Dann Corbit wrote: ↑Sun Jan 19, 2025 5:46 pm
Another category that may be of similar interest is a single winning move and all other moves only draw or worse.
It is clearly easier to find positions of this type andf a lot of studies are of this type.
[d]1rkb4/pNp5/8/2N4p/8/5B2/4K3/8 w - - 0 1 bm Bc6
2n5/3N1p2/p2pPP1n/p1p4P/PkP5/1P1p2N1/KP1p1rbp/3B4 w - - bm e7;
2N5/p4P2/p4RqP/P2krb2/5RNK/2PPB1P1/3P1pP1/4nQ2 b - - bm Nf3+;
2N5/p7/q1p2p2/kp6/3P4/PP3K2/5P2/2B5 w - - bm Bd2+;
2r1n1nK/q1pp2pp/p6r/3BPN1p/8/3Np1P1/1pR1P2B/5k2 w - - bm Nh4;
2R2Nn1/brp1k1K1/3pPp1p/4p1p1/2Nr4/3n4/2p2p2/3B4 w - - bm Ne3;
2R5/3pqNp1/4pp1p/1pp4k/5P1P/1B5K/8/8 w - - bm Re8;
2R5/kp6/2p1BK2/p1P5/8/P1p2bB1/2q5/8 w - - bm Bb8+;
3b1n1n/1p3P1P/1p1K1P1P/3PN3/p7/Pp1p4/1P1P4/1RBk4 b - - bm Ke2;
3K3k/b1ppP1pB/r1r1p3/6pb/1n4N1/qpp2p2/QRR5/8 w - - bm Bg6;
3K4/2P5/3p4/1p1kp3/1P1pp3/1P2Pp2/p3P2b/1N6 w - - bm Nc3+;
3n1k2/5B2/R7/R2PK3/1NP2P1n/3PP1P1/pppppppp/8 w - - bm Kf6;
3N2bK/6rp/R3p1B1/N1k1P2n/2Pn1pp1/2Bp2rq/4pP1b/8 w - - bm Ndc6;
3n2K1/1R6/2pNp1N1/2pkP1p1/4p1qP/bpp1Bp2/2P4n/8 w - - bm Nc8;
3n2K1/5pbB/7k/7p/4N1pp/B2p4/4bp2/7r w - - bm Bb4;
3N3K/2Bnp3/8/3k4/1P5R/3p1P1p/1p1PP1p1/6Nq w - - bm e3;
3nBBK1/3PRPP1/4P1n1/4N1kr/3PPp2/5P2/8/6b1 b - - bm Nxe5;
3q1q2/1PNkPqPR/1Q1q1b1K/1R1q1q1Q/1Q1n1n1P/1NPqRqBB/2q3b1/1r5r w - - bm Q4xd6+;
3r3r/1bk2p2/1p2p3/NpP1qPp1/PP4p1/1N1BP3/4Q1P1/2R2RK1 b - - bm g3;
3R4/4K1b1/8/N1n1k3/n1R1pp2/2P3P1/2N1p1P1/q7 w - - bm Rd5+;
4K2k/6pp/2p1N3/R6R/1p4n1/1B1p1qpp/7p/8 w - - bm Ra8;
4Kbr1/4p1p1/4P1Pp/7P/3np3/p1p1k1P1/p1p3P1/N1N1B3 w - - bm Kd7;
4N3/1p5p/1P3p1k/qp2pK2/1b4P1/4pPP1/2R3P1/8 w - - bm Rc1;
4N3/5b2/2qrr1n1/2bp2Rp/Rn2k3/1P1Np1P1/1K2P3/5B2 w - - bm Bg2+;
4R3/1pKB1ppr/8/2pk3p/2p1p3/b1P3N1/r4Pp1/3bn3 w - - bm Bc8;
4R3/1pKB1ppr/8/2pk3p/2p1p3/b1P3N1/r4Pp1/n2b4 w - - bm Bc8;
4R3/n2BP2p/3B2P1/2P2R2/1P2N1P1/3P3n/brpK1p1r/bkN1Q3 b - - bm f1=N+;
5b1r/1K1p4/1B1P1p2/pkN5/p5R1/p1pPp1N1/2br4/8 w - - bm Ne2;
5B2/1p1n4/6p1/1P2R3/3p3q/8/p1K1P1Q1/k3b3 w - - bm Re3;
5B2/rbp2K2/q2PR3/rp2p1k1/Bn4P1/n1P2p1p/5P1N/b1N5 w - - bm Bh6+;
5bbq/ppp1prnr/4Ppkn/5p2/5PP1/6N1/8/1K5Q w - - bm Qh5+;
5k2/2p4p/7q/4BKNB/7p/p3p2P/P1P1P3/8 w - - bm Bg7+;
5n1B/n1p5/P1P2p1N/1p4Nb/1K1P2PR/3rp2Q/2k2PBR/2b5 b - - bm Ba3+;
5n2/p3p2p/P5P1/1K1PP1P1/1p4p1/1kp1Q1P1/prr5/qb1B4 w - - bm Qb6;
