mg vs eg eval

[joost.van...@gmail.com]

I was looking at how similar endgame and midgame eval are (some kind of internal consistency), and found that there can be very large differences. I wonder if there could be something to be learned from fens where there is a large difference. For example, the following is a set of fens (encountered during a standard bench) for which the final evaluation is nearly a draw, but that near-draw value is the result of very different (and thus opposite sign) mg and eg evals. Often these fens are not a draw.

    eval       mg       eg                               FEN

=================================================================================================

       0     -115      378                  r3k3/p1ppqp1r/Bn4p1/8/4p1N1/8/PKP2PQP/3R3R w q - 0 17
       7      358     -109                  1r1k3r/p1pp1p2/Bn4p1/8/4R3/7p/PPP2PPP/R5K1 w - - 1 19
      -2      208     -690              r3k3/p1Qp1pb1/q5p1/4N3/4n1Pr/7p/PPP2P1P/1K1R3R b q - 0 17
      -3      747     -150                4r3/pp1n1k1p/6p1/2p1P1B1/P3N3/1bP5/6PP/4R1K1 b - - 4 25
      -7      164     -274                     8/pp1n1k1p/6p1/2q1r1Q1/4NB2/8/b5PP/4R2K w - - 0 27
      -5      241     -389                    4r1k1/pp1n3p/6pQ/3q4/2bpR3/5PN1/P2B2KP/8 w - - 0 25
       6      185     -272                    4r1k1/pp1n3p/6pQ/4p3/8/q2bB1NP/P4RP1/6K1 w - - 1 25
      -9      538     -117                4r3/pp1n1k1p/6p1/P1p1P3/4N3/1bP5/3B2PP/4R1K1 b - - 2 25
      -2      798     -159                4rk2/pp1n3p/3N2p1/2p5/2Pp4/2Pb4/P2B2PP/4R1K1 b - - 1 23
       8      432     -253                 6k1/pp2Q2p/4n1p1/2p1q3/2b5/2P3N1/P2B2PP/6K1 w - - 0 25
      -3      315     -104                   8/p2n2kp/1p1q2p1/2p5/2P5/2P1Q3/P2B2PP/5K2 w - - 0 26
      -8     -235      422           r3q1k1/ppp3p1/1b1pp2p/8/1n2PQ1B/5P1P/PPPK2P1/3R3R b - - 0 19
       5      137     -425                rq5k/ppp5/1bnpNr1p/8/4P1Q1/7P/PPP2PP1/1K1R3R b - - 0 19
      -5      264     -375                     r4qk1/ppp5/2npp3/7Q/8/3R3P/PPP2bP1/2K4R b - - 1 22
      -6     -157      270     N2q1r1k/1p2n1pp/1p1p4/4p2Q/4P1P1/1b1P1p1P/PPP2P2/3R1RK1 b - - 0 18
       1      211     -258                 r5k1/2p2p1p/p4p2/2p1r3/4P3/4Q3/P1q3PP/R4RK1 w - - 0 21
       5      338     -496            r1bb3r/p1k2p1p/2nN3n/6p1/3p4/6B1/PPP2PPP/2K1RB1R w - - 2 18
       3      242     -199               r3k1nr/ppB2p1p/4b3/n5p1/3R4/8/PPP2PPP/2K2B1R b kq - 0 16
       2      217     -321               r1bb1k2/pp2nN1r/7p/6p1/2Bn4/6B1/PPP2PPP/2KR3R w - - 0 18
       2      808     -685               r2b2nr/pp3p1p/2k1b3/8/2N5/6p1/PPP2PP1/2KR1B1R w - - 0 18
      -6      199     -317           r1b2knr/1p3N1p/1bn5/1p1B2p1/3p4/6B1/PPP2PPP/2KR3R b - - 0 17
       5      261     -152             1r1r4/p3k2p/2n1p3/1N4p1/3p4/6P1/PPP2n1P/2K1RBR1 w - - 0 22
      -4      186     -292             r2b1knr/pp3p1p/3N4/3bB1p1/3n4/8/PPP2PPP/2KR1B1R w - - 2 17
      -6      305     -193                N1bb2kr/1p3N1p/p1n5/6p1/2Bp4/8/PPP2PPP/2KR3n w - - 0 19
       2      661     -560            2rb2nr/pp3p1p/2n1k3/5N2/3p1p1P/8/PPP2PP1/2KR1B1R b - - 0 17
      -3      169    -1131          r3qrk1/ppp1n1pp/4n3/3p1B2/3P2P1/1PP1BN1Q/P6P/4RbK1 w - - 0 23
      -6      471    -1020          r1b1q1k1/ppp1n2p/4np2/3pN3/3P3Q/2P1p3/PPB3PP/4R1K1 w - - 0 22
       4      257     -156          r1b1r3/pp2n1pp/2p3k1/3p2B1/3Pp2P/2P5/PP1N2P1/4RRK1 w - - 2 25
       2      114     -415           r2q1rk1/ppp3pp/4n1n1/3pNp1Q/3P4/2P1B3/PPB3PP/5RK1 w - - 1 21
      -9      178     -368                 7Q/r1q1kp2/ppp2p2/4r2b/2B5/1N6/PPP3PP/4R1K1 b - - 4 22
       0      186     -356              6k1/r1q2p1p/ppp1r2Q/4Pp2/6b1/1N1B2PP/PPP5/R5K1 b - - 0 20
       2      292     -531                      5knQ/r4ppB/ppp1r3/8/8/1N2q3/PPP3PP/R6K w - - 0 24
