83 nice Testpositions (Long and big post).

[Dann Corbit]

I think that all of these are in doubt:

Code: Select all

r1b1r1k1/p2q1p1p/np2nBp1/3p1N1Q/8/6PB/P4P1P/1R2R1K1 w - - acd 50; acs 3598; bm Nh6+; c3 "Qh6"; ce 1380; pm Nh6+; pv Nh6+; 
r2qk2r/pp3ppp/2p1pn2/4n3/1b6/3P2PP/PPPN1PB1/R1BQK2R b KQkq - acd 50; acs 3599; bm Qc7; c0 "There is no key move. It seems likely that a5 Ned7 O-O Rc8 Be7 Qc7 all draw."; c3 "h5"; ce -54; pm Qc7; pv Qc7 Qe2; 
4r2k/Q5pp/3pN3/1P1P1n2/P4P2/2r4q/6RP/3R2K1 w - - acd 52; acs 17565; am Qd7; bm Qf7; c0 "Qf7 wins, Re1 looks drawish"; c3 "Re1"; ce 684; id "Vincent Lejeune - hard CCC 2008 - Pos 64"; pm Qf7; pv Qf7 Rec8 Ng5 R3c7 Nxh3 Rxf7 Rb1 Ne3 Ra2 Nxd5 b6 Rb7 a5 h6 Rd2 Nxb6 Rxb6 Rbc7 Kg2 Kh7 Rbxd6 Kg8 a6 Kf7 f5 Rc4 Rd7+ Kf6 a7 Ra4 Nf2 Kxf5 Rxg7 h5 Re2 Rd8 Rge7 Rg8+ Kf3 Rc8 R2e5+ Kg6 Nd3 Rf8+ Ke3 Ra3 Rd5 h4 Kd4 Rf5 Nc5 Rf8 Nd7 Ra8 Ne5+ Kf5 Nc4+ Kf6 Nxa3 Kxe7 Nb5 Kf6 Kc5 Ke7 Kb6; 
r3r3/p2n1pk1/2qpp2p/1p5n/3PN2p/1PPQ1P2/P1BB1P2/R6K w - - acd 53; acs 3599; bm c4; c0 "There is no best move, since Bxh6+ c4 Rg1+ and Qe3 all win easily and a4 Qe2 also probably win."; c3 "Bxh6+"; ce 776; pm c4; pv c4; 
r1b1r1k1/p3qp1p/1ppp1npb/4n3/1PPBP3/2N2N1P/P1Q1BPP1/3RR1K1 w - - acd 54; acs 3599; bm a4; c0 "It is not at all clear which is best out of:a4 Na4 Qb3 Nb1 Qb2"; c3 "c5"; ce 46; pm a4; pv a4; 
r1b2rn1/1ppn3k/p4q1p/3P1p2/1p3Q2/1NNB4/P5PP/4RRK1 w - - acd 54; acs 3599; bm Bxf5+; c0 "Bxf5+ is clearly best and Ne2 Qxb4 Ne4 Na4 Re6 may also win."; c3 "Re6"; ce 362; id "GS2930.08"; pm Bxf5+; pv Bxf5+; 
r1b1r1k1/1p1nnpp1/pq1Bp2p/8/4N3/6Q1/2PRB1PP/5R1K w - - acd 56; acs 3594; bm Be5; c0 "Be5 wins and is best, Bc7 also wins, and Bh5 may win"; c3 "Bc7"; cce 407; ce 608; pm Be5; pv Be5 Ng6 Bxg7 Kxg7 Qc3+ e5 Rd6 Qd8 Qg3 Qe7 Bc4 f6 Bd3 Rg8 Nxf6 Nf8 Rb6 Be6 Nxg8 Kxg8 Bxg6 Qg5 Qxg5 hxg5 Be4 Rd8 Kg1 Rd4 Bxb7 Bc4 c3 Rd2 Re1 Kf7 Rb4 Be6 Rxe5 Kf6 Re3 Rd6 Ra4 Rd7 Be4 Rd6 Ra5 g4 Bd3 Rc6 Be2 Bc8 Ra4 Ng6 Bxg4 Bxg4 Rxg4 Ne5 Rf4+ Kg5 Rf1 Ng6 Rg3+ Kh6 Rf5 Ne7 Ra5 Ng6 Rg4 Nf8 c4 Nd7 Kf2 Nc5 Ke3 Re6+ Kd4 Nb3+ Kd5 Re1 Rxa6+ Kh5; 
r1bqk2r/ppp1ppbp/2n3p1/3p4/2PP1B2/2P1P3/P4PPP/R2QKBNR w KQkq - acd 58; acs 3599; am cxd5; bm cxd5; c0 "I am not convinced that Nf3 is better than cxd5, which have about the same score"; c3 "Nf3"; ce 76; pm cxd5; pv cxd5 Qxd5 Bxc7 O-O Nf3 Bg4 Be2 Rac8 Bg3 Na5 Rc1 Qxa2 Ra1 Qd5 Ra3 e5 h3 Bxf3 Bxf3 Qd8 O-O b6 Be2 Qe7 Qa1 h5 Ba6 Rc7 Rd1 Rd8 Bb5 Rdc8 Bd3 Rd8 Ra4 Rc6 Bb5 Rcc8 Rb1 Qe6 Qb2 Bf6 Ba6 Rc6 Qa1 Bg7 Rf1 Rc7 Bb5 Kh7 Rb1 Kg8 Be2 Rdc8 Ba6 Rd8 c4 exd4 Bxc7 dxe3 Bxd8 Bxa1 Rbxa1 exf2+ Kxf2 Nb3 Re1; 
rnb1k2Q/1p5p/p7/4p3/4q3/8/PPP2R1P/2K5 b - - acd 60; acs 3599; am Kd7; bm Kd7; c0 "It is possible that both Kd7 and Ke7 draw, or that neither draw.  This problem seems quite unclear."; c3 "Ke7"; cce -0; ce -805; pm Kd7; pv Kd7 Rd2+; 
r1b1q1k1/1p2rppn/p3p2p/3pP2Q/3P4/K1PBN2P/PP4P1/R4R2 w - - acd 61; acs 3599; bm Rf6; c3 "Bxh7+ Rf6"; ce 227; pm Rf6; pv Rf6 Nxf6 exf6 Rc7 Rf1 Qf8+ Kb3 e5 Qxe5 Rc6 fxg7 Qxg7 Qf4 Kh8 g4 Be6 h4 f6 a4 Re8 Ka2 Rg8 a5 Re8 Bc2 Rd8 Kb1 b5 Ka2 Rf8 Re1 Rd8 Bd3 Bc8 Re2 Kg8 Rg2 Qf8 Nc2 Re6 Kb1 Kh8 Rg1 Red6 Rf1 Qg7 Ne3 Rc6 Rf2 Be6 Ka2 Re8 Rf1 Bd7 Rf3 Be6 Bc2 Bd7 Rf2 Be6 Re2 Rd8 Re1 Kg8 Rf1 Bc8 Kb1 Be6; 
2kr1nr1/pbp1qp2/1p2pn1p/8/3Pp2Q/P1P3N1/2PBBPPP/R3K2R w KQ - acd 67; acs 3599; am O-O; bm Nh5; c0 "Nh5 definitely draws, and several other moves may also draw"; c3 "Bxh6"; ce 0; id "ST3_07 - Koenigssicherheit"; pm Nh5; pv Nh5 Rxg2 Qxf6 Qxf6 Nxf6 Rg6 Ng4 e3 Rg1 exd2+ Kxd2 Nd7 Bd3 f5 Ne3 Rdg8 Rg3 Kd8 Nc4 h5 Rag1 Rxg3 Rxg3 Rg4 Be2 Rh4 h3 Ke8 Rg6 Rxh3 Rxe6+ Kf7 Rh6 Kg7 Re6; 
8/3bkp2/n7/1rN1P3/2R3P1/2R4K/2p5/8 b - - acd 77; acs 3599; am Nb4; bm Nxc5 Rxc5; c3 "Rxc5"; ce 0; pm Nxc5 Rxc5; pv Nxc5 Rxc5 Rb4 R5c4 Rb5 Rxc2 Rxe5 Rc7 Rd5 R2c5 Rd1 Rg5 Kf8 Ra7 Rh1+ Kg2 Rd1 Kf3 Rd4 Ra6 Be6 Ra8+ Ke7 Ra7+ Kf8 Kg3 Rd3+ Kh4 Rd1 Ra8+ Ke7 Rgg8 Rc1 Ra7+ Kf6 Ra5 Rc3 Rh5 Rc1 Rb5 Rh1+ Kg3 Rg1+ Kf4 Rf1+ Ke3 Re1+ Kf2 Re5 Rbb8 Rc5 Kg3 Rc4 Rb6 Rc3+ Kh4 Rc1 Rb4 Rh1+ Kg3 Rg1+ Kh2 Rf1 Rb7 Ra1 Kg3 Rg1+ Kf4 Rf1+ Ke3 Re1+ Kf2 Rc1 Rb4 Ke5 Rg5+ Kd6 Rf4 Ke7 Re5 Rb1 Re3 Rd1 Rf5 Rd2+ Kf3 Rd6 Kg3 Ra6 Kh4; 

