The existence of even positions and books

[CRoberson]

It is known that playing styles exist. It seems that there are multiple good playing styles and that not one of them is perfect. IOW, there exists multiple playing styles that are effectively the best not just one best. Given that each playing style is different, they have different strengths and weaknesses.

Assuming the above is correct, the existence of truly even positions is a bit more questionable. In that, I mean finding a large list of even positions is more difficult than one may think, thus creating even opening books is more difficult than most seem to think.

For instance, taking a list of openings and cutting them off at move 5 or 10 is full of error. There are some openings that are very tricky between moves 5 and 10 while others are not. Also, there are openings that are tricky between moves 10 and 15 while others are not. One could say that given sufficient time an entity should be able to see through the tricky part. The flip side of that statement is that there exists some tricky positions that can't be thought through properly at blitz time controls. Given that, it is difficult to invent a set of positions or an opening book that is even for all time controls.

This also implies that is difficult to produce a varied book that is even for all playing styles. With that, throw in all time controls and the effort gets more difficult. Or it gets much easier: limit all openings/positions to the simple openings such as 1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 and those like that. Thus, Sicilian defense is tossed out of the book. That is a problem since the Sicilian is blacks best chance of a win against e4.

Thus, the strength of a chess playing entity is only best measure by using a book that is optimized for that entity.

This becomes entertaining when we see people trying to create a book for all programs. Also, when some try to find a universally great opening book, but only use one program as the testing engine. Even more entertaining is when a engine author creates a book and uses Rybka to tune it. Again the same issues come up: 1 good book for all TC's and 1 good book for all entities.

GM's know better than this. Each has his/her own own opening repertoire which they know quite deeply. In match play, they may diverge from that preferred repertoire for something more playable against a particular opponent. This has been going on for decades.

So, one perfect book can't exist nor can one perfect book for all perfect engines exist (assuming style differences). Also, books must vary based on specific opponents. I have seen openings that Telepath will consistently do well against a particular (quite strong) opponent, but not that well against some other opponents of similar strength.

If people are searching for that one perfect book and having loads of fun/enjoyment doing it, then that is great. I am discussing the issue not the fun.

[Dann Corbit]

Drawish positions (analysis depth at least 20 plies):

Code: Select all

1r1q1rk1/1bp1ppbp/p1np1np1/1p6/2PP4/2N1B1PP/PP1NPPB1/R2Q1RK1 w - -
1r1r1bk1/1bq2p1p/pn2p1p1/2p1P3/5P2/P1NBB3/1P3QPP/R2R2K1 b - -
1r2qrk1/3bn3/pp1p3p/n1p1p1p1/P1P5/B1PP1NPP/2Q2PB1/1R2R1K1 w - -
1r2qrk1/p3npbp/b1np4/2p2PN1/4p3/1NPPB1PB/P3Q2P/R1R3K1 b - -
2br3k/1r2q1p1/2p1pp1p/p6P/1bBpPBQ1/1P1R4/2P2PP1/1R4K1 w - -
2kr1bnr/1pp3pp/p1p1qp2/4p3/3PP1b1/N1P2N2/PP2QPPP/R1BR2K1 w - -
2r1rb2/1bq2p1k/3p1np1/p1p5/1pP1P1P1/PP2BPN1/2Q3P1/R2R1BK1 b - -
3qr1k1/1brnbp1p/pp2p1pB/3n3N/3PN3/PB3Q2/1P3PPP/3RR1K1 w - -
4r1k1/1pb3qp/p1b1r1p1/P1Pp4/3P1p2/2BB4/1R1Q1PPP/1R4K1 b - -
4r1k1/1pp1rb1p/p1nq4/3p1pPp/2nP1P2/P1QB1P2/1P2RBNK/2R5 w - -
b2rrbk1/2q2p1p/pn1p2p1/1p4P1/2nNPB1P/P1N3Q1/1PP3B1/1K1RR3 w - -
r1b1k2r/1p1nbppp/1qn1p3/p2pP3/3P4/3B1N2/PP1N1PPP/R1BQR1K1 w kq -
r1b1k2r/1p1nbppp/pq1ppn2/6B1/4PP2/1NN2Q2/PPP3PP/2KR1B1R b kq -
r1b1k2r/1p1nbppp/pq1ppn2/6B1/4PP2/1NN2Q2/PPP3PP/R3KB1R w KQkq -
r1b1k2r/1pq2ppp/p2ppn2/2b1n3/3NPP2/2NBB2P/PPP3P1/R2Q1R1K b kq -
r1b1k2r/2qnbppp/p2pp3/1p4P1/3NPP2/2N2Q2/PPP4P/2KR1B1R w kq -
r1b1k2r/pp1pppbp/2n1n1p1/qN6/2P5/2N1Q1P1/PP2PPBP/R1B1K2R w KQkq -
r1b1kb1r/1pqp1ppp/p1n1pn2/8/3NP3/2N3P1/PPP2PBP/R1BQ1RK1 b kq -
r1b1kb1r/pp1n1ppp/4p3/3pP3/3q4/3B4/PP1N1PPP/R1BQ1RK1 w kq -
r1b1kb1r/pp3ppp/2p1p3/q2n2B1/3P4/2PBPN2/P2Q1PPP/R3K2R w KQkq -
r1b1r1k1/1pq1bppp/p1nppn2/8/P2NPP2/2N1BB2/1PP3PP/R2Q1R1K b - -
r1b2rk1/1p1nbppp/p2p1n2/8/P3PB2/2NB1N2/1qP3PP/R2Q1R1K w - -
r1b2rk1/1p1nbppp/pq1p1n2/8/P3PB2/2NB1N2/1PP3PP/R2Q1R1K b - -
r1b2rk1/1p1nbppp/pq1p1n2/8/P3PB2/2NB1N2/1PP3PP/R2Q1RK1 w - -
r1b2rk1/1pq1bppp/p2p4/P3p3/2N1n3/4B3/1PP1BPPP/R2QK2R w KQ -
r1b2rk1/2q1bppp/p2p1n2/np2p3/3pP3/2P2N1P/PPBN1PP1/R1BQR1K1 w - -
r1b2rk1/2q1bppp/p2p1n2/np2p3/3PP3/5N1P/PPBN1PP1/R1BQR1K1 b - -
r1b2rk1/2qn1ppp/p1pb1n2/1p2p3/3PP3/2N2N2/PPQ1BPPP/R1BR2K1 w - -
r1b2rk1/pp2ppbp/3p1np1/q3P1B1/3Q4/2N2P2/PPP3PP/2KR1B1R w - -
r1b2rk1/ppq1ppbp/5np1/n1p3B1/2B1P3/2P2N2/PP1NQPPP/R4RK1 w - -
r1b3k1/ppq2pnp/1n1pr1p1/2p3B1/2P5/2PBR2P/P1QN1PP1/4R1K1 w - -
r1bq1r1k/bpp2ppp/p1np4/4p3/4P1n1/1BPP1N2/PP1NRPPP/R1BQ2K1 w - -
r1bq1rk1/1pp1ppbp/p2p1np1/n2P4/2P5/2N2NP1/PP2PPBP/R1BQ1RK1 w - -
r1bq1rk1/1pp2ppp/p1np1n2/2b1p3/4P3/1BPP1N2/PP3PPP/RNBQ1RK1 w - -
r1bq1rk1/1ppnb1pp/p1p2p2/4p3/2N1P3/3P1N2/PPP2PPP/R1BQ1RK1 w - -
r1bq1rk1/2p1bppp/p2p1n2/np2p3/4P3/1BP2N1P/PP1P1PP1/RNBQR1K1 w - -
r1bq1rk1/2p2p2/2pp3p/p1b3pn/3NP3/2P3B1/PP1N1PPP/R2Q1RK1 b - -
r1bq1rk1/4bppp/p2p1n2/npp1p3/4P3/2PP1N2/PPB2PPP/RNBQR1K1 w - -
r1bq1rk1/5ppp/p1np1b2/1p1Np3/4P3/N1P5/PP3PPP/R2QKB1R w KQ -
r1bq1rk1/p3ppb1/1p1p2p1/2pP2Np/PnP4P/2N1P1P1/1P1Q1PB1/R3R1K1 w - -
r1bq1rk1/p4pbp/1pn1pnp1/3p4/3P1N2/2N1P1P1/PP3PBP/R1BQ1RK1 w - -
r1bq1rk1/pp1np1bp/2p1p1p1/8/6n1/2N2N2/PPP1BPPP/R1BQR1K1 w - -
r1bq1rk1/pp1pppbp/2n2np1/1Bp5/4P3/2P2N2/PP1P1PPP/RNBQR1K1 w - -
r1bq1rk1/pp1pppbp/2n2np1/8/2BNP3/2N1B3/PPP2PPP/R2QK2R w KQ -
r1bq1rk1/pp1pppbp/2n2np1/8/3NP3/1BN1B3/PPP2PPP/R2QK2R b KQ -
r1bq1rk1/pp3pp1/n3pb1p/3P4/3p4/P1N1PN2/1PQ2PPP/R3KB1R w KQ -
r1bq1rk1/ppp2ppp/2np1n2/2b1p3/2P5/2NP1NP1/PP2PPBP/R1BQ1RK1 b - -
r1bq1rk1/pppn1pbp/3p1np1/4p3/2PPP3/2N2NP1/PP3PBP/R1BQ1RK1 b - -
r1bqk1nr/pp1pppbp/2n3p1/2p5/2P1P3/2N4P/PP1P1PP1/R1BQKBNR w KQkq -
r1bqk1nr/pp1pppbp/2n3p1/2p5/2P5/2N1P1P1/PP1P1PBP/R1BQK1NR b KQkq -
r1bqk1nr/pp1pppbp/2n3p1/2p5/2P5/P1N3P1/1P1PPPBP/R1BQK1NR b KQkq -
r1bqk1nr/pp1pppbp/2n3p1/2p5/4P3/3P2P1/PPP2PBP/RNBQK1NR w KQkq -
r1bqk1nr/pp2ppbp/2np2p1/2p5/4P3/2NP2P1/PPP2PBP/R1BQK1NR w KQkq -
r1bqk2r/5ppp/p1n1p3/1pnpP3/5P2/2N2N2/PPPQ2PP/R3KB1R w KQkq -
r1bqk2r/p1pp1ppp/2p2n2/8/1b2P3/2N5/PPP2PPP/R1BQKB1R w KQkq -
r1bqk2r/pp3pp1/2pp1n1p/2b3B1/3NP3/1PN5/1PP2PPP/R2QK2R w KQkq -
r1bqk2r/pppp1ppp/2n2n2/8/1bBPP3/5N2/PP3PPP/RNBQK2R w KQkq -
r1bqkb1r/3n1pp1/p2ppn1p/1p6/3NP1P1/2N1BP2/PPPQ3P/2KR1B1R b kq -
r1bqkb1r/pp1pnppp/2n1p3/2p5/4P3/3P1N2/PPP1QPPP/RNB1KB1R w KQkq -
r1bqkb1r/pp1ppppp/2n5/2pnP3/8/2P2N2/PP1P1PPP/RNBQKB1R w KQkq -