      -2      206     -400               2b2k2/r1q2p1B/ppp2p2/4r3/8/1N4Q1/PPP3PP/2R3K1 b - - 0 21
       1      363     -270                   2b5/r1q2pkp/ppp2p2/8/3N1Q2/3B4/PPP3PP/4K3 b - - 0 22
      -4      196     -386             2b1r1k1/r1q2p1p/ppp5/4p3/8/1N1BQ2P/PPP3P1/4R1K1 w - - 0 20
       8     -259      180               4r1k1/2r2pp1/pRp2n2/7p/6b1/1N1B2P1/PPP2KP1/R7 b - - 1 21
      -2      431     -450                    4r1k1/3r1p2/ppp2Q1p/8/8/2qB4/P1P3PP/5RK1 b - - 1 25
       4      240     -443                 2b1r1k1/r4p1p/ppp2p2/8/8/1N1B4/PqP2QPP/5RK1 w - - 0 21
       6      107     -962       2rqk2r/1pp3b1/p1np1np1/3Q1P1p/2P1P3/8/PP2B1PP/R1B2R1K b - - 2 17
      -3       99     -340             2r1k3/2pn1pbr/p2p1qP1/3P3p/4P3/8/PP3QPP/R1B2RK1 b - - 0 19
       6      248     -266           2rnk2r/ppp2p2/6pb/1N2p2p/2P2Pn1/8/PP2B1PP/R1B2RK1 w - - 0 16
      -7     -165      583               2rq2k1/ppp3b1/3p4/3Pn2Q/3Nn3/6PP/PP6/R1B1K2R w KQ - 1 21
       1      198     -447                  5r2/p5kp/bp2Qpp1/8/2Pr1P2/6R1/q2N2PP/4R1K1 w - - 1 30
      -9     -146      730            r4rk1/2N2ppp/2PQ1bb1/8/2P1p2P/pP2B3/P2R1Pq1/2K4R b - - 0 24
       7     -237      165           1rr3k1/2p2ppp/2Pp1bb1/8/p1PNp3/4B2P/PP1R1PP1/1K5R w - - 1 23
      -1     -259      164              1rr3k1/2p1bppp/2Pp2b1/8/p1PNp3/4B3/PPKR1PPP/7R w - - 0 22
      -2     -252      156            r1r3k1/2p1bpp1/2Pp2bp/1N6/p1P1p3/4B3/PPKR1PPP/7R b - - 1 22
       2     -281      184          2rr2k1/2p1bppp/2P3b1/p2p4/P1PNp3/4BP2/1P1R2PP/2K4R b - - 0 22
       6     -116      286             r2q1rk1/2p2ppp/2Pp1b2/5Q2/p1PRp3/4B3/PPK2PPP/7R w - - 1 23
       0     1673     -112                        r7/5Rpp/4p3/3pN1k1/3P4/4P1P1/r4PKP/8 w - - 1 30
       1      357     -116                7r/1pb2ppp/1Pr1pk2/1R1p4/3P4/4PN2/5PPP/1R4K1 w - - 1 26
      -8      507     -180                4k2r/1R3pp1/4p3/p2pN1bp/3P4/4P3/2r2PPP/1R4K1 w - - 2 27
       8      389     -118               7r/1pbk1Npp/1p2p3/1R1p4/3P4/4P3/P1r2PPP/1R4K1 b - - 0 24
      -3      307     -107            1b2k2r/1p4pp/1p2p3/PR1p1p2/3P4/4PN2/2r2PPP/1R4K1 w - - 1 25
      -1      344     -116                r7/2b4p/3kpp2/p2p2P1/3P4/2r1PN2/1R3PP1/1R4K1 b - - 0 28
      -7     -834      154                        6k1/p4p1p/P5p1/8/1p3q2/1P5P/1N4K1/Q7 b - - 1 35
       6    -1134      124                             1Q6/6pk/8/P7/1Pq2p1p/5P2/3K1P2/8 b - - 4 6
       0     1072     -111                               8/Q7/2k3p1/P7/8/5P1p/4KP2/6q1 w - - 9 11
       0      254     -190                      1r6/4q3/2PpBk1b/8/5pQ1/1p1P4/1P2KP2/3N4 w - - 2 6
       6      509     -189                   1r6/4q2Q/2Pp1k2/4p3/5p2/1p1P2P1/1P2KP2/3N4 w - - 3 6
      -3      332     -255                  2r5/2q3k1/3p3b/4pQ2/4Bp2/1p1P2P1/1P2KP2/3N4 b - - 3 4
       3      245     -179                     2r5/6k1/3p1q2/3Bp1b1/4Q3/1p1PN3/1P2KP2/8 w - - 0 8
       9     -343      274                       5r2/3Q4/2Pp3k/4p1q1/8/1BNP4/1P2KP2/2b5 w - - 7 8
      -6      759     -305                         7r/2q5/3p1k2/3Bp3/8/1p1PP1Q1/1P2K3/8 w - - 0 9
       7      452     -327                  1r6/2q3k1/2Pp3b/4pQ2/2B5/1p1P2p1/1P2KP2/3N4 b - - 3 4
      -5      298     -124                         1r6/2k3q1/2PpQ3/4p3/8/1B1P4/1P2KP2/8 w - - 1 9
      -5     -399      291                      5r1k/2P5/3p2Qb/3Bp3/5q2/1p1PN3/1P2KP2/8 b - - 1 6
       9     -413      328                      5r1k/2PQ4/3p1q1b/3Bp3/8/1p1PN3/1P2KP2/8 w - - 3 9
       2      485     -186                        1r5k/2q5/3p4/3Bp3/6Q1/1p1P4/1P3P2/5K2 b - - 1 8
       4      258     -186                  4r3/2q3k1/3p3b/3Bp2Q/5p2/1p1P2P1/1P2KP2/3N4 b - - 3 4
       5      813     -447                    r4n2/pp3k2/2p1p2p/4b3/2P1rPQP/8/P4P1K/6R1 w - - 0 7

