The most awesome fit I've ever seen in my model: all recorded games at standard time controls with both players rated 2780+, combining analysis by Komodo 10 and Stockfish 7. The point is that the fitting process is equating only the frequency of matching on the first and second moves, the matching on moves of equal value to the first move (often there are two tied-top moves or three or more), and the projected versus actual error judged by the computer chess programs. All of the other indices from 3rd-best to 30th-best are coming along "for free". The index fit line is judging each index with equal weight. Thus a low value means the model is accurately projecting the probabilities of all moves, not just the top ones. Previously, I've considered index-fit values below 0.01 to be good; this is 25 times lower. The model itself is definitely not "overfitted"---it is severely underfitted (i.e., "theory-headed"), but it must be doing something right. Obtained 9/13/16 i mDelta Swing ProjVal Sigma Actual Proj% Actual% 2sigma range z-score 1 0.00 0.0374 26742.36 82.46: 26742.00 57.17%: 57.17% 56.82%--57.53%, z = -0.00 2 0.21 0.0219 8405.08 67.38: 8405.00 18.02%: 18.02% 17.74%--18.31%, z = -0.00 3 0.33 0.0145 3975.79 52.18: 3956.00 8.59%: 8.55% 8.37%-- 8.82%, z = -0.38 4 0.40 0.0094 2278.28 41.47: 2267.00 4.98%: 4.96% 4.80%-- 5.16%, z = -0.27 5 0.46 0.0058 1438.72 33.86: 1403.00 3.18%: 3.10% 3.03%-- 3.33%, z = -1.05 6 0.51 0.0032 1005.19 28.73: 988.00 2.25%: 2.21% 2.12%-- 2.37%, z = -0.60 7 0.56 0.0005 706.70 24.37: 720.00 1.59%: 1.62% 1.48%-- 1.70%, z = +0.55 8 0.60-0.0020 523.13 21.13: 548.00 1.18%: 1.24% 1.09%-- 1.28%, z = +1.18 9 0.65-0.0040 399.00 18.62: 368.00 0.90%: 0.83% 0.82%-- 0.99%, z = -1.66 10 0.69-0.0053 309.57 16.43: 291.00 0.70%: 0.66% 0.63%-- 0.78%, z = -1.13 11 0.72-0.0075 234.78 14.40: 236.00 0.53%: 0.54% 0.47%-- 0.60%, z = +0.09 12 0.76-0.0085 176.34 12.53: 188.00 0.40%: 0.43% 0.35%-- 0.46%, z = +0.93 13 0.80-0.0096 141.40 11.27: 140.00 0.32%: 0.32% 0.27%-- 0.38%, z = -0.12 14 0.83-0.0107 108.95 9.86: 116.00 0.25%: 0.27% 0.21%-- 0.30%, z = +0.72 15 0.87-0.0111 77.99 8.42: 82.00 0.18%: 0.19% 0.14%-- 0.22%, z = +0.48 16 0.90-0.0116 64.54 7.65: 68.00 0.15%: 0.16% 0.12%-- 0.19%, z = +0.45 17 0.94-0.0118 46.67 6.52: 57.00 0.11%: 0.14% 0.08%-- 0.14%, z = +1.58 18 0.98-0.0130 34.31 5.62: 61.00 0.08%: 0.15% 0.06%-- 0.11%, z = +4.75 19 1.01-0.0131 27.12 5.00: 32.00 0.07%: 0.08% 0.04%-- 0.09%, z = +0.98 20 1.05-0.0135 18.91 4.20: 27.00 0.05%: 0.07% 0.03%-- 0.07%, z = +1.93 21 1.08-0.0136 14.40 3.63: 19.00 0.04%: 0.05% 0.02%-- 0.06%, z = +1.27 22 1.13-0.0134 13.86 3.52: 22.00 0.04%: 0.06% 0.02%-- 0.06%, z = +2.31 23 1.17-0.0128 8.61 2.81: 10.00 0.02%: 0.03% 0.01%-- 0.04%, z = +0.49 24 1.21-0.0124 6.61 2.47: 6.00 0.02%: 0.02% 0.00%-- 0.03%, z = -0.25 25 1.25-0.0117 5.20 2.18: 6.00 0.01%: 0.02% 0.00%-- 0.03%, z = +0.37 26 1.30-0.0107 2.97 1.65: 4.00 0.01%: 0.01% -0.00%-- 0.02%, z = +0.63 27 1.34-0.0100 2.35 1.48: 4.00 0.01%: 0.01% -0.00%-- 0.02%, z = +1.12 28 1.39-0.0091 1.37 1.14: 6.00 0.00%: 0.02% -0.00%-- 0.01%, z = +4.06 29 1.43-0.0079 