5n2/p3p2p/P5P1/1K1PP1P1/1pp3p1/1k4P1/prr5/qbQB4 w - - bm Qe3+;
5Q2/3qq3/2kqNq2/1qqNRNn1/1qNRPKPn/2qQBbrP/3qQrP1/4bB2 w - - bm Qa8+;
5r2/1bb1Npkp/r3n1p1/3N3P/3n4/q3QP2/p3BKP1/1R5R w - - bm Qh6+;
6B1/n1pp4/6PR/1N2pP1R/PNKb2PB/2Pn2PP/P1b4k/1r6 b - - bm Rxb4+;
I expect to roughly double the number presented so far. So roughly 120 in total.
With more effort, I could expand that quite a bit, but I expect it will take a while to work the bugs out of the seul set.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
First positions seem to be not good test positions because it is too easy to find the right move or there is nothing good and everything is losing.
2n5/3N1p2/p2pPP1n/p1p4P/PkP5/1P1p2N1/KP1p1rbp/3B4 w - - acd 999; bm e7
chest see mate in 9 in few seconds
Other alternatives are losing except maybe Nb8 Bb7 e7 that is probably also losing against Re2 but I did not find a mate score against white.
[d]1Nn5/1b2Pp2/p2p1P1n/p1p4P/PkP5/1P1p2N1/KP1pr2p/3B4 w - - 1 3
2N5/p4P2/p4RqP/P2krb2/5RNK/2PPB1P1/3P1pP1/4nQ2 b - - acd 93; bm Nf3+
stockfish can see mate in 46 against itself in 1 second and it is verified by chest in defender mode.
It is position when everything is losing so not a good test position.
2N5/p7/q1p2p2/kp6/3P4/PP3K2/5P2/2B5 w - - acd 61; bm Bd2+
Not a good test position(too easy to find Bd2+ not with a winning score)
The main problem is to find 4.Ba5+ after 1.Bd2+ b4 2.Bxb4+ Kb5 3.Nd6+ Kb6
[d]8/p7/qkpN1p2/8/1B1P4/PP3K2/5P2/8 w - - 3 4 bm Ba5+
stockfish not with multi-pv does not seem to see it in a reasonable time and it seems it does too much pruning(maybe better to avoid null move pruning when the score is bad)
There are also positions where the best move is only a draw and all other moves are losing.
Some of them are pretty easy, but some chess engines will get the wrong answer. I know this because I have a database table called alternate evals and it contains the positions and evaluations of every attempt to analyze the position along with the move and score.
So I looked in this table for positions that do not have any alternative moves with scores bigger than zero.
That is on the one hand. On the other hand, some of the analysis will have been performed years ago, and today's engines will all find a good move in a fraction of a second on decent hardware.