[SapphireBrand]

Do you happen to know which positions are winning?

[algot...@gmail.com]

it is always interesting to have a different angle to study a problem.

studying where comes the difference in some positions might be enlightening.

but one must consider the following facts

the midgame eval: when more pieces on the board.

the endgame eval: when few pieces on the board.

for example, kingdanger is mostly mg, and not much eg,

and there are bonus such as space which are mg only and not scored as soon as queens are off the board.

others, such as threat on pawns and initiative are mostly eg, etc, etc...

also a quick look at our current piece values will produce some large difference, 

a soon as white or black material is different

  PawnValueMg   = 128,   PawnValueEg   = 213,

  KnightValueMg = 782,   KnightValueEg = 865,

  BishopValueMg = 830,   BishopValueEg = 918,

  RookValueMg   = 1289,  RookValueEg   = 1378,

  QueenValueMg  = 2529,  QueenValueEg  = 2687,

so it is highly expected that mg and eg totals will be quite different.

the only important value is the final value 

which is an interpolation between the 2 according to current nonpawnmaterial mg value

ScaleFactor sf = scale_factor(eg_value(score));

    v =  mg_value(score) * int(me->game_phase())

        + eg_value(score) * int(PHASE_MIDGAME - me->game_phase()) * sf / SCALE_FACTOR_NORMAL;

    v /= PHASE_MIDGAME;

[joost.van...@gmail.com]

should have added search evaluation as well, now last column search (note different units, internal eval vs. search in cp).

