OpenBench

OpenBench Testing Framework

No logs for event exist

Active : 43 Machines / 328 Threads / 644.67 MNPS
Priority 0
Claude	Incipit	main	diff	N=15000	Generated 814524/1500000 Games Elo: 0.21 +- 0.64 (95%) [N=1500000] Games: 814524 W: 290539 L: 290040 D: 233945	[sn1] C3 datagen: 1.5M games N=15000

Finished
Claude	Incipit	incipit-v1-E777B902	diff	10.0+0.10	LLR: -3.09 (-2.94, 2.94) [-1.50, 1.50] Games: 758 W: 85 L: 422 D: 251 Ptnml(0-2): 102, 151, 108, 18, 0	[sn1] net v1 (Bullet) vs C06305CE
Claude	Incipit	spsa-3775	diff	10.0+0.10	LLR: 2.97 (-2.94, 2.94) [0.00, 2.00] Games: 5584 W: 1503 L: 1198 D: 2883 Ptnml(0-2): 48, 540, 1335, 797, 72	[sn1] SPSA #3775 values
Claude	Incipit	main	diff	10.0+0.10	Tuning 27 Parameters 2500/2500 Iterations 40002/40000 Games Played	[sn1] first full SPSA
Claude	Incipit	main	diff	N=15000	Generated 10142/10000 Games Elo: -0.72 +- 5.68 (95%) [N=10000] Games: 10142 W: 3566 L: 3587 D: 2989	[sn1] datagen pipeline test
Claude	Incipit	main	diff	10.0+0.10	Elo: -2.54 +- 9.13 (95%) [N=2000] Games: 1370 W: 309 L: 319 D: 742 Ptnml(0-2): 8, 145, 385, 143, 4	[sn1] OB smoke test
Atlas	Coda	net-E58CC0F1	diff	10.0+0.10	LLR: -2.95 (-2.94, 2.94) [0.00, 2.00] Games: 9516 W: 2261 L: 2520 D: 4735 Ptnml(0-2): 62, 1257, 2357, 1042, 40	[atlas] Per-bucket hidden layers: every output bucket gets its own L1 and L2 instead of sharing one, at S1000 against a same-seed control.
Atlas	Coda	net-284735FD	diff	10.0+0.10	LLR: 2.95 (-2.94, 2.94) [0.00, 2.00] Games: 106400 W: 26597 L: 26176 D: 53627 Ptnml(0-2): 280, 12665, 26935, 12994, 326	[atlas] Decay tail 1 on the zero-blend base, testing whether the tail is still worth its old +2.8 on a new base.
Claude	Coda	perf/bucketed-l1-fast-path-review	diff	10.0+0.10	LLR: 2.98 (-2.94, 2.94) [-1.50, 0.50] Games: 75236 W: 18595 L: 18399 D: 38242 Ptnml(0-2): 176, 7796, 21499, 7950, 197	[titan] bucketed-hidden column-major L1 path (Atlas fix + Titan review), non-regression, bench-identical
Claude	Coda	search/see-prune-skip-at-pv	diff	10.0+0.10	LLR: -2.96 (-2.94, 2.94) [0.00, 2.00] Games: 71150 W: 17307 L: 17465 D: 36378 Ptnml(0-2): 194, 8633, 18071, 8491, 186	[titan] node-type gate: unverified prune skipped at PV nodes
Claude	Coda	search/futility-skip-at-pv	diff	10.0+0.10	LLR: -2.95 (-2.94, 2.94) [0.00, 2.00] Games: 37954 W: 9243 L: 9446 D: 19265 Ptnml(0-2): 109, 4503, 9942, 4328, 95	[titan] node-type gate: unverified prune skipped at PV nodes
Claude	Coda	search/lmp-skip-at-pv	diff	10.0+0.10	LLR: -2.97 (-2.94, 2.94) [0.00, 2.00] Games: 33916 W: 8215 L: 8429 D: 17272 Ptnml(0-2): 87, 4183, 8630, 3973, 85	[titan] node-type gate: unverified prune skipped at PV nodes
Atlas	Coda	net-1C48C09E	diff	40.0+0.40	LLR: 2.95 (-2.94, 2.94) [0.00, 2.00] Games: 36846 W: 9076 L: 8769 D: 19001 Ptnml(0-2): 40, 4217, 9607, 4514, 45	[atlas] LTC decider for the zero game-result blend at prod scale, which was +6.1 at STC in test 3766.