So, this is how I have derived the positions. And the reason I did not deliver them all at once is that I am analyzing each position with the bleeding edge stockfish at one minute per position with thirty threads and mpv=3. That way I know it takes SF at least one second to find the solution and SF does not find any alternatives. Of course, given a longer time some of them will be refuted, unless all of the positions are driven to + and - checkmate values, we can never know for sure. There are a few like that.
The class of positions which is either a win for SF for a single move and otherwise a draw or worse is much larger, but perhaps less interesting especially because it is not an answer to your question. I am giving only those positions with a win for a single move and a losing score for all other moves.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
6K1/8/1p6/p3Q2N/3p4/3r4/n5p1/3n1brk w - - bm Kf8;
6K1/n2ppBP1/R1p4k/2b3p1/6p1/5pP1/3N3p/2r5 w - - bm Ra4;
7K/2Np2p1/2nB2B1/r3P3/1Ppkp1R1/7P/2bPPbpP/R1r3q1 w - - bm Rxa5;
8/1pk2pNP/2p4b/1PPpnp1B/1K2p2R/Q2b1r2/2qNrB2/1R3n2 w - - bm Qa5+;
8/2N5/1Nbk1n2/7p/6Pb/2K1Qp2/3P3P/6nq w - - bm Qe5+;
8/3P3k/n2K3p/2p3n1/1b4N1/2p1p1P1/8/3B4 w - - bm Nf6+;
8/3P4/n2K2kp/2p3nN/1b6/2p1p1P1/8/3B4 w - - bm Bc2+;
8/3qq3/2kqNq2/QqqNRNn1/1qNRPKPn/2qQBbrP/3qQrP1/4bB2 w - - bm Qa8+;
8/4Kr2/1p6/p3Q2N/3p4/8/n5p1/3n1brk w - - bm Kd8;
8/4p2r/1B2RPb1/2pp1p1p/P2k1K2/1P1Pp1p1/1Pp1P1N1/6br w - - bm Bc7;
8/4p3/7p/4K3/5N2/2p1ppp1/2pbrkp1/2nbrqn1 w - - bm Ke6;
8/5pNK/5B1P/5R2/1p4k1/rp1ppppb/4rb2/n2n1q2 w - - bm Rd5;
8/6Bk/5K2/2P2P2/4PPBn/6RP/1ppppppp/8 w - - bm Ra3;
8/6N1/4p3/7p/1N3P1p/p1pK2Pp/P1R3pb/4k1r1 w - - bm Nf5;
8/7p/4K2p/3N3P/pp1p3P/2PP1Q1P/3Pppp1/4rkr1 w - - bm Nf4;
8/7p/6pP/6Pp/2N5/5P1P/5KPR/1k1q4 w - - bm Rh1;
8/8/1B6/4N1p1/4p3/4p3/N1pQp1PK/2bb1kr1 w - - bm Qe1+;
8/8/1K5p/p7/2N5/1p6/ppp3P1/kb1R2B1 w - - bm Rc1;
8/8/3KNN2/p3p3/rnp1p3/1pk5/bp1n4/qrb1N3 w - - bm Kd7;
8/8/8/2p1Q3/8/6Kp/1ppp4/krqb2br w - - bm Qe8;
8/8/8/p1K4p/k1ppppqp/rp5B/1P6/6R1 w - - bm Bxg4;
8/b6r/q3R3/p3PP2/3NpnP1/2KBP2p/k2Np1Pp/r3Rbn1 w - - bm Rxa1+;
8/p2Kp1p1/rp6/1p6/brpn4/Rnk5/4p1p1/2B1N1N1 w - - bm Kc8;
I hope to finish up tomorrow morning, when my batch of 1024 analysis trials will be complete. It would have finished long ago, but my machine did a reboot, probably due to automatic updates.
Even if these positions are not interesting to you, I have become intensely interested in them. After all, consider your initial and reasonable question as to whether any such positions even exist. And it is a very reasonable question. Today's chess engines are highly tuned to examine favorable board features and today's static evaluation is like the evaluation of a human GM. And considering that one move is fantastic and all the other moves are terrible yet the engine has to do a lot of searching to find them puts these positions into a special class for me.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Dann Corbit wrote: ↑Sun Jan 19, 2025 7:10 am
In other words should we drive the best move clear to winning checkmate and all the other moves clear to losing by checkmate or are approximations good enough?