r1bqkb1r/pp2pppp/2np1n2/2p5/4P3/2P2N2/PP1PBPPP/RNBQK2R w KQkq -
r1bqkb1r/pppn1ppp/4pn2/3p2B1/3P4/2N2N2/PPP1PPPP/R2QKB1R w KQkq -
r1bqkbnr/1pp3pp/p1p2p2/4p3/4P3/2N2N2/PPPP1PPP/R1BQK2R w KQkq -
r1bqkbnr/pp1p1ppp/2n5/1N2p3/4P3/8/PPP2PPP/RNBQKB1R b KQkq -
r1bqkbnr/pp2pp1p/2np2p1/2p5/4P3/5NP1/PPPP1PBP/RNBQK2R w KQkq -
r1bqkbnr/pp2pp1p/2np2p1/2p5/4PP2/2NP4/PPP3PP/R1BQKBNR w KQkq -
r1bqkbnr/pppp1ppp/2n5/4p3/3PP3/5N2/PPP2PPP/RNBQKB1R b KQkq -
r1bqnr1k/1p2n1bp/8/p1pP2pB/P1P1Pp2/2N5/1B1N2PP/R2Q1RK1 b - -
r1bqr1k1/pp2bppp/2p2n2/4p3/2P5/2N3P1/PPQ1PPBP/R1B2RK1 w - -
r1bqr1k1/ppp2pbp/3p1np1/1N1P4/2PpP3/3B4/PP3PPP/R1BQ1RK1 w - -
r1bqr1k1/pppp1ppp/2n2n2/6N1/2P1p3/2P2PP1/P2PP1BP/R1BQ1RK1 b - -
r1r3k1/3nqpp1/1pp1pn1p/p2pN3/2PP1Q2/1P4P1/P2BPPKP/2RR4 b - -
r2q1rk1/1bp1bppp/p1np1n2/1p2p3/3PP3/1BP2N1P/PP3PP1/RNBQR1K1 b - -
r2q1rk1/1p1nbppp/p3pn2/2p5/2Q5/1P2PN2/PB1P1PPP/RN3RK1 w - -
r2q1rk1/bpp2pp1/p1npbn1p/4p3/4P3/1BPP1N1P/PP1N1PP1/R1BQR1K1 w - -
r2q1rk1/p1p2ppp/2p1pb2/3n1b2/3P4/P4N1P/1PP2PP1/RNBQ1RK1 w - -
r2qk2r/1b1nbppp/pp1p1n2/4p3/P3P3/2N3P1/1PP1NPBP/R1BQ1RK1 w kq -
r2qk2r/pp1n1pp1/2pbpnp1/8/2pP4/2N1P3/PPQBBPPP/R3K2R w KQkq -
r2qkb1r/1b1n1pp1/p2ppn1p/8/Np1NP1PP/4BP2/PPPQ4/2KR1B1R b kq -
r2qr1k1/1bp1bppp/p1np1n2/1p2p3/3PP3/1BP2N1P/PP1N1PP1/R1BQR1K1 b - -
r2qr1k1/pb2bpp1/5n1p/1p1pN3/2pP1BP1/P1R1P3/1PQ2P1P/1B3RK1 w - -
r2qr1k1/pp3ppp/2n2nb1/1P4B1/3p4/P2B1P2/2P1N1PP/R2Q1RK1 b - -
r2qrbk1/1b1n1pp1/p2p3p/1ppP4/Pn2P3/R4N1P/1P1N1PP1/1BBQR1K1 b - -
r2qrbk1/1b3pp1/p2p1n1p/1ppP4/Pn2P3/5N1P/1P1N1PP1/RBBQR1K1 b - -
r2qrbk1/1bp2pp1/p1np1n1p/1p6/P2PP3/5N1P/1PBN1PP1/R1BQR1K1 b - -
r2qrbk1/1bp2ppp/p1np1n2/1p2p3/P2PP3/1BP2N1P/1P1N1PP1/R1BQR1K1 b - -
r2r2k1/p2n1p2/1q3n1p/3p2p1/2p5/P1N1P1B1/1PQ2PPP/R2R2K1 b - -
r2r2k1/pb3q2/1p1n1p1p/1B1p2p1/1QpP4/2P1P3/P3NPPP/1R2R1K1 b - -
r3k2r/1bqnbppp/pp1ppn2/8/2PNP3/2N1B1P1/PP3PBP/R2QR1K1 w kq -
r3kb1r/pp1bnppp/1qn1p3/3pP3/1P1P2P1/P4N2/1B3P1P/RN1QKB1R w KQkq -
r3kb1r/pp1n1ppp/2p1p1b1/q7/2BP2PP/2N5/PPP2P2/R1BQK2R w KQkq -
r3r1k1/1pq2pbp/p1ppbnp1/4n3/2P1PB2/1NN2P2/PP1Q2PP/R3RBK1 w - -
r4rk1/pp1nqpp1/2pbpn1p/3p1b2/2PP4/1PN3P1/PB1NPPBP/R2Q1RK1 w - -
r4rk1/pp1qppbp/2p1b1p1/n7/P2PpB1N/2P3P1/2P1QPBP/1R2R1K1 w - -
r4rk1/pppq1ppp/2n2n2/2bp4/3P2b1/2NBBN2/PP3PPP/R2Q1RK1 w - -
rn1q1rk1/1b2bppp/pp1p1n2/4p3/4P3/1NN1B3/PPP1BPPP/R2Q1R1K w - -
rn1q1rk1/1bp1bppp/p3pn2/8/1p1P4/2NQ1NP1/PP2PPBP/R1B2RK1 w - -
rn1q1rk1/pb3ppp/1p3b2/2pp4/3P4/P1N2NP1/1P2PPBP/R2QK2R w KQ -
rn1q1rk1/pbp2ppp/1p1ppn2/6B1/2PP4/P1Q1PP2/1P4PP/R3KBNR b KQ -
rn1q1rk1/pbpp1ppp/1p2pn2/6B1/2PP4/P1Q1P3/1P3PPP/R3KBNR b KQ -
rn1q1rk1/pbpp1ppp/1p2pn2/6B1/2PP4/P1Q2P2/1P2P1PP/R3KBNR b KQ -
rn1q1rk1/pbpp1ppp/1p2pn2/8/1PP5/P1Q2N2/1B1PPPPP/R3KB1R b KQ -
rn1qk2r/p1ppbppp/bp2pn2/8/2PP4/1P3NP1/P2BPP1P/RN1QKB1R w KQkq -
rn1qk2r/p3bppp/bpp1pn2/3p4/2PP4/1PB2NP1/P3PPBP/RN1QK2R w KQkq -
rn1qkb1r/1b3ppp/p3pn2/1pp5/3P4/1B2PN2/PP2QPPP/RNBR2K1 b kq -
rn1qkb1r/pb1p1ppp/1p2pn2/8/2PpP3/P1N2N2/1PQ2PPP/R1B1KB1R w KQkq -
rn1qkb1r/pp2pppp/2p2n2/3p4/6b1/3P1NP1/PPP1PPBP/RNBQK2R w KQkq -
rn1qkb1r/pp2pppp/2p2n2/4Nb2/P1pP4/2N5/1P2PPPP/R1BQKB1R b KQkq -
rn1qkb1r/ppp2ppp/4p3/3n1b2/3P4/5N1P/PPP1BPP1/RNBQK2R b KQkq -
rn2k3/p4ppr/1p2p1n1/2ppP1Bp/q2P3P/P1P2QN1/2P2PP1/R4K1R w q c6
rn2kb1r/1bq1pp1p/p2p1np1/8/1p1NPPP1/2NBBQ2/PPP4P/R3K2R w KQkq -
rn3rk1/pp2ppbp/q1p1bnp1/3p4/2PP4/1PNQ1NPP/P3PPB1/R1B2RK1 w - -
rnb1k2r/ppq1nppp/4p3/2ppP3/3P2Q1/P1P5/2P2PPP/R1B1KBNR w KQkq -
rnb1kb1r/1p3ppp/pq1ppn2/6B1/3NPP2/2N5/PPP3PP/R2QKB1R w KQkq -
rnb1kb1r/1p3ppp/pq1ppn2/6B1/4PP2/1NN5/PPP3PP/R2QKB1R b KQkq -
rnb2rk1/1p2bppp/p1qp1n2/4p3/4P1P1/1NN5/PPP1BP1P/R1BQ1R1K w - -
rnb2rk1/2q1bppp/pp1ppn2/8/P2NPP2/2N5/1PP1B1PP/R1BQ1R1K w - -
rnbq1rk1/5pbp/p2p1np1/1ppP2B1/4P3/2N5/PP1NBPPP/R2QK2R w KQ -
rnbq1rk1/p1pp1ppp/1p2pn2/8/2PP4/P1Q5/1P2PPPP/R1B1KBNR w KQ -
rnbq1rk1/pp2nppp/4p3/2ppP3/3P2Q1/P1PB4/2P2PPP/R1B1K1NR b KQ -
rnbq1rk1/pp2ppbp/3p1np1/2p5/2PPPP2/2N2N2/PP4PP/R1BQKB1R w KQ -
rnbq1rk1/pp3pbp/3ppnp1/2pP4/2P1PP2/2N2N2/PP4PP/R1BQKB1R w KQ -
rnbq1rk1/ppp1ppbp/3p1np1/8/2PPP3/2N1BP2/PP4PP/R2QKBNR b KQ -
rnbq1rk1/ppp1ppbp/3p1np1/8/2PPPP2/2N2N2/PP4PP/R1BQKB1R b KQ -
rnbq1rk1/ppp1ppbp/3p1np1/8/2PPPP2/2N5/PP4PP/R1BQKBNR w KQ -
rnbq1rk1/pppp1ppp/5n2/2b1p3/2B1P3/2NP4/PPP2PPP/R1BQK1NR w KQ -
rnbqk2r/1p2bppp/p2ppn2/6B1/3NPP2/2N2Q2/PPP3PP/R3KB1R b KQkq -
rnbqk2r/1p2ppb1/p2p4/6pp/3NP1n1/2N3B1/PPP1BPPP/R2QK2R w KQkq -
rnbqk2r/1pp1bp1p/p3pp2/8/3PN3/5NP1/PPP2P1P/R2QKB1R b KQkq -
rnbqk2r/pp1p1ppp/4pn2/2P5/1bP5/2N5/PPQ1PPPP/R1B1KBNR b KQkq -
rnbqk2r/ppp1ppbp/3p1np1/8/2PPPP2/2N5/PP4PP/R1BQKBNR b KQkq -
rnbqk2r/ppp1ppbp/5np1/3p4/3P4/P1P2N2/1P2PPPP/RNBQKB1R w KQkq -
rnbqkb1r/1p3ppp/p2ppn2/6B1/3NPP2/2N5/PPP3PP/R2QKB1R b KQkq -
rnbqkb1r/p1p2ppp/1p2pn2/3p4/4P3/3P1N2/PPPN1PPP/R1BQKB1R w KQkq -
rnbqkb1r/pp1p1ppp/4pn2/2p5/2PP4/2N2N2/PP2PPPP/R1BQKB1R b KQkq -
rnbqkb1r/pp1p1ppp/4pn2/2p5/3P4/5NP1/PPP1PP1P/RNBQKB1R w KQkq -
rnbqkb1r/pp1ppppp/5n2/2p5/2P5/1P3N2/P2PPPPP/RNBQKB1R b KQkq -
rnbqkb1r/pp2pppp/2p2n2/8/2pPP3/2N2N2/PP3PPP/R1BQKB1R b KQkq -
rnbqkb1r/pp3ppp/4pn2/2pp4/3P4/2PBPN2/PP3PPP/RNBQK2R b KQkq -
rnbqkb1r/pp3ppp/4pn2/2pp4/3P4/4PN2/PPPN1PPP/R1BQKB1R w KQkq -
rnbqkb1r/pp4pp/4pn2/2pp1pB1/3P4/2N1P1P1/PPP1NPBP/R2QK2R b KQkq -
rnbqkb1r/ppp1pppp/5n2/3p4/3P4/5NP1/PPP1PP1P/RNBQKB1R b KQkq -
rnbqkb1r/ppp1pppp/5n2/3p4/3P4/6P1/PPP1PPBP/RNBQK1NR b KQkq -
rnbqkb1r/ppp1pppp/5n2/3p4/5P2/5N2/PPPPP1PP/RNBQKB1R w KQkq -
rnbqkb1r/ppp1pppp/5n2/3p4/8/1P3N2/P1PPPPPP/RNBQKB1R w KQkq -
rnbqkb1r/ppp1pppp/5n2/3p4/8/5NP1/PPPPPP1P/RNBQKB1R w KQkq -
rnbqkb1r/ppp2ppp/4pn2/3p4/3P4/2N2N2/PPP1PPPP/R1BQKB1R w KQkq -
rnbqkb1r/pppp1ppp/5n2/4p3/2P5/2N1P3/PP1P1PPP/R1BQKBNR b KQkq -
rnbqkb1r/pppp1ppp/5n2/4p3/2P5/6P1/PP1PPP1P/RNBQKBNR w KQkq -
rnbqkbnr/ppp2ppp/4p3/3p4/2P5/1P3N2/P2PPPPP/RNBQKB1R b KQkq -

