BeyondCritics wrote:In this position Nb8-a6 seems bad on general grounds and Nd7 and Ne8-a6 are nearly equal positionally. Other options seem clearly worse. In any case black faces a difficult defense: No space, limited center control and no counterplay in sight.
Btw i personally don't think it is a good test position. If black plays Nd7 then white is better. If black plays Ne8-c7 white is better too, so what?
Give this test position to different engines. One engine will choose Nd7 and another engine will choose Ne8-c7. What did you learn? Not much.
In a good position the engines faces difficult, pivotal choices, which are uncorrelated. Something like a pawn sacrifice or so. You will then extract maximum information.
Just my 2 cent.
I agree that Na6 isn't quite as good as Nd7 and Nc7, but it isn't easily to be shown, and as for test positions I don't think we should look at the ones with clear single best moves only anymore.
Of course you get more information about the engine's search for these bms or ams by these classical test positions, but if you want to test the engines's evals for moves not as clearly to be discriminated you need to judge these evals compared to each other in "quiet" "positional" positions too.
Here is the latest SF dev. with empty hash on 4MV mode, depth 34:
rn1qnrk1/pp2ppbp/3p2p1/P1pP4/4P3/2N2B2/1PP2PPP/R1BQ1RK1 b - - 0 1
Analysis by Stockfish 030115 64 POPCNT:
1. +/= (0.65): 10...Nc7 11.Be2 Nba6 12.Bc4 Nb8 13.Bg5 Nd7 14.Be2 Rb8 15.Qd2 Qe8 16.Rfe1 Qd8 17.h3 Re8 18.Be3 Rf8 19.Ra3 Be5 20.Ra2 Nf6 21.Bh6 Re8 22.Bc4 a6 23.Na4 Nd7 24.Be3 Nb5 25.c3 Bg7 26.Raa1 Qc7 27.f4
2. +/= (0.66): 10...Nd7 11.Be2 Nc7 12.Bg5 Rb8 13.Qd2 Qe8 14.Rfe1 Qd8 15.h3 Bd4 16.Reb1 Nf6 17.Bd3 Be5 18.Re1 Nd7 19.f4 Bd4+ 20.Kh1 Re8 21.Bh4 Qc8 22.Rab1 b5 23.axb6 Rxb6 24.b3 Rb4 25.Rbd1 Bf6 26.Bg3 e5 27.Ra1 exf4
3. +/= (0.67): 10...Na6 11.Bg5 Nec7 12.Re1 Nb8 13.Be2 Nd7 14.Qd2 Bd4 15.Rf1 Re8 16.Kh1 a6 17.f3 Rb8 18.Rfb1 Bf6 19.Be3 Be5 20.Rf1 Bf6 21.Rfd1 b5 22.axb6 Rxb6 23.Ra2 Bg7 24.Kg1 Qc8 25.Na4 Rb8 26.Rda1 Bd4 27.Nc3 Bxe3+ 28.Qxe3 Qb7 29.b3
4. +/- (0.72): 10...a6 11.Be2
And here is a variant SF with Black played against K8 on 240'+30", 12 threads and 8G hash each, ponder on after
1. d4 c5 2. d5 Nf6 3. Nc3 d6 4. e4 g6 5. Nf3 Bg7 6. Be2 Bg4 7. O-O Bxf3 8. Bxf3 O-O 9. a4 Ne8 10. a5 Nd7 11. Be2 Nc7 12. Be3
12...b6
which was the first SF-deviation at backward solving from Leiva Rodriguez-Lozano, Bogota 2013, quoted in .pgn of my first posting.
1. d4 c5 2. d5 Nf6 3. Nc3 d6 4. e4 g6 5. Nf3 Bg7 6. Be2 Bg4 7. O-O Bxf3 8. Bxf3 O-O 9. a4 Ne8 10. a5 Nd7 11. Be2 Nc7 12. Be3 b6 13. axb6 Nxb6 14. Qd3 Qd7 15. h4 f5 16. Bc1 a5 17. h5 gxh5 18. Bxh5 fxe4 19. Nxe4 Qf5 20. Bf3 Ncxd5 21. Qd1 Be5 22. Ra3 Kh8 23. Ng5 Bf4 24. Bxf4 Nxf4 25. Bxa8 Qxg5 26. Rg3 Qf6 27. Bf3 a4 28. Qa1 Nc4 29. b3 axb3 30. Qxf6+ Rxf6
Compared to the variant in my second posting of K8 playing with Black after 10...Na6 against SF,
1. d4 c5 2. d5 Nf6 3. Nc3 d6 4. e4 g6 5. Nf3 Bg7 6. Be2 Bg4 7. O-O Bxf3 8. Bxf3 O-O 9. a4 Ne8 10. a5 Na6
11. Be2 Nec7 12. Bg5 Qd7 13. f4 b5 14. axb6 axb6 15. Bg4 Qd8 16. f5 b5 17. Qe1 Ne8 18. fxg6 hxg6 19. Qh4 Nf6 20. e5 dxe5 21. Rxf6 exf6 22. Ne4 fxg5 23. Nxg5 Qxg5 24. Qxg5 e4 25. h4 Bxb2 26. Rd1 Bd4+ 27. Rxd4 cxd4 28. h5 Kh7 29. hxg6+ fxg6 30. Qe7+ Kh6 31. Qxe4 Nc5 32. Qxd4 Nb7 33. Qb4 Rad8 34. Qxb5 Nd6 35. Qd3
there is much less dynamics in the first one variant than in the second one, and to relativate the engines' evals over the moves of the one and the other one variant at forward and at backward solving yet shows some more difference than there is in comparing the ones of MV at the starting position.
It's a good test position as for my pov to judge evals together with search (Can you really discriminate those two of them at all at positional testing?
If you want to, you'll have to have quite certain kinds of test positions for sure.) in an opening- position of some dynamics but several less forced lines with eval- progressings of small differences
Peter.