While I agree that eval is obviously more important that either mg or eg component, there might be info hidden.

    eval       mg       eg                               FEN                                        search(d24)

===============================================================================================================

       0     -115      378                  r3k3/p1ppqp1r/Bn4p1/8/4p1N1/8/PKP2PQP/3R3R w q - 0 17      cp   244

       7      358     -109                  1r1k3r/p1pp1p2/Bn4p1/8/4R3/7p/PPP2PPP/R5K1 w - - 1 19      cp     0
      -2      208     -690              r3k3/p1Qp1pb1/q5p1/4N3/4n1Pr/7p/PPP2P1P/1K1R3R b q - 0 17      cp   743
      -3      747     -150                4r3/pp1n1k1p/6p1/2p1P1B1/P3N3/1bP5/6PP/4R1K1 b - - 4 25      cp    40
      -7      164     -274                     8/pp1n1k1p/6p1/2q1r1Q1/4NB2/8/b5PP/4R2K w - - 0 27      cp  1194
      -5      241     -389                    4r1k1/pp1n3p/6pQ/3q4/2bpR3/5PN1/P2B2KP/8 w - - 0 25      cp  2200
       6      185     -272                    4r1k1/pp1n3p/6pQ/4p3/8/q2bB1NP/P4RP1/6K1 w - - 1 25      cp   -92
      -9      538     -117                4r3/pp1n1k1p/6p1/P1p1P3/4N3/1bP5/3B2PP/4R1K1 b - - 2 25      cp     0
      -2      798     -159                4rk2/pp1n3p/3N2p1/2p5/2Pp4/2Pb4/P2B2PP/4R1K1 b - - 1 23      cp   222
       8      432     -253                 6k1/pp2Q2p/4n1p1/2p1q3/2b5/2P3N1/P2B2PP/6K1 w - - 0 25      cp     0
      -3      315     -104                   8/p2n2kp/1p1q2p1/2p5/2P5/2P1Q3/P2B2PP/5K2 w - - 0 26      cp   -28
      -8     -235      422           r3q1k1/ppp3p1/1b1pp2p/8/1n2PQ1B/5P1P/PPPK2P1/3R3R b - - 0 19      cp   447
       5      137     -425                rq5k/ppp5/1bnpNr1p/8/4P1Q1/7P/PPP2PP1/1K1R3R b - - 0 19      cp   735
      -5      264     -375                     r4qk1/ppp5/2npp3/7Q/8/3R3P/PPP2bP1/2K4R b - - 1 22      cp   284
      -6     -157      270     N2q1r1k/1p2n1pp/1p1p4/4p2Q/4P1P1/1b1P1p1P/PPP2P2/3R1RK1 b - - 0 18      cp   532
       1      211     -258                 r5k1/2p2p1p/p4p2/2p1r3/4P3/4Q3/P1q3PP/R4RK1 w - - 0 21      cp   -13
       5      338     -496            r1bb3r/p1k2p1p/2nN3n/6p1/3p4/6B1/PPP2PPP/2K1RB1R w - - 2 18      cp   371
       3      242     -199               r3k1nr/ppB2p1p/4b3/n5p1/3R4/8/PPP2PPP/2K2B1R b kq - 0 16      cp   286
       2      217     -321               r1bb1k2/pp2nN1r/7p/6p1/2Bn4/6B1/PPP2PPP/2KR3R w - - 0 18      cp   357
       2      808     -685               r2b2nr/pp3p1p/2k1b3/8/2N5/6p1/PPP2PP1/2KR1B1R w - - 0 18      cp  -383
      -6      199     -317           r1b2knr/1p3N1p/1bn5/1p1B2p1/3p4/6B1/PPP2PPP/2KR3R b - - 0 17      cp   186
       5      261     -152             1r1r4/p3k2p/2n1p3/1N4p1/3p4/6P1/PPP2n1P/2K1RBR1 w - - 0 22      cp   292
      -4      186     -292             r2b1knr/pp3p1p/3N4/3bB1p1/3n4/8/PPP2PPP/2KR1B1R w - - 2 17      cp   826
      -6      305     -193                N1bb2kr/1p3N1p/p1n5/6p1/2Bp4/8/PPP2PPP/2KR3n w - - 0 19      cp   699