[CRoberson]

Drawish positions (analysis depth at least 20 plies):

Hi Dann,

While the existence of drawish positions is a fact, my point is about how easy or difficult it is to find them and prove that they are even. Certainly, taking a book of GM moves and cutting them off at an arbitrary depth isn't the way to do it.

I looked at 2 of your positions quickly. How many engines did you use to at least 20 ply to test them as drawish ?

I don't know how to prove (with current technology) that all of these positions are even. I don't believe a 20 ply search is enough to prove it. In fact, I am sure that is insufficient for the first 2 positions. The horizon effect has impact on those positions beyond 20 ply.

Numerous Shootouts or large scale Monte Carlo analysis might be the best that can be done.

This discussion begs for a definition of even positions. While drawish positions are certainly even, do there exist positions that are even but are not drawish? As in, both sides have even chances at winning. Are those positions guaranteed drawish?

[bob]

Drawish positions (analysis depth at least 20 plies):

Hi Dann,

While the existence of drawish positions is a fact, my point is about how easy or difficult it is to find them and prove that they are even. Certainly, taking a book of GM moves and cutting them off at an arbitrary depth isn't the way to do it.

I looked at 2 of your positions quickly. How many engines did you use to at least 20 ply to test them as drawish ?

I don't know how to prove (with current technology) that all of these positions are even. I don't believe a 20 ply search is enough to prove it. In fact, I am sure that is insufficient for the first 2 positions. The horizon effect has impact on those positions beyond 20 ply.

Numerous Shootouts or large scale Monte Carlo analysis might be the best that can be done.

This discussion begs for a definition of even positions. While drawish positions are certainly even, do there exist positions that are even but are not drawish? As in, both sides have even chances at winning. Are those positions guaranteed drawish?

I think you are searching for the pot 'o gold at the end of the rainbow. What is "even" to one program is anything but to another. Some "even" positions lead to endgames a program has no clue about. Others lead to endgames it plays like a GM. And not all programs are created equal in this regard. I think that trying to find a set that works for everyone is futile.

I chose mine, not because any one program plays them well or thinks they are equal, but rather because they represent a wide cross-section of potential openings, so that I can be fairly comfortable in making changes and verifying that they don't work well in some types of positions, while wrecking many more types.

Tuning the book for tournaments is a completely different task that is quite painful and time-consuming, hence the professional book authors that do nothing but that stuff. I'm happy enough with the positions I am using, for the purpose I am using them.

[Dann Corbit]

Drawish positions (analysis depth at least 20 plies):

Hi Dann,

While the existence of drawish positions is a fact, my point is about how easy or difficult it is to find them and prove that they are even. Certainly, taking a book of GM moves and cutting them off at an arbitrary depth isn't the way to do it.

I looked at 2 of your positions quickly. How many engines did you use to at least 20 ply to test them as drawish ?

I don't know how to prove (with current technology) that all of these positions are even. I don't believe a 20 ply search is enough to prove it. In fact, I am sure that is insufficient for the first 2 positions. The horizon effect has impact on those positions beyond 20 ply.

Numerous Shootouts or large scale Monte Carlo analysis might be the best that can be done.

This discussion begs for a definition of even positions. While drawish positions are certainly even, do there exist positions that are even but are not drawish? As in, both sides have even chances at winning. Are those positions guaranteed drawish?

It is easy to come up with 100 positions that are forced draws (by ring-around-the-rosey chase) but of course these positions are not at all interesting.

The positions I supplied are probably not perfectly drawish. But if you are looking for a set of positions to test engines with, they are close enough to even for all practical purposes.

I filtered for lots of other practical things such as:
1. Board is highly populated
2. No checks in the pv

Of course, thorough testing will probably show that 2% of the supplied positions really do contain a significant advantage for one side or the other. At the point that these positions are discovered, simply throw them out.

It may also be true that some of the supplied positions are only even if played almost perfectly (it is not unlikely, for instance, that some of them are gambits analyzed to 30 plies and very likely to crash and burn at blitz if played improperly).

I guess that really your ultimate goal is not fully clear to me.