I think approximations are also very interesting, especially for humans interested in puzzles. So I am keeping copies of the sets that you have posted. Thank you. Are some of the positions easier for humans (to at least guess the best move) than for engines?
8/pKb2prr/3k1P2/2p2B1p/4Pp2/2PN4/p7/4B1n1 w - - bm Nb2;
8/r6b/3R3q/2PP3p/1PnpN3/p2PBK2/pP1pN2k/1nbR3r w - - bm Rxh1+;
b1B4K/1r1N2R1/kB4pp/3p1p2/P2pp3/1P3P1n/3P3q/b3n1r1 w - - bm b4;
B1n5/5p1n/N1p2P1P/bN4p1/RP2P1K1/Q2pr3/RBP2k2/5b2 b - - bm Bh3+;
b1r1r2k/b1qn1p1p/5PpB/3pn3/pp1N4/R1P4Q/4NRPP/1B4K1 w - - bm Bg7+;
B7/8/1p2p1N1/1n2Pp2/1P3Pp1/6p1/2ppPp1P/R1nkbK2 w - - bm Nh4;
b7/r1RNB2K/1p2p3/q2pNkpp/1p3p1P/3P1R2/3Pp3/r7 w - - bm Nf7;
Bk1n3q/p7/P1P5/1p6/1P2p3/4P3/1n6/K4Q2 w - - bm Qf4+;
BKn1q3/BRPk1P1R/3p3N/1p2n1p1/2p2r2/3b2p1/N2bp2P/4r3 w - - bm f8=N+;
bn2B3/rn2Bp1p/p6p/k2N1p2/8/2bp1p2/5P1p/3K3N w - - bm Kc1;
bn2B3/rn2Bp1p/p7/k2N1p2/7p/3p1p2/1K3P1p/4b2N w - - bm Kc1;
k1K5/1N1p1p2/1Ppp4/6N1/5p1p/8/1b3pp1/3n2rq w - - bm Ne6;
k1r5/prp4p/Pp6/2P5/3N2p1/6KB/8/8 w - - bm c6;
K3b3/pNNp4/2kP1p2/P1B1p3/5P1b/3p1p2/3P1pp1/1n1r1rq1 w - - bm f5;
K6k/2bpNR2/8/5P2/1p3p2/2nn2p1/3p2p1/R7 w - - bm Ra6;
K6k/pp1n1R2/1r2p1Rp/b1p1p3/5p2/3p4/1B6/5b2 w - - bm Rfg7;
K7/1p2pN2/1B2B3/3P3n/1R1pkpP1/P7/PpbPPb2/1q3r2 w - - bm Bc7;
K7/5BN1/2p5/3RBP2/2pPk2P/pPPbpp2/2p1pq1p/2b4n w - - bm Ne8;
K7/p5p1/rpNb2p1/1p2p3/kPp5/qr4N1/3Q2p1/n2B2Rb w - - bm Ne4;
Kn6/prp1N2p/8/rq1pP1n1/1p1PkP1P/P1P1P3/3PB2b/4N3 w - - bm Bg4;
N1b3Kb/3pp1Bp/p2N4/kp1PrP1P/r1nP2rq/P3pp2/QN3P2/5B2 w - - bm Qb3;
N1K1k1br/3p1n1r/2p2pN1/3p1N2/2P2q1P/8/P2P4/8 w - - bm Nc7+;
N1K2b2/1p1Q4/kB6/Pp4p1/3P1q2/n1p4n/N4pbp/r7 w - - bm Qd6;
N1K5/1B4n1/2Pp4/k1NP2r1/Pp3n2/1Ppp4/Q2B4/rq6 w - - bm Nb6;
n1n1b3/P1P3p1/P1P1K1p1/3NP3/7p/4p1pP/4P1P1/4kBR1 b - - bm Kd2;
nb1B4/qrnPp3/rkp1Pp2/p1p1pp2/2P1Np2/P7/1PN1K1P1/8 w - - bm a4;
nB5K/2PN3P/3PNP1P/1Rp1PR2/1pP5/p3rPpb/kppp2pb/q1n2B1r w - - bm Bd3;
r1b5/1pKp4/pP1P4/P6p/3pn2p/1P1k1p2/1P3p2/3B1N1N w - - bm Kd8;
r1b5/1pKp4/pP1P4/P6p/3pn3/1P1k1p1p/1P3p2/3B1N1N w - - bm Kd8;
r1b5/1pKp4/pP1P4/P7/3pn2p/1P1k1p2/1P3p1p/3B1N1N w - - bm Kd8;
r1brn2q/2pP3n/2P5/1P1k2pP/2R2pP1/p1B1pK2/B1p5/8 w - - bm Bb4;
r3n3/bR1p4/1p1Np3/1pk1K3/2pp4/8/2N5/8 w - - bm Rc7+;
r6r/6pk/p7/4p3/4Npnq/8/1RBR1Qn1/3b1K2 w - - bm Ng5+;
r7/pk5p/1p5R/nN3Qp1/2P2bP1/5P2/PPKNrq2/8 w - - bm Rxh7+;
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
chesskobra wrote: ↑Mon Jan 20, 2025 2:26 pm