       2      661     -560            2rb2nr/pp3p1p/2n1k3/5N2/3p1p1P/8/PPP2PP1/2KR1B1R b - - 0 17      cp   883
      -3      169    -1131          r3qrk1/ppp1n1pp/4n3/3p1B2/3P2P1/1PP1BN1Q/P6P/4RbK1 w - - 0 23    mate     7
      -6      471    -1020          r1b1q1k1/ppp1n2p/4np2/3pN3/3P3Q/2P1p3/PPB3PP/4R1K1 w - - 0 22      cp  1094
       4      257     -156          r1b1r3/pp2n1pp/2p3k1/3p2B1/3Pp2P/2P5/PP1N2P1/4RRK1 w - - 2 25      cp   126
       2      114     -415           r2q1rk1/ppp3pp/4n1n1/3pNp1Q/3P4/2P1B3/PPB3PP/5RK1 w - - 1 21      cp     5
      -9      178     -368                 7Q/r1q1kp2/ppp2p2/4r2b/2B5/1N6/PPP3PP/4R1K1 b - - 4 22      cp  1804
       0      186     -356              6k1/r1q2p1p/ppp1r2Q/4Pp2/6b1/1N1B2PP/PPP5/R5K1 b - - 0 20      cp  2349
       2      292     -531                      5knQ/r4ppB/ppp1r3/8/8/1N2q3/PPP3PP/R6K w - - 0 24      cp   163
      -2      206     -400               2b2k2/r1q2p1B/ppp2p2/4r3/8/1N4Q1/PPP3PP/2R3K1 b - - 0 21      cp   287
       1      363     -270                   2b5/r1q2pkp/ppp2p2/8/3N1Q2/3B4/PPP3PP/4K3 b - - 0 22      cp  2465
      -4      196     -386             2b1r1k1/r1q2p1p/ppp5/4p3/8/1N1BQ2P/PPP3P1/4R1K1 w - - 0 20      cp    39
       8     -259      180               4r1k1/2r2pp1/pRp2n2/7p/6b1/1N1B2P1/PPP2KP1/R7 b - - 1 21      cp    51
      -2      431     -450                    4r1k1/3r1p2/ppp2Q1p/8/8/2qB4/P1P3PP/5RK1 b - - 1 25      cp   530
       4      240     -443                 2b1r1k1/r4p1p/ppp2p2/8/8/1N1B4/PqP2QPP/5RK1 w - - 0 21      cp  -184
       6      107     -962       2rqk2r/1pp3b1/p1np1np1/3Q1P1p/2P1P3/8/PP2B1PP/R1B2R1K b - - 2 17      cp  1323
      -3       99     -340             2r1k3/2pn1pbr/p2p1qP1/3P3p/4P3/8/PP3QPP/R1B2RK1 b - - 0 19      cp   540
       6      248     -266           2rnk2r/ppp2p2/6pb/1N2p2p/2P2Pn1/8/PP2B1PP/R1B2RK1 w - - 0 16      cp   101
      -7     -165      583               2rq2k1/ppp3b1/3p4/3Pn2Q/3Nn3/6PP/PP6/R1B1K2R w KQ - 1 21      cp   315
       1      198     -447                  5r2/p5kp/bp2Qpp1/8/2Pr1P2/6R1/q2N2PP/4R1K1 w - - 1 30      cp  -299
      -9     -146      730            r4rk1/2N2ppp/2PQ1bb1/8/2P1p2P/pP2B3/P2R1Pq1/2K4R b - - 0 24      cp   314
       7     -237      165           1rr3k1/2p2ppp/2Pp1bb1/8/p1PNp3/4B2P/PP1R1PP1/1K5R w - - 1 23      cp   -51
      -1     -259      164              1rr3k1/2p1bppp/2Pp2b1/8/p1PNp3/4B3/PPKR1PPP/7R w - - 0 22      cp   -29
      -2     -252      156            r1r3k1/2p1bpp1/2Pp2bp/1N6/p1P1p3/4B3/PPKR1PPP/7R b - - 1 22      cp   141
       2     -281      184          2rr2k1/2p1bppp/2P3b1/p2p4/P1PNp3/4BP2/1P1R2PP/2K4R b - - 0 22      cp   166
       6     -116      286             r2q1rk1/2p2ppp/2Pp1b2/5Q2/p1PRp3/4B3/PPK2PPP/7R w - - 1 23      cp     0
       0     1673     -112                        r7/5Rpp/4p3/3pN1k1/3P4/4P1P1/r4PKP/8 w - - 1 30      cp   208
       1      357     -116                7r/1pb2ppp/1Pr1pk2/1R1p4/3P4/4PN2/5PPP/1R4K1 w - - 1 26      cp   572
      -8      507     -180                4k2r/1R3pp1/4p3/p2pN1bp/3P4/4P3/2r2PPP/1R4K1 w - - 2 27      cp   744