[Norm Pollock]

When you say "even" positions, aren't you implying that you want to have starting positions for the engines that are evaluated by both engines as "0.00"? If so, then aren't you saying that want starting positions that are slightly in Black's favor relative to the evaluation of the original opening position, which evaluates (just a guess) at around "+0.10" in White's favor?

To me, the idea of a "fair" or "even" opening book or epd file, is one in which neither side improves upon the original opening position, and therefore all starting positions for the engines are around "+0.10" millipawns in White's favor.

[Kirill Kryukov]

It is known that playing styles exist. It seems that there are multiple good playing styles and that not one of them is perfect. IOW, there exists multiple playing styles that are effectively the best not just one best. Given that each playing style is different, they have different strengths and weaknesses.

Assuming the above is correct, the existence of truly even positions is a bit more questionable. In that, I mean finding a large list of even positions is more difficult than one may think, thus creating even opening books is more difficult than most seem to think.

Any discussion about existance of even positions should begin with defining a context. Examples of a context: super-GM vs super-GM classic TC, internet bullet, CCRL 40/40 engine vs engine list, humans 2600+, humans 2400-2600, all human games, human games after 1990, etc. We also can imagine the "perfect player" context, like 32-men EGTB. Some positions may be even in one context and not in other.

(Also "even" may need some kind of definition, like 50% draw rate).

While we can expect at least some overlap between the sets of even positions from different contexts, ultimately only statistics gathered for particular position in particular context can tell whether a position is truly even in that context. I think there is no magic workaround here.

I am trying to evaluate evenness of opening positions that occurred in my engine tourney: Link. However the number of games is probably not enough for any interesting discussion.

For instance, taking a list of openings and cutting them off at move 5 or 10 is full of error. There are some openings that are very tricky between moves 5 and 10 while others are not. Also, there are openings that are tricky between moves 10 and 15 while others are not. One could say that given sufficient time an entity should be able to see through the tricky part. The flip side of that statement is that there exists some tricky positions that can't be thought through properly at blitz time controls. Given that, it is difficult to invent a set of positions or an opening book that is even for all time controls.

Agreed. Moreover I think it's difficult to invent a set of positiosn that are even for ONE time control.

I think that the best we can do is use many candidate positions and later remove the ones that turned out uneven.

This also implies that is difficult to produce a varied book that is even for all playing styles. With that, throw in all time controls and the effort gets more difficult. Or it gets much easier: limit all openings/positions to the simple openings such as 1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 and those like that. Thus, Sicilian defense is tossed out of the book. That is a problem since the Sicilian is blacks best chance of a win against e4.

You don't need an "even book for all playing styles". What is a purpose of an even opening book? To provide a variety of opening for particular event. Therefore it's enough if the book is "even" in the broad context of that event. If a particular pair of players always take a particular opening to 1-0, that might be a problem for that pair, but not for the event as a whole. I think it's OK to not worry about particular pairings, otherwise too much work is needed to ensure the evenness of any position (and then, all games will end up as draws).

Actually what I am trying to achieve is: Finding a set of positions that are even in the broad context of the event (white score is close to 50%), while producing few draws (say, 20% or less). Ideally I also need to make sure there is no strong correlation between engine strength and white score for each position.

Thus, the strength of a chess playing entity is only best measure by using a book that is optimized for that entity.

If you are interested in "engine+book" entities, then yes. If you are interested in "engine" entities, then no.

This becomes entertaining when we see people trying to create a book for all programs. Also, when some try to find a universally great opening book, but only use one program as the testing engine. Even more entertaining is when a engine author creates a book and uses Rybka to tune it. Again the same issues come up: 1 good book for all TC's and 1 good book for all entities.

GM's know better than this. Each has his/her own own opening repertoire which they know quite deeply. In match play, they may diverge from that preferred repertoire for something more playable against a particular opponent. This has been going on for decades.

So, one perfect book can't exist nor can one perfect book for all perfect engines exist (assuming style differences). Also, books must vary based on specific opponents. I have seen openings that Telepath will consistently do well against a particular (quite strong) opponent, but not that well against some other opponents of similar strength.

If people are searching for that one perfect book and having loads of fun/enjoyment doing it, then that is great. I am discussing the issue not the fun.