FuryBench

Active : 13 Machines / 554 Threads / 402.29 MNPS
Priority 0
Caissa Witek	stats-per-thread-counters	smp diff	2.0+0.02	ELO 0.65 +- 3.69 (95%) LLR 1.08 (-2.94, 2.94) [-4.00, 0.00] GAMES N: 10212 W: 2607 L: 2588 D: 5017 PENTA [67, 1245, 2470, 1250, 74]
Integral Fury	hmc_inputs_and_scaling	diff	N=20000	ELO 1.07 +- 1.51 (95%) LLR 0.32 (-2.25, 2.89) [0.00, 3.00] GAMES N: 86720 W: 25874 L: 25606 D: 35240 PENTA [2035, 10180, 18609, 10554, 1982]
Quanticade DarkNeutrino	master 25K node datagen.	diff	N=25000	ELO -0.02 +- 0.24 (95%) GAMES N: 5606682/30000000 W: 1967690 L: 1967989 D: 1671003
Caissa Witek	pv-pruning-margins Search: PV-node margins for history, futility and SEE pruning	diff	10.0+0.10	ELO 0.29 +- 0.94 (95%) LLR -1.00 (-2.94, 2.94) [0.00, 2.00] GAMES N: 139828 W: 32974 L: 32856 D: 73998 PENTA [465, 16652, 35559, 16776, 462]
Caissa Witek	hist-bonus-overflow	diff	10.0+0.10	ELO -2.75 +- 2.97 (95%) LLR -1.33 (-2.94, 2.94) [-3.00, 0.00] GAMES N: 14036 W: 3238 L: 3349 D: 7449 PENTA [51, 1732, 3560, 1627, 48]
Priority -1
Integral Fury	hmc_inputs Optimizations	diff	10.0+0.10	ELO 0.83 +- 2.07 (95%) LLR 0.08 (-2.25, 2.89) [0.00, 3.00] GAMES N: 28518 W: 7069 L: 7001 D: 14448 PENTA [87, 3382, 7251, 3454, 85]
Integral Fury	spsa please ascend me!	smp diff	20.0+0.20	TUNE 106 parameters ITER 5,787 / 20,000 GAMES 23,148 / 80,000
Caissa Witek	rfp-qsearch-verify	diff	10.0+0.10	ELO 0.16 +- 1.01 (95%) LLR -1.38 (-2.94, 2.94) [0.00, 2.00] GAMES N: 120220 W: 28278 L: 28223 D: 63719 PENTA [365, 14410, 30510, 14455, 370]
Caissa Witek	singular-beta-history	diff	10.0+0.10	ELO -0.07 +- 1.21 (95%) LLR -1.58 (-2.94, 2.94) [0.00, 2.00] GAMES N: 83510 W: 19673 L: 19691 D: 44146 PENTA [257, 10045, 21177, 10011, 265]
Priority -2
Illumina Mergener	neg-extensions Negative extensions	diff	40.0+0.40	ELO 3.06 +- 7.50 (95%) LLR 0.31 (-2.25, 2.89) [0.00, 5.00] GAMES N: 2044 W: 423 L: 405 D: 1216 PENTA [6, 221, 549, 241, 5]
Illumina Mergener	neg-extensions Negative extensions	diff	8.0+0.08	ELO -5.13 +- 7.89 (95%) LLR -1.08 (-2.25, 2.89) [0.00, 5.00] GAMES N: 2166 W: 465 L: 497 D: 1204 PENTA [11, 282, 533, 242, 15]
Illumina Mergener	fix-non-pawn-corrhist Fix non pawn corrhist	diff	8.0+0.08	ELO -3.95 +- 5.40 (95%) LLR -1.51 (-2.25, 2.89) [0.00, 5.00] GAMES N: 3254 W: 683 L: 720 D: 1851 PENTA [7, 321, 1010, 280, 9]