I think approximations are also very interesting, especially for humans interested in puzzles. So I am keeping copies of the sets that you have posted. Thank you. Are some of the positions easier for humans (to at least guess the best move) than for engines?
Some of the problems are not nearly so difficult as the others. There are problems that SF will not solve in three minutes with 30 cores.
I used special mate solvers to solve some of them, and some took a very long time. A few required back and forth analysis using the author supplied pv along with my generated, step-wise pvs.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
There are some corrections (three positions got overwritten in the main table with inferior information).
There are also additional annotations in the seul-bon-ana.epd file that should help trace some of the positions if people want to know the background.
I guess that a thorough pounding of this test set will probably weed out some positions, but I guess there will still be at least 100 out of 120 correct in the end.
Annotations like HHdbVI.1634134.039a.88375 in field c9 refer to (for instance) the database HHdbVI as converted to EPD, at line #1634134, with move 039a (39th move for white) and game #88375.
So if you have HHdbVI you can probably find those referenced positions that have HHdbVI notes.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
For this position [d]2N5/p4P2/p4RqP/P2krb2/5RNK/2PPB1P1/3P1pP1/4nQ2 b - - I have this pv:
Nf3+ gxf3 Qh5+ Kxh5 Bg6+ Kxg6 Rg5+ Kh7 Rg7+ Kh8 Rh7+ Kg8 Rh8+ Kg7 Rh7+ Kf8 Rh8+ Ke7 Re8+ Kd7 Rd8+ Kc7 Rd7+ Kb8 Rb7+ Ka8 Rb8+ Kxa7 Rb7+ Kxa6 Ra7+ Kb5 Rxa5+ Kb4 Rb5+ Ka3 Rb3+ Ka2 Rb2+ Ka1 Ra2+ Kb1 Rb2+ Kc1 Rb1+ Kc2 Rb2+ Kd1 Rb1+ Ke2 Re1+ Kxf2 Re2+ Kg1 Rg2+
with a score of 32676, but maybe there was a bug, because I only have one mate solution, and usually I have a lot of them (I try different engines on mate problems to see how well they fare at solving).
I will re-examine this one later.
Right now, the electric company is watching me and I am supposed to curtail my electricity usage between 8AM and 10AM.
I am one of the biggest electricity using households in Federal Way, and I get constant requests to know if I want training to reduce my electric power usage.
Good thing I am not running my industrial servers right now. Several of them have 4 power supplies in the server with over 1000 watts per power supply. I imagine my electric meter would be emitting smoke, and whining in agony as it furiously spins.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.