FuryBench Testing Framework

home toIntegralPlentyChessObsidianCaissaQuanticadeSerendipitySpaghetSchoenemannClarityIlluminaPawnocchioVineAnura
Finished
Clarity Vast
bestmove
Get best move from PV table I expect null moves, which I'll fix later (hopefully later today)
crash diff8.0+0.08
ELO    4.35 +- 3.68 (95%)
LLR    2.95 (-2.25, 2.89) [-3.00, 1.00]
GAMES  N: 9754 W: 2332 L: 2210 D: 5212
PENTA  [55, 1104, 2444, 1212, 62]
Caissa Witek
fix-bmi2
Fix pext/pdep and MulHi64 falling back to software on non-Windows builds Also, with pext working, restore -DUSE_BMI2 in the makefile native probe. No functional change.
diff5.0+0.05
ELO    3.01 +- 1.84 (95%)
LLR    2.95 (-2.94, 2.94) [0.00, 2.00]
GAMES  N: 37274 W: 9297 L: 8974 D: 19003
PENTA  [203, 4093, 9721, 4418, 202]
Caissa Witek
avxvnni_no_bmi2
AVX-VNNI support
diff5.0+0.05
ELO    3.86 +- 2.15 (95%)
LLR    2.95 (-2.94, 2.94) [0.00, 2.00]
GAMES  N: 28568 W: 7166 L: 6849 D: 14553
PENTA  [148, 3252, 7222, 3459, 203]
Illumina Mergener
pre-v4-tune
Tuning
diff40.0+0.40
ELO    3.15 +- 2.52 (95%)
LLR    2.89 (-2.25, 2.89) [0.00, 5.00]
GAMES  N: 18394 W: 3865 L: 3698 D: 10831
PENTA  [36, 2074, 4817, 2227, 43]
Integral Fury
ti_tweaks
smp diff20.0+0.20
ELO    12.31 +- 5.43 (95%)
LLR    2.91 (-2.25, 2.89) [0.00, 4.00]
GAMES  N: 3868 W: 1029 L: 892 D: 1947
PENTA  [4, 395, 998, 534, 3]
Caissa Witek
eval-84-144B
Optimize StaticExchangeEvaluation Measured bench speedup: +0.233% No functional change.
diff10.0+0.10
ELO    1.35 +- 1.06 (95%)
LLR    2.97 (-2.94, 2.94) [0.00, 2.00]
GAMES  N: 110222 W: 26151 L: 25722 D: 58349
PENTA  [337, 13024, 28013, 13347, 390]
Caissa Witek
eval-84-144B
Optimize StaticExchangeEvaluation Measured bench speedup: +0.233% No functional change.
diff10.0+0.10
ELO    8.26 +- 3.10 (95%)
LLR    2.95 (-2.94, 2.94) [0.00, 2.00]
GAMES  N: 10860 W: 2946 L: 2688 D: 5226
PENTA  [1, 1087, 2998, 1341, 3]
Integral Fury
ti_tweaks
diff40.0+0.40
ELO    5.94 +- 2.98 (95%)
LLR    2.90 (-2.25, 2.89) [0.00, 2.50]
GAMES  N: 13274 W: 3415 L: 3188 D: 6671
PENTA  [16, 1507, 3363, 1736, 15]
Caissa Witek
see_speedup
Optimize StaticExchangeEvaluation Measured bench speedup: +0.233% No functional change.
diff4.0+0.04
ELO    0.37 +- 1.12 (95%)
LLR    2.96 (-2.94, 2.94) [-2.00, 0.00]
GAMES  N: 105178 W: 25923 L: 25810 D: 53445
PENTA  [735, 11966, 27065, 12097, 726]
Integral Fury
main
Save draw score to correction history if there's an upcoming repetition (#241)
time odds diff10.0+0.10
ELO    56.35 +- 8.81 (95%)
LLR    2.93 (-2.25, 2.89) [0.00, 2.50]
GAMES  N: 1474 W: 482 L: 245 D: 747
PENTA  [1, 75, 359, 290, 12]
Integral Fury
ti_tweaks
diffN=25000
ELO    39.42 +- 6.22 (95%)
GAMES  N: 5116/5000 W: 1840 L: 1262 D: 2014
PENTA  [71, 439, 1072, 793, 183]
Integral Fury
ti_tweaks
diff10.0+0.10
ELO    8.23 +- 3.51 (95%)
LLR    2.96 (-2.25, 2.89) [0.00, 2.50]
GAMES  N: 9376 W: 2469 L: 2247 D: 4660
PENTA  [28, 933, 2556, 1131, 40]
Caissa Witek
makefile-fix
Makefile: Fix net download and INCBIN build race Force HTTP/1.1 with retries (some hosts hit HTTP/2 PROTOCOL_ERROR on the GitHub CDN) and download via a temporary file, so a partial transfer is never mistaken for a complete net. Make Evaluate.o and the per-arch targets depend on the net file, since it's embedded at compile time via INCBIN and could otherwise be read while still downloading under make -j.