       8      389     -118               7r/1pbk1Npp/1p2p3/1R1p4/3P4/4P3/P1r2PPP/1R4K1 b - - 0 24      cp   173
      -3      307     -107            1b2k2r/1p4pp/1p2p3/PR1p1p2/3P4/4PN2/2r2PPP/1R4K1 w - - 1 25      cp   408
      -1      344     -116                r7/2b4p/3kpp2/p2p2P1/3P4/2r1PN2/1R3PP1/1R4K1 b - - 0 28      cp    64
      -7     -834      154                        6k1/p4p1p/P5p1/8/1p3q2/1P5P/1N4K1/Q7 b - - 1 35      cp    38
       6    -1134      124                             1Q6/6pk/8/P7/1Pq2p1p/5P2/3K1P2/8 b - - 4 6      cp     0
       0     1072     -111                               8/Q7/2k3p1/P7/8/5P1p/4KP2/6q1 w - - 9 11      cp     0
       0      254     -190                      1r6/4q3/2PpBk1b/8/5pQ1/1p1P4/1P2KP2/3N4 w - - 2 6      cp   220
       6      509     -189                   1r6/4q2Q/2Pp1k2/4p3/5p2/1p1P2P1/1P2KP2/3N4 w - - 3 6      cp   417
      -3      332     -255                  2r5/2q3k1/3p3b/4pQ2/4Bp2/1p1P2P1/1P2KP2/3N4 b - - 3 4      cp     0
       3      245     -179                     2r5/6k1/3p1q2/3Bp1b1/4Q3/1p1PN3/1P2KP2/8 w - - 0 8      cp     0
       9     -343      274                       5r2/3Q4/2Pp3k/4p1q1/8/1BNP4/1P2KP2/2b5 w - - 7 8      cp   358
      -6      759     -305                         7r/2q5/3p1k2/3Bp3/8/1p1PP1Q1/1P2K3/8 w - - 0 9      cp     0
       7      452     -327                  1r6/2q3k1/2Pp3b/4pQ2/2B5/1p1P2p1/1P2KP2/3N4 b - - 3 4      cp     0
      -5      298     -124                         1r6/2k3q1/2PpQ3/4p3/8/1B1P4/1P2KP2/8 w - - 1 9      cp  -102
      -5     -399      291                      5r1k/2P5/3p2Qb/3Bp3/5q2/1p1PN3/1P2KP2/8 b - - 1 6      cp  6534
       9     -413      328                      5r1k/2PQ4/3p1q1b/3Bp3/8/1p1PN3/1P2KP2/8 w - - 3 9      cp  1871
       2      485     -186                        1r5k/2q5/3p4/3Bp3/6Q1/1p1P4/1P3P2/5K2 b - - 1 8      cp   135
       4      258     -186                  4r3/2q3k1/3p3b/3Bp2Q/5p2/1p1P2P1/1P2KP2/3N4 b - - 3 4      cp    94
       5      813     -447                    r4n2/pp3k2/2p1p2p/4b3/2P1rPQP/8/P4P1K/6R1 w - - 0 7      cp  1262

[SapphireBrand]

If the mg/eg and search values are from the same perspective (e.g., both side-to-move, or both from the perspective of white) then this sample has more search results whose sign agrees with the mg value. That is, I count mg as the better predictor by a 42-19 margin, with 10 draws.

Otherwise let me know and I can redo the formula to match.

That seems like suggestive evidence.

Also curious: how did you select this sample?

Thanks

[joost.van...@gmail.com]

values are 'raw', unfortunately, so search is side-to-move, where mg/eg & eval are white perspective.

the sample was based on instrumenting evaluate.cpp with code to write eval, mg, eg, and running the standard bench. From these results, I just picked every Nth configuration which had an -10 < eval < 10 and some other criterium for eg mg (i.e. both different enough from zero). It is of course easy to generate more of these.