code: Select all

Depth 22 seems to be a good balance between quality and speed.

How can u confirm this. Am curious to know. AFAIK even numbers have a blind spot .

[noobpwnftw]

How can u confirm this. Am curious to know. AFAIK even numbers have a blind spot .

Roughly, at this depth it can hit "best" move with 70% probability, then among top 5 moves I have a statistically sound chance to be correct.
Then this process is done recursively so generally it wouldn't miss important lines, in practice, better moves can be discovered by self-play using deeper searches.