diff10.0+0.10
ELO    4.17 +- 10.80 (95%)
GAMES  N: 1000/1000 W: 240 L: 228 D: 532
PENTA  [2, 108, 271, 114, 5]
Clarity Vast
tb-filtering
TB optimal move filtering at root. No elo but looks nice in tournaments :)
crash diff8.0+0.08
ELO    1.62 +- 2.40 (95%)
LLR    2.89 (-2.25, 2.89) [-3.00, 1.00]
GAMES  N: 20766 W: 4783 L: 4686 D: 11297
PENTA  [102, 2214, 5651, 2317, 99]
Caissa Witek
move-overhead-fix
diff10.0+0.10
ELO    0.58 +- 1.60 (95%)
LLR    2.95 (-2.94, 2.94) [-3.00, 0.00]
GAMES  N: 44418 W: 10552 L: 10478 D: 23388
PENTA  [129, 4900, 12073, 4982, 125]
Caissa Witek
move-overhead-fix
diff1.0+0.00
ELO    27.92 +- 7.05 (95%)
LLR    3.00 (-2.94, 2.94) [-2.00, 0.00]
GAMES  N: 4378 W: 1381 L: 1030 D: 1967
PENTA  [88, 426, 895, 607, 173]
Illumina Mergener
output-buckets
Output buckets
diff8.0+0.08
ELO    10.21 +- 5.67 (95%)
LLR    2.91 (-2.25, 2.89) [0.00, 5.00]
GAMES  N: 4424 W: 1075 L: 945 D: 2404
PENTA  [35, 469, 1080, 587, 41]
Caissa Witek
move-overhead-fix
diff0.1+0.00
ELO    62.74 +- 9.71 (95%)
LLR    3.40 (-2.94, 2.94) [0.00, 2.00]
GAMES  N: 2396 W: 796 L: 368 D: 1232
PENTA  [23, 205, 420, 421, 129]
Caissa Witek
move-overhead-fix
diff0.2+0.00
ELO    50.26 +- 8.71 (95%)
LLR    3.32 (-2.94, 2.94) [0.00, 2.00]
GAMES  N: 2896 W: 927 L: 511 D: 1458
PENTA  [45, 241, 543, 491, 128]
Quanticade DarkNeutrino
opts
Optimizations
diff4.0+0.04
ELO    5.96 +- 3.19 (95%)
LLR    3.02 (-2.25, 2.89) [0.00, 3.00]
GAMES  N: 12070 W: 3064 L: 2857 D: 6149
PENTA  [51, 1275, 3194, 1446, 69]
Quanticade DarkNeutrino
net57
How much is it without the slowdown
diffN=5000
ELO    13.79 +- 2.53 (95%)
GAMES  N: 40102/40000 W: 13973 L: 12382 D: 13747
PENTA  [1418, 4160, 7797, 4765, 1911]
Quanticade DarkNeutrino
simp-ttpv-cutnode
search.c: Simplify ttpv && cutnode adjustment
diff10.0+0.10
ELO    1.30 +- 2.16 (95%)
LLR    2.96 (-2.94, 2.94) [-3.00, 1.00]
GAMES  N: 24928 W: 5872 L: 5779 D: 13277
PENTA  [34, 2950, 6406, 3037, 37]
Integral Fury
nnue_threat_inputs
Incremental threat updates
diffN=20000
ELO    39.13 +- 9.07 (95%)
LLR    2.94 (-2.25, 2.89) [0.00, 2.50]
GAMES  N: 2550 W: 919 L: 633 D: 998
PENTA  [43, 214, 536, 378, 104]
Integral Fury
nnue_threat_inputs
Merge branch 'main' into nnue_threat_inputs
diffN=20000
ELO    32.51 +- 8.13 (95%)
LLR    2.99 (-2.25, 2.89) [0.00, 2.50]
GAMES  N: 3076 W: 1089 L: 802 D: 1185
PENTA  [55, 272, 646, 461, 104]
Quanticade DarkNeutrino
lmr-remove-check
search.c: Remove was in check lmr
diff10.0+0.10
ELO    2.94 +- 3.00 (95%)
LLR    2.96 (-2.94, 2.94) [-3.00, 1.00]
GAMES  N: 12986 W: 3128 L: 3018 D: 6840
PENTA  [20, 1487, 3378, 1579, 29]