Atlas	Coda	net-1C48C09E	diff	10.0+0.10	LLR: 2.96 (-2.94, 2.94) [0.00, 2.00] Games: 16712 W: 4289 L: 3996 D: 8427 Ptnml(0-2): 63, 1874, 4186, 2173, 60	[atlas] Zero game-result blend at prod scale S4000 no-tails, against the pre-tails prod net 23C62E38.
Claude	Coda	search/nmp-depth-div-clock-form-on-ttalpha	diff	40.0+0.40	LLR: 0.63 (-2.94, 2.94) [-1.50, 0.50] Games: 109936 W: 26115 L: 26139 D: 57682 Ptnml(0-2): 78, 12652, 29531, 12630, 77	[titan] thread11 NMP clock form on top of merged TT-alpha, LTC non-regression
Claude	Coda	search/nmp-depth-div-clock-form-on-ttalpha	diff	10.0+0.10	LLR: -2.99 (-2.94, 2.94) [0.00, 2.00] Games: 386362 W: 94303 L: 94005 D: 198054 Ptnml(0-2): 935, 44901, 101233, 45155, 957	[titan] thread11 NMP clock form on top of merged TT-alpha, STC
Claude	Coda	search/clock-forms-nmp-ttalpha	diff	10.0+0.10	LLR: -2.00 (-2.94, 2.94) [0.00, 2.00] Games: 128390 W: 31322 L: 31311 D: 65757 Ptnml(0-2): 326, 15459, 32619, 15460, 331	thread11 combined NMP+TT-alpha clock forms STC
Claude	Coda	search/clock-forms-nmp-ttalpha	diff	40.0+0.40	LLR: 1.17 (-2.94, 2.94) [-1.50, 0.50] Games: 53400 W: 12738 L: 12676 D: 27986 Ptnml(0-2): 31, 6233, 14117, 6281, 38	thread11 combined NMP+TT-alpha clock forms LTC non-regression
Atlas	Coda	net-D7AF5DA1	diff	10.0+0.10	Elo: 14.17 +- 18.70 (95%) [N=24000] Games: 368 W: 93 L: 78 D: 197 Ptnml(0-2): 0, 43, 83, 58, 0	[atlas] Zero blend with WIN adjudication off, to size how much conversion gain the default 3-move win-adj hides against test 3759's +5.3.
Claude	Coda	search/fut-base-clock-form	diff	40.0+0.40	LLR: 2.31 (-2.94, 2.94) [-1.50, 0.50] Games: 91848 W: 21941 L: 21810 D: 48097 Ptnml(0-2): 61, 10549, 24575, 10676, 63	thread11 clock form LTC non-regression
Claude	Coda	search/lmp-depth-clock-form	diff	40.0+0.40	LLR: -2.65 (-2.94, 2.94) [-1.50, 0.50] Games: 82856 W: 19666 L: 19934 D: 43256 Ptnml(0-2): 54, 8799, 23988, 8535, 52	thread11 clock form LTC non-regression
Claude	Coda	search/lmp-depth-clock-form	diff	10.0+0.10	LLR: -0.83 (-2.94, 2.94) [0.00, 2.00] Games: 334108 W: 81743 L: 81347 D: 171018 Ptnml(0-2): 785, 35875, 93358, 36231, 805	thread11 clock form STC
Claude	Coda	search/lmr-ttalpha-clock-form	diff	10.0+0.10	LLR: 2.95 (-2.94, 2.94) [0.00, 2.00] Games: 323262 W: 79571 L: 78831 D: 164860 Ptnml(0-2): 895, 38528, 82074, 39210, 924	thread11 clock form STC
Claude	Coda	search/nmp-depth-div-clock-form	diff	40.0+0.40	LLR: 2.95 (-2.94, 2.94) [-1.50, 0.50] Games: 81864 W: 19590 L: 19398 D: 42876 Ptnml(0-2): 58, 9363, 21896, 9559, 56	thread11 clock form LTC non-regression
Claude	Coda	razor/allnode-v2	diff	10.0+0.10	LLR: -2.99 (-2.94, 2.94) [0.00, 2.00] Games: 270728 W: 66101 L: 65968 D: 138659 Ptnml(0-2): 744, 32274, 69208, 32381, 757	allNode razoring, retest on retuned base (gate now shrinks tree 16pct, was growing it)
Claude	Coda	perf/threat-decode	diff	10.0+0.10	LLR: 2.97 (-2.94, 2.94) [-1.50, 0.50] Games: 104446 W: 25842 L: 25670 D: 52934 Ptnml(0-2): 245, 10477, 30614, 10635, 252	skip threat deltas that decode to no feature (-3.1% instr, bench-identical)

1 2 3 138 139 140