1.31 1.08: 1.00 0.00%: 0.00% -0.00%-- 0.01%, z = -0.28 30 1.47-0.0062 0.57 0.74: 0.00 0.00%: 0.00% -0.00%-- 0.01%, z = -0.77 Index fits, x10,000: 0.0004435, wtd. 0.0004435; mass 0.0005316, wtd. 0.0005316 Name ProjVal St.Dev Actual; Proj% Actual% 2sigma range z-score ScaledFalloffW 2109.49 20.07: 2109.49 0.0451: 0.0451 0.0442--0.0460, z = +0.0000, adj +0.0000 MoveMatchWtd 26742.36 82.46: 26742.00 57.17%: 57.17% 56.82%--57.53%, z = -0.00, adj -0.00 EqualValueMatch 30071.92 79.65: 30072.00 64.29%: 64.29% 63.95%--64.63%, z = +0.00, adj +0.00 PlayedMoveMatch 23566.06 77.14: 46773.00 50.38%: 100.0% 50.05%--50.71%, z = +300.84 Selection Test ProjVal St.Dev Actual; Proj% Actual% 2sigma range z-score AdvancingMove 28570.75 65.57: 29412.00 61.08%: 62.88% 60.80%--61.36%, z = +12.83, engm% = 62.01 BishopMove 7667.94 47.18: 7616.00 16.39%: 16.28% 16.19%--16.60%, z = -1.10, engm% = 16.05 Capture 10143.15 37.58: 10827.00 21.69%: 23.15% 21.53%--21.85%, z = +18.20, engm% = 22.55 Castling 450.79 13.80: 550.00 0.96%: 1.18% 0.90%-- 1.02%, z = +7.19, engm% = 0.97 CheckingMove 2371.46 25.48: 2536.00 5.07%: 5.42% 4.96%-- 5.18%, z = +6.46, engm% = 5.42 EngineMove 26742.36 82.46: 26742.00 57.17%: 57.17% 56.82%--57.53%, z = -0.00, engm% = 100.00 EqualTopMove 30071.92 79.65: 30073.00 64.29%: 64.30% 63.95%--64.63%, z = +0.01, engm% = 100.00 Error150 118.03 9.70: 170.00 0.25%: 0.36% 0.21%-- 0.29%, z = +5.36, engm% = 0.00 Error35 1418.74 31.51: 1348.00 3.03%: 2.88% 2.90%-- 3.17%, z = -2.24, engm% = 0.00 Error70 483.85 19.03: 512.00 1.03%: 1.09% 0.95%-- 1.12%, z = +1.48, engm% = 0.00 EvalGoesToZero 6619.03 36.36: 6679.00 14.15%: 14.28% 14.00%--14.31%, z = +1.65, engm% = 14.17 KingMove 5891.69 41.51: 5859.00 12.60%: 12.53% 12.42%--12.77%, z = -0.79, engm% = 12.43 KnightMove 6739.48 43.74: 7203.00 14.41%: 15.40% 14.22%--14.60%, z = +10.60, engm% = 14.77 NonCapture 36630.44 37.57: 35946.00 78.32%: 76.85% 78.15%--78.48%, z = -18.22, engm% = 77.45 PawnMove 10123.81 56.07: 10196.00 21.64%: 21.80% 21.40%--21.88%, z = +1.29, engm% = 22.24 PlayedMove 23566.06 77.14: 46773.00 50.38%: 100.0% 50.05%--50.71%, z= +300.84, engm% = 57.17 Promotion 24.69 2.05: 26.00 0.05%: 0.06% 0.04%-- 0.06%, z = +0.64, engm% = 0.04 QueenMove 5959.01 38.09: 5748.00 12.74%: 12.29% 12.58%--12.90%, z = -5.54, engm% = 12.63 RetreatingMove 8926.67 52.90: 8461.00 19.09%: 18.09% 18.86%--19.31%, z = -8.80, engm% = 18.56 RookMove 10391.72 52.43: 10151.00 22.22%: 21.70% 21.99%--22.44%, z = -4.59, engm% = 21.87 SamePieceAsPrev 1241.64 15.27: 1328.00 2.65%: 2.84% 2.59%-- 2.72%, z = +5.66, engm% = 2.76 SidewaysMove 9276.17 55.61: 8900.00 19.83%: 19.03% 19.59%--20.07%, z = -6.76, engm% = 19.44 Note confirmation of beliefs in retreating moves being harder for human players to find and higher tendency to capture, move Knights, and move the same piece as on the previous move. However, also note from the last column that my model is also under-projecting the latter quantities relative to the computer's recommendations---this has been the main caveat all summer in being satisfied with how my upgraded model is being fitted.