I ran multi-pv analysis on 13 of the problems that seemed a bit sticky. Attached you will find the log file and resulting EPD records.

[jp]

But we found 2 solutions for the starting position , both draw, so it's not a good test.

Black-Diamond-XI find this quickly

Is the feeling that the fortress detection code is totally reliable? (i.e. no false positives e.g. a "fortress" that can be broken. False negatives are okay.)

[jhellis3]

Is the feeling that the fortress detection code is totally reliable?

I will try to answer, but I'm not sure what "totally reliable" actually means.... To put it most simply, it will be as reliable as the robustness of the underlying search. So yes, you can get false positives. But this is not due to anything the fortress code is doing (it just accelerates the process). Examples of where this might happen is where breaking the fortress requires a very large sacrifice (queen) or multiple significant sacrifices.

But it is important to put things into context:
1) If the default search explores such moves it is a non-issue.
2) If the default search does not explore such moves, there may be a problem. But this problem is one that already existed (not created by fortress detection) and if anything the FD code may slightly help depending on to what extent stats based pruning is performed in the search.

My suggestion for best practice when studying possible fortress positions is to use something with FD code with a high contempt for the side trying to break the fortress. The contempt will help ensure sacrificial lines are explored, while the FD code will help provide optimal resistance for the defending side as well as more accurate scoring.

Also, while the code does effectively detect fortresses, I feel that language sort of masks what is really going on... Quite simply, it is a recursive shuffle extension which continues on any shuffle (repeating) move/position until one of three things happens:
1) 3 fold repetition
2) 50 move rule
3) A non-shuffling move is selected

As shuffling moves approach the 50 move rule they are linearly reduced in value towards draw in order to more accurately reflect the likely true value of the position while also encouraging the selection of non-repeating alternative moves (should they exist).

[jp]

I will try to answer...

Quite simply, it is a recursive shuffle extension which continues on any shuffle (repeating) move/position until one of three things happens:
1) 3 fold repetition
2) 50 move rule
3) A non-shuffling move is selected

As shuffling moves approach the 50 move rule they are linearly reduced in value towards draw in order to more accurately reflect the likely true value of the position while also encouraging the selection of non-repeating alternative moves (should they exist).

Thanks.