Axiom BOINC Session Log — s0303ar/as (~10:05-10:19 UTC, March 3, 2026) ===================================================================== PART 1: VALIDATION & CREDIT (Session s0303ar, ~10:05 UTC) ========================================================= Credit Awarded: 317 results, 1,681.2 total credit, 6 users, 13 hosts ---------------------------------------------------------------------- | User | Credit | Hosts | |-----------------------|---------|------------------------------------------------------------| | ChelseaOilman | +595.2 | Foxtrot-3, Dell-9520, Dell-XPS, Delta-2 | | Steve Dodd | +557.6 | DadOld-PC, Dads-PC, Dad-Workstation | | WTBroughton | +257.9 | achernar, sirius, capella | | zombie67 [MM] | +159.6 | rose | | Anandbhat | +79.7 | DESKTOP-11MAEMP | | marmot | +31.2 | XYLENA | Total: 1,681.2 credit (well under 10K safety cap) Counters updated: credited_count.txt: 7,791 -> 8,108 (+317) total_results.txt: 173,152 Cleanup: System Healthy ---------------------------------------------------------------------- - 0 stuck tasks (>12h running on dead hosts) - 0 tasks >48 hours - 0 broken experiments — all active experiments healthy - No cleanup actions required - All findings 100% consistent with prior data KEY SCIENTIFIC FINDINGS — THIS SESSION ---------------------------------------------------------------------- 1. Finding #62 (Extended Width): 386+ -> 447+ valid seeds — BREAKTHROUGH! +61 wdextwidth seeds. w96=100%, w128=100%, w256=100%, w512=100%. ALL WIDTHS 100%! Combined with #59: inverse CP at ALL widths w24-w512. 2. Finding #52 (Percolation): 576+ -> 631+ seeds. +55 percolation seeds. ER p_c robustly 0.337-0.343 (vs theory 0.25). ROBUST. 3. Finding #60 (Task Generality): 426+ -> 464+ seeds — STRONGLY CONFIRMED! +38 wdtasksweep seeds. compositional=100%, spiral=100%, teacher-student=~95%, hidden_manifold=~86%. TASK-GENERAL! 4. Finding #59 (Width Transition): 439+ -> 476+ seeds — BEYOND DEFINITIVE. +37 wdwidthtrans seeds. ALL widths w24-w64 at 100%. 5. Finding #49 (WD x LR): 438+ -> 469+ seeds. +31 wdlr seeds. LR schedule changes HOW MUCH, not WHICH timing is optimal. 6. Finding #51 (SVD Rank Intervention): 849+ -> 870+ seeds. +21 svdrank seeