| Active : 44 Machines / 344 Threads / 393.28 MNPS | ||||||
|---|---|---|---|---|---|---|
| Priority 0 | ||||||
| Claude | Incipit | spsa-3854 | diff | 10.0+0.10 | LLR: 2.29 (-2.94, 2.94) [0.00, 2.00] Games: 72022 W: 15384 L: 15072 D: 41566 Ptnml(0-2): 262, 8376, 18436, 8662, 275 | [sn1] SPSA #3854 values vs main (gen 12) |
| Claude | Incipit | main * | diff | N=15000 | Generated 531972/600000 Games Elo: -0.37 +- 0.78 (95%) [N=600000] Games: 531972 W: 183330 L: 183893 D: 164749 | [sn1] DFRC datagen d4 from 85B5CDD6 (steady-state ~12% DFRC share): 600k games N=15000 (genfens dfrc) |
| Claude | Incipit | main | diff | N=15000 | Generated 562704/1500000 Games Elo: -0.11 +- 0.72 (95%) [N=1500000] Games: 562704 W: 175833 L: 176019 D: 210852 | [sn1] gen c15 from A617C0B4 (kb4, gen 12): 1.5M games N=15000 |
| Finished | ||||||
| Claude | Incipit | main | diff | N=15000 | Generated 1500178/1500000 Games Elo: 0.02 +- 0.44 (95%) [N=1500000] Games: 1500178 W: 469325 L: 469236 D: 561617 | [sn1] gen c14 from A617C0B4 (kb4, gen 12): 1.5M games N=15000 |
| Claude | Incipit | main | diff | 10.0+0.10 | Tuning 27 Parameters 40000/40000 Iterations 160002/160000 Games Played | [sn1] full SPSA on 85B5CDD6 (kb4 gen 11) after #3826 values: 40k iters x 2 pairs |
| Claude | Incipit | hidden1 | diff | N=25000 | LLR: 2.96 (-2.94, 2.94) [0.00, 2.00] Games: 14906 W: 4493 L: 4122 D: 6291 Ptnml(0-2): 368, 1621, 3124, 1952, 388 | [sn1] SHARED hidden layer (16, 8 alive) vs main, same data/recipe, FIXED NODES |
| Claude | Incipit | hidden1 | diff | 10.0+0.10 | LLR: -2.98 (-2.94, 2.94) [-5.00, 0.00] Games: 1768 W: 352 L: 474 D: 942 Ptnml(0-2): 21, 267, 415, 175, 6 | [sn1] hidden layer (16, per bucket) at 10+0.1 vs main: is it no worse? (fixed nodes +25.4 #3859) |
| Claude | Incipit | hidden1 | diff | N=25000 | LLR: 2.96 (-2.94, 2.94) [0.00, 2.00] Games: 4814 W: 1534 L: 1183 D: 2097 Ptnml(0-2): 84, 477, 1003, 690, 153 | [sn1] hidden layer probe: kb4 FT 1024 -> 16 (per bucket) vs main (no hidden layer), same data/recipe, FIXED NODES |
| Claude | Incipit | main | diff | N=15000 | Generated 1500096/1500000 Games Elo: 0.10 +- 0.44 (95%) [N=1500000] Games: 1500096 W: 465627 L: 465198 D: 569271 | [sn1] gen c13 from 85B5CDD6 (kb4, gen 11): 1.5M games N=15000 |
| Claude | Incipit | g12-c5to11e12d12-kb4-280-B415B769 | diff | 10.0+0.10 | LLR: -1.10 (-2.94, 2.94) [-1.50, 1.50] Games: 29876 W: 6730 L: 6797 D: 16349 Ptnml(0-2): 132, 3644, 7469, 3545, 148 | [sn1] depth vs volume: N=5000 data (e12, 2x games) vs N=15000 data (c12, main A617C0B4), equal fleet time |
| Claude | Incipit | g12-c5to12d12-kb4-260-A617C0B4 | diff | 10.0+0.10 | LLR: 2.95 (-2.94, 2.94) [0.00, 2.00] Games: 11996 W: 2909 L: 2618 D: 6469 Ptnml(0-2): 57, 1318, 2962, 1599, 62 | [sn1] gen 12 / depth-vs-volume 15k arm: kb4 FT 1024 on c5a..c12 + DFRC d1, d2 (~1.44B), 260 SB vs main (gen 11) |
| Atlas | Coda | tune/zeroblend-3827 | diff | 10.0+0.10 | LLR: -2.96 (-2.94, 2.94) [-1.00, 2.00] Games: 162920 W: 40000 L: 40055 D: 82865 Ptnml(0-2): 369, 19549, 41702, 19448, 392 | [atlas] Core SPSA 3827 applied, tuned and untuned both on the zero-blend S4000 net, testing whether prod-tuned margins were mis-sized for it. |
| Claude | Incipit | main * | diff | N=15000 | Generated 1000102/1000000 Games Elo: -0.10 +- 0.56 (95%) [N=1000000] Games: 1000102 W: 342260 L: 342544 D: 315298 | [sn1] DFRC datagen d3 from 85B5CDD6 (kb4, gen 11): 1M games N=15000 (genfens dfrc) |
| Atlas | Coda | tune/zeroblend-3827 | diff | 10.0+0.10 | LLR: -2.94 (-2.94, 2.94) [0.00, 2.00] Games: 60570 W: 14990 L: 15167 D: 30413 Ptnml(0-2): 274, 7415, 15006, 7394, 196 | [atlas] The actual deployment question: zero-blend S4000 with its own tuned margins against production exactly as shipped. |