Finished
Integral Fury	hmc_inputs	diff	N=25000	ELO 0.90 +- 2.07 (95%) LLR -0.09 (-2.25, 2.89) [0.00, 3.00] GAMES N: 43298 W: 12657 L: 12545 D: 18096 PENTA [880, 5115, 9549, 5223, 882]
Integral Fury	hmc_inputs_and_scaling	diff	10.0+0.10	ELO 0.66 +- 2.23 (95%) LLR 0.16 (-2.25, 2.89) [0.00, 3.00] GAMES N: 6794 W: 1136 L: 1123 D: 4535 PENTA [1, 236, 2911, 247, 2]
Quanticade Anematode	move-lover move lover	diff	4.0+0.04	ELO 2.30 +- 1.76 (95%) LLR 2.94 (-2.25, 2.89) [0.00, 3.00] GAMES N: 39656 W: 9923 L: 9660 D: 20073 PENTA [193, 4321, 10543, 4572, 199]
Integral Fury	hmc_inputs Optimizations	diff	10.0+0.10	ELO 0.59 +- 2.56 (95%) LLR 0.08 (-2.25, 2.89) [0.00, 3.00] GAMES N: 5326 W: 904 L: 895 D: 3527 PENTA [2, 188, 2274, 197, 2]
Integral Fury	hmc_inputs Optimizations	diff	10.0+0.10	ELO 0.67 +- 1.46 (95%) LLR 0.35 (-2.25, 2.89) [0.00, 3.00] GAMES N: 19790 W: 3652 L: 3614 D: 12524 PENTA [0, 885, 8087, 923, 0]
Caissa Witek	tm-cross-thread-instability TimeManager: Cross-thread best move instability Each search thread counts how often its own root move changes between iterations. The main thread sums those counters, normalizes by thread count and folds the result into the soft time limit as one more multiplicative factor, so time allocation sees the helper threads disagreeing instead of only the main thread's own stability streak.	smp diff	5.0+0.05	ELO -5.48 +- 3.48 (95%) LLR -2.95 (-2.94, 2.94) [0.00, 3.00] GAMES N: 9516 W: 2180 L: 2330 D: 5006 PENTA [12, 1212, 2460, 1062, 12]
Integral Fury	spsa_4500	smp diff	20.0+0.20	ELO 5.50 +- 9.79 (95%) LLR 0.27 (-2.25, 2.89) [0.00, 3.00] GAMES N: 1010 W: 262 L: 246 D: 502 PENTA [0, 97, 295, 113, 0]
Caissa Witek	qs-node-delta-prune QSearch: Node-level delta pruning Bail out of a quiescence node before move generation when even winning the most valuable piece on the board (plus a queen promotion when a pawn stands on the relative 7th) cannot reach alpha.	diff	10.0+0.10	ELO -4.59 +- 2.71 (95%) LLR -2.68 (-2.94, 2.94) [0.00, 2.00] GAMES N: 15914 W: 3629 L: 3839 D: 8446 PENTA [53, 1915, 4210, 1747, 32]
Integral Fury	spsa_4500	diff	10.0+0.10	ELO -2.28 +- 3.56 (95%) LLR -1.41 (-2.25, 2.89) [0.00, 3.00] GAMES N: 9440 W: 2249 L: 2311 D: 4880 PENTA [27, 1140, 2450, 1074, 29]
Quanticade DarkNeutrino	D4D2F8C3BDC4446D 2 speedups both doing the same thing tho	diff	4.0+0.04	ELO 2.56 +- 2.67 (95%) LLR 1.50 (-2.25, 2.89) [0.00, 3.00] GAMES N: 17914 W: 4417 L: 4285 D: 9212 PENTA [101, 1990, 4655, 2098, 113]
Quanticade DarkNeutrino	ttpv-reduction search.c: Reduce more for tt_pv ahead of pruning	diff	60.0+0.60	ELO -0.70 +- 3.99 (95%) LLR -0.48 (-2.25, 2.89) [0.00, 3.00] GAMES N: 6426 W: 1490 L: 1503 D: 3433 PENTA [1, 710, 1803, 699, 0]
Quanticade DarkNeutrino	lmr-cutnode-ttmove search.c: Reduce more if this node is cutnode and we dont have TT move	diff	10.0+0.10	ELO -2.64 +- 2.91 (95%) LLR -2.26 (-2.25, 2.89) [0.00, 3.00] GAMES N: 13306 W: 3064 L: 3165 D: 7077 PENTA [20, 1592, 3529, 1493, 19]
Quanticade DarkNeutrino	multicut-corr plenty stuff (search is depressing man)	diff	10.0+0.10	ELO 0.18 +- 1.20 (95%) LLR -2.27 (-2.25, 2.89) [0.00, 3.00] GAMES N: 79832 W: 19016 L: 18975 D: 41841 PENTA [123, 9357, 20951, 9326, 159]
Quanticade DarkNeutrino	tune Tuned 4700 iters	smp diff	20.0+0.20	ELO 1.07 +- 2.69 (95%) LLR 0.28 (-2.25, 2.89) [0.00, 3.00] GAMES N: 13630 W: 3211 L: 3169 D: 7250 PENTA [0, 1424, 3928, 1460, 3]
Quanticade DarkNeutrino	master Source: Add and make use of removing lowest bit (#430)	diff	60.0+0.60	ELO 90.78 +- 7.70 (95%) GAMES N: 1816/10000 W: 698 L: 234 D: 884 PENTA [0, 50, 362, 478, 18]

1 2 3 370 371 372

FuryBench Testing Framework