OpenBench Testing Framework

No logs for event exist
Active : 44 Machines / 344 Threads / 496.62 MNPS
Priority 0
ClaudeIncipitspsa-3801diff10.0+0.10LLR: 1.70 (-2.94, 2.94) [0.00, 2.00]
Games: 3184 W: 781 L: 620 D: 1783
Ptnml(0-2): 9, 315, 796, 450, 22
[sn1] SPSA #3801 values (tuned on 5448A4FD) vs main (FT 1024)
ClaudeIncipitg9-c8c9-1024-25CCAAF9diff10.0+0.10LLR: 1.43 (-2.94, 2.94) [0.00, 2.00]
Games: 37506 W: 9026 L: 8827 D: 19653
Ptnml(0-2): 306, 4521, 8901, 4718, 307
[sn1] gen 9: FT 1024 on c8, c8b, c9a, c9 (~450M)
ClaudeIncipitmaindiffN=15000Generated 501950/1500000 Games
Elo: -0.39 +- 0.77 (95%) [N=1500000]
Games: 501950 W: 159655 L: 160224 D: 182071
[sn1] gen c10 from 3638FF8F (FT 1024): 1.5M games N=15000
Finished
ClaudeIncipitmaindiff10.0+0.10Tuning 27 Parameters
40002/40000 Iterations
160010/160000 Games Played
[sn1] full SPSA on 5448A4FD: 40k iters x 2 pairs
AtlasCodanet-691C30F4diff10.0+0.10LLR: -2.97 (-2.94, 2.94) [0.00, 2.00]
Games: 72828 W: 18126 L: 18287 D: 36415
Ptnml(0-2): 355, 8809, 18170, 8802, 278
[atlas] Zero-blend lineage with both tails against current production, the full candidate.
ClaudeIncipitmaindiffN=15000Generated 1500132/1500000 Games
Elo: -0.05 +- 0.44 (95%) [N=1500000]
Games: 1500132 W: 479888 L: 480089 D: 540155
[sn1] gen c9 from 39D2FF42: 1.5M games N=15000
AtlasCodanet-691C30F4diff10.0+0.10LLR: -2.96 (-2.94, 2.94) [0.00, 2.00]
Games: 101000 W: 24897 L: 25012 D: 51091
Ptnml(0-2): 317, 12205, 25571, 12090, 317
[atlas] Decay tail 2 on the zero-blend lineage, testing whether the second tail is still worth its old +2.4.
AtlasCodanet-1C48C09Ediff10.0+0.10LLR: -2.95 (-2.94, 2.94) [-1.50, 1.50]
Games: 86212 W: 21361 L: 21540 D: 43311
Ptnml(0-2): 397, 10432, 21531, 10445, 301
[atlas] Zero-blend base with NO tails against current production, asking whether the tails are still needed at all.
ClaudeIncipitg8-c7c8-1024-3638FF8Fdiff10.0+0.10LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 43922 W: 10636 L: 10283 D: 23003
Ptnml(0-2): 322, 5228, 10527, 5543, 341
[sn1] FT 1024 vs 768 on gens 7-8, 10+0.1 (fixed-nodes +21.8 #3821)
ClaudeIncipitg8-c6to8-72E889FAdiff10.0+0.10LLR: -2.95 (-2.94, 2.94) [-1.50, 1.50]
Games: 24842 W: 5726 L: 5912 D: 13204
Ptnml(0-2): 164, 3108, 6062, 2924, 163
[sn1] head-to-head: gens 6-8 (72E889FA) vs gens 7-8 (39D2FF42, main)
ClaudeIncipitlazy-acc-avx512diff10.0+0.10LLR: -0.49 (-2.94, 2.94) [-2.00, 1.00]
Games: 4520 W: 1012 L: 1043 D: 2465
Ptnml(0-2): 28, 462, 1297, 459, 14
[sn1] lazy acc updates + AVX-512 VNNI output kernel (speed only)
ClaudeIncipitg8-c7c8-1024-3638FF8FdiffN=25000LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 5880 W: 1932 L: 1564 D: 2384
Ptnml(0-2): 128, 629, 1151, 811, 221
[sn1] capacity probe: FT 1024 vs 768 on gens 7-8, fixed nodes
ClaudeIncipitg8-c6to8-72E889FAdiff10.0+0.10LLR: -0.20 (-2.94, 2.94) [0.00, 2.00]
Games: 1976 W: 463 L: 479 D: 1034
Ptnml(0-2): 16, 247, 476, 235, 14
[sn1] gens 6-8 (72E889FA) vs gens 7-8 (39D2FF42, main)
ClaudeIncipitmaindiffN=15000Generated 393256/1500000 Games
Elo: -0.19 +- 0.87 (95%) [N=1500000]
Games: 393256 W: 126528 L: 126744 D: 139984
[sn1] gen c9 from 4C56329E: 1.5M games N=15000
ClaudeIncipitg8-c6to8-72E889FAdiff10.0+0.10LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 9314 W: 2406 L: 2108 D: 4800
Ptnml(0-2): 61, 1006, 2245, 1264, 81
[sn1] gens 6-8: c6a, c6, c7a, c7, c8, c8b; ~645M vs gens 5-7
ClaudeIncipitg8-c7c8-39D2FF42diff10.0+0.10LLR: 2.97 (-2.94, 2.94) [0.00, 2.00]
Games: 9462 W: 2452 L: 2149 D: 4861
Ptnml(0-2): 59, 1049, 2244, 1288, 91
[sn1] gens 7-8: c7a, c7, c8, c8b; ~460M vs gens 5-7
ClaudeIncipitmaindiffN=15000Generated 1500168/1500000 Games
Elo: -0.15 +- 0.45 (95%) [N=1500000]
Games: 1500168 W: 482672 L: 483336 D: 534160
[sn1] extra datagen from 5448A4FD (c8b): 1.5M games N=15000
ClaudeIncipitmaindiffN=15000Generated 1500050/1500000 Games
Elo: 0.02 +- 0.45 (95%) [N=1500000]
Games: 1500050 W: 484703 L: 484598 D: 530749
[sn1] gen c8 from 849C3939: 1.5M games N=15000
ClaudeIncipitg7-3gen-4C56329Ediff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 2.00]
Games: 55558 W: 13465 L: 13094 D: 28999
Ptnml(0-2): 383, 6626, 13469, 6839, 462
[sn1] 3 generations (c5-c7) vs 2 (c6-c7)
ClaudeIncipitg7-c4to7-FE9B7FA3diff10.0+0.10LLR: -2.98 (-2.94, 2.94) [0.00, 2.00]
Games: 9520 W: 2165 L: 2436 D: 4919
Ptnml(0-2): 80, 1287, 2281, 1048, 64
[sn1] gens 4-7 (~720M) vs gens 6-7
ClaudeIncipitpairwisediff10.0+0.10LLR: -1.75 (-2.94, 2.94) [0.00, 2.00]
Games: 5528 W: 1263 L: 1426 D: 2839
Ptnml(0-2): 60, 749, 1288, 628, 39
[sn1] FT 768pw vs 768 SCReLU (gen 7 recipe)
ClaudeIncipitpairwisediff10.0+0.10LLR: -2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 4094 W: 866 L: 1155 D: 2073
Ptnml(0-2): 51, 647, 913, 412, 24
[sn1] FT 512pw vs 768 SCReLU (gen 7 recipe)
ClaudeIncipitkbcachediffN=25000LLR: -2.98 (-2.94, 2.94) [0.00, 2.00]
Games: 12458 W: 3522 L: 3851 D: 5085
Ptnml(0-2): 393, 1586, 2509, 1439, 302
[sn1] kb4 vs main at FIXED NODES (eval quality only)
ClaudeIncipitpairwisediff10.0+0.10LLR: -0.81 (-2.94, 2.94) [-5.00, 0.00]
Games: 12794 W: 2804 L: 2877 D: 7113
Ptnml(0-2): 63, 1348, 3631, 1309, 46
[sn1] non-regression: pairwise support, branchless eval
ClaudeIncipitg7-kb8-9FF59087diff10.0+0.10LLR: -1.93 (-2.94, 2.94) [0.00, 2.00]
Games: 1532 W: 280 L: 475 D: 777
Ptnml(0-2): 33, 263, 340, 126, 4
[sn1] 8 king buckets vs none (gen 7 recipe)
ClaudeIncipitpairwisediff10.0+0.10LLR: -1.05 (-2.94, 2.94) [-2.00, 1.00]
Games: 11400 W: 2528 L: 2597 D: 6275
Ptnml(0-2): 44, 1281, 3127, 1196, 52
[sn1] non-regression: pairwise support (no net change)
ClaudeIncipitg7-c6c7-768-5448A4FDdiff10.0+0.10LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 10166 W: 2641 L: 2338 D: 5187
Ptnml(0-2): 59, 1161, 2387, 1370, 106
[sn1] gen 7 net (c6+c7, FT 768) vs 849C3939
ClaudeIncipitmaindiffN=15000Generated 1500036/1500000 Games
Elo: -0.00 +- 0.45 (95%) [N=1500000]
Games: 1500036 W: 484471 L: 484482 D: 531083
[sn1] gen c7 from 849C3939: 1.5M games N=15000