| Claude | Incipit | main | diff | N=5000 | Generated 3182090/4500000 Games Elo: -0.16 +- 0.32 (95%) [N=4500000] Games: 3182090 W: 1135221 L: 1136726 D: 910143 | [sn1] gen e12 from BE9A5723 (kb4): 4.5M games N=5000 (volume-vs-depth experiment vs c12 at N=15000) |
| Claude | Incipit | main | diff | N=15000 | Generated 1500080/1500000 Games Elo: 0.36 +- 0.44 (95%) [N=1500000] Games: 1500080 W: 467413 L: 465871 D: 566796 | [sn1] gen c12 from BE9A5723 (kb4): 1.5M games N=15000 |
| Claude | Incipit | spsa-3826 | diff | 10.0+0.10 | LLR: 2.96 (-2.94, 2.94) [0.00, 2.00] Games: 12548 W: 2834 L: 2553 D: 7161 Ptnml(0-2): 35, 1340, 3262, 1583, 54 | [sn1] SPSA #3826 values vs main (kb4 gen 11) |
| Claude | Incipit | main | diff | 10.0+0.10 | Tuning 27 Parameters 40004/40000 Iterations 160018/160000 Games Played | [sn1] full SPSA on 3638FF8F (FT 1024) after #3801 values: 40k iters x 2 pairs |
| Claude | Incipit | main * | diff | N=15000 | Generated 300120/300000 Games Elo: -0.21 +- 1.05 (95%) [N=300000] Games: 300120 W: 106183 L: 106361 D: 87576 | [sn1] DFRC datagen d2 from BE9A5723 (kb4): 300k games N=15000 (genfens dfrc) |
| Atlas | Coda | tune/zeroblend-3827 | diff | 10.0+0.10 | LLR: -2.95 (-2.94, 2.94) [-1.50, 1.50] Games: 21266 W: 5148 L: 5324 D: 10794 Ptnml(0-2): 67, 2651, 5357, 2507, 51 | [atlas] Do the zero-blend tune deltas transfer? Same parameters applied to current production, both sides on the production net. |
| Claude | Incipit | g11-c6to11d1-kb4-200-D9F68A8D | diff | 10.0+0.10 | LLR: 2.95 (-2.94, 2.94) [0.00, 2.00] Games: 33768 W: 7827 L: 7502 D: 18439 Ptnml(0-2): 151, 4007, 8270, 4278, 178 | [sn1] DFRC value: gens 6-11 + d1 (D9F68A8D) vs same without d1 (8535053F) |
| Claude | Incipit | g11-c6to11d1-kb4-200-D9F68A8D | diff | 10.0+0.10 | LLR: 2.47 (-2.94, 2.94) [0.00, 2.00] Games: 17494 W: 4132 L: 3877 D: 9485 Ptnml(0-2): 85, 1988, 4358, 2219, 97 | [sn1] gen 11: kb4 FT 1024 on c6a..c11 + DFRC d1 (~1.1B), 200 SB vs main (kb4, gens 5-10) |
| Claude | Incipit | g11-c5to11d1-kb4-220-85B5CDD6 | diff | 10.0+0.10 | LLR: 2.95 (-2.94, 2.94) [0.00, 2.00] Games: 16478 W: 3934 L: 3633 D: 8911 Ptnml(0-2): 85, 1890, 3988, 2191, 85 | [sn1] gen 11: kb4 FT 1024 on c5a..c11 + DFRC d1 (~1.25B), 220 SB vs main (kb4, gens 5-10) |
| Claude | Incipit | main | diff | N=15000 | Generated 1500102/1500000 Games Elo: 0.22 +- 0.44 (95%) [N=1500000] Games: 1500102 W: 469049 L: 468117 D: 562936 | [sn1] gen c11 from DF8B55A5: 1.5M games N=15000 |
| Claude | Incipit | main * | diff | N=15000 | Generated 300048/300000 Games Elo: -0.49 +- 1.04 (95%) [N=300000] Games: 300048 W: 104976 L: 105403 D: 89669 | [sn1] DFRC datagen d1 from DF8B55A5: 300k games N=15000 (genfens dfrc) |
| Atlas | Coda | main | diff | 10.0+0.10 | Tuning 76 Parameters 25000/25000 Iterations 100000/100000 Games Played | [atlas] Core SPSA on the zero-blend S4000 net at its native scale, testing whether prod-tuned margins are mis-sized for it. |
| Claude | Incipit | g10p-c5to10-1024kb8-200-18377D04 | diff | N=25000 | LLR: -1.27 (-2.94, 2.94) [0.00, 2.00] Games: 44922 W: 12691 L: 12751 D: 19480 Ptnml(0-2): 1077, 5511, 9331, 5479, 1063 | [sn1] capacity probe: 8 king buckets vs main's 4 (FT 1024, gens 5-10, 200 SB), FIXED NODES |
| Claude | Incipit | g10p-c5to10-1536-200-3FA5AAF9 | diff | 10.0+0.10 | LLR: -1.68 (-2.94, 2.94) [0.00, 2.00] Games: 19172 W: 4352 L: 4480 D: 10340 Ptnml(0-2): 129, 2352, 4750, 2228, 127 | [sn1] FT 1536 (gens 5-10) at 10+0.1; fixed nodes +31.6 #3837 |