1600-rated engine draws latest Drawfish

[tpoppins]

[pgn][Event "test"] [Site "Dual X5670"] [Date "2018.07.13"] [Round "1"] [White "Shallow Blue 1.1.0 64-bit"] [Black "Stockfish 070718 64-bit"] [Result "1/2-1/2"] [ECO "A41"] [GameDuration "00:10:35"] [GameEndTime "2018-07-13T04:22:48.139 Eastern Daylight Time"] [GameStartTime "2018-07-13T04:12:12.630 Eastern Daylight Time"] [Opening "Old Indian"] [PlyCount "31"] [TimeControl "40/1260"] [Variation "Tartakower (Wade) Variation"] 1. d4 {book} d6 {book} 2. Nf3 {book} Bg4 {book} 3. c4 {book} e5 {book} 4. dxe5 {book} Bxf3 {book} 5. exf3 {-0.11/8 32s} dxe5 {-0.45/29 67s} 6. Qb3 {+0.28/9 30s} Nc6 {0.00/24 8.2s} 7. Qxb7 {+0.83/8 32s} Nd4 {0.00/29 23s} 8. Bd3 {+1.19/8 32s} Nf6 {0.00/30 18s} 9. O-O {+1.29/7 32s} Be7 {0.00/31 47s} 10. Nc3 {+1.39/8 29s} Rb8 {0.00/33 13s} 11. Qxa7 {+1.70/8 32s} Ra8 {0.00/37 16s} 12. Qb7 {+0.09/8 33s} Rb8 {0.00/41 18s} 13. Qa6 {0.00/8 33s} Ra8 {0.00/42 32s} 14. Qb7 {0.00/9 25s} Rb8 {0.00/44 18s} 15. Qa6 {0.00/9 25s} Ra8 {0.00/45 15s} 16. Qb7 {0.00/9 25s, Draw by 3-fold repetition} 1/2-1/2 [/pgn]
Full game log on Pastebin: link -- includes all SF UCI options

Shallow Blue is a UCI engine by Rhys Rustad-Elliott, rated 1565 Elo on the CCRL blitz (cf SF9 at 3491 Elo). It's 1726 on the 40/40 list but that's based on a mere 50 games. A barebones single-threaded engine with no configurable options, using 24 MB of RAM. The game was run under CCRL 40/40 rules: default settings and for SF 512 MB hash, 5-men Syzygy TBs and the default contempt 21. Actual time control: 40/21 (this is what the Xeon X5670 benchmarks at).

[d]rn1qkbnr/ppp2ppp/3p4/4P3/2P5/5b2/PP2PPPP/RNBQKB1R w KQkq - 0 5
This is the book exit. SF evaluates it as +0.63 at d=44, but after 5.exf3 the eval goes down to +0.35 (same depth). So we can rule out this line as being bad for Black.

Actually, Shallow Blue is nothing special -- it's just the lowest-rated engine I had installed at the moment that is present on both of our lists. The result should be reproducible with lots of other sub-2000 engines; for example I tried Gargamella -- a 16-year old Xboard engine rated 1680 @40/4 that runs in 3MB of RAM -- and it made a still shorter draw (in 15 moves) at 40/4.

Yes, the default contempt and the emphasis on testing at hyper-bullet TCs do amazing wonders for Stockfish. In human terms, this would be like Magnus Carlsen walking into your local club, sitting down to play a 1000-rated guy and only managing a draw.

[Eelco de Groot]

I just don't think it is easy to avoid White drawing the opening, even if Stockfish knew the opponent is weak, and played another move somewhere, the chances of a draw may still be higher than White making a fatal mistake. Stockfish does not know the rating of his opponent but even if he knew he can't take too many risks. Just the first move already, other moves don't promise much in eval, would be somewhat risky.


[d]rn1qkbnr/ppp2ppp/3p4/4P3/2P5/5P2/PP3PPP/RNBQKB1R b KQkq -

Engine: Ancalagon 20171105 (512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G.

36 25:50 +0.48 5...dxe5 6.Qxd8+ Kxd8 7.Nc3 Nd7
8.Be3 c6 9.O-O-O Kc7 10.h4 b6 11.Ne4 Be7
12.h5 Ngf6 13.h6 g6 14.Ng5 Rhf8
15.Bd3 Rae8 16.Rhe1 Bc5 17.Bc2 a5
18.Bxc5 Nxc5 (2.961.617.276) 1910

36 25:50 +0.77 5...Nc6 6.exd6 Bxd6 7.Nc3 Qe7+ 8.Be3 O-O-O
9.Qc2 Nd4 10.Qa4 Nf5 11.Qxa7 Nxe3
12.Qxe3 Bc5 13.Qe2 Qxe2+ 14.Bxe2 Nf6
15.Rd1 Rxd1+ 16.Nxd1 Rd8 17.a3 Bd4
18.Bd3 g6 (2.961.617.276) 1910

36 25:50 +0.97 5...Qe7 6.Be2 Nc6 7.exd6 Qxd6 8.O-O O-O-O
9.Nc3 Qe6 10.Qa4 Bc5 11.Bf4 Nf6
12.Rfd1 Rd4 13.Rxd4 Bxd4 14.Bf1 Qf5
15.Bg3 Nh5 16.Nb5 Nxg3 17.Nxd4 Nxd4
18.hxg3 a6 (2.961.617.276) 1910

A draw is not so bad, as long as you get to play enough games.

[Uri Blass]

I just don't think it is easy to avoid White drawing the opening, even if Stockfish knew the opponent is weak, and played another move somewhere, the chances of a draw may still be higher than White making a fatal mistake. Stockfish does not know the rating of his opponent but even if he knew he can't take too many risks. Just the first move already, other moves don't promise much in eval, would be somewhat risky.


[d]rn1qkbnr/ppp2ppp/3p4/4P3/2P5/5P2/PP3PPP/RNBQKB1R b KQkq -

Engine: Ancalagon 20171105 (512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G.

36 25:50 +0.48 5...dxe5 6.Qxd8+ Kxd8 7.Nc3 Nd7
8.Be3 c6 9.O-O-O Kc7 10.h4 b6 11.Ne4 Be7
12.h5 Ngf6 13.h6 g6 14.Ng5 Rhf8
15.Bd3 Rae8 16.Rhe1 Bc5 17.Bc2 a5
18.Bxc5 Nxc5 (2.961.617.276) 1910

36 25:50 +0.77 5...Nc6 6.exd6 Bxd6 7.Nc3 Qe7+ 8.Be3 O-O-O
9.Qc2 Nd4 10.Qa4 Nf5 11.Qxa7 Nxe3
12.Qxe3 Bc5 13.Qe2 Qxe2+ 14.Bxe2 Nf6
15.Rd1 Rxd1+ 16.Nxd1 Rd8 17.a3 Bd4
18.Bd3 g6 (2.961.617.276) 1910

36 25:50 +0.97 5...Qe7 6.Be2 Nc6 7.exd6 Qxd6 8.O-O O-O-O
9.Nc3 Qe6 10.Qa4 Bc5 11.Bf4 Nf6
12.Rfd1 Rd4 13.Rxd4 Bxd4 14.Bf1 Qf5
15.Bg3 Nh5 16.Nb5 Nxg3 17.Nxd4 Nxd4
18.hxg3 a6 (2.961.617.276) 1910

A draw is not so bad, as long as you get to play enough games.

Stockfish with Contempt=100 can avoid 9...Be7 and choose 9...Rb8 and I guess that it will practically win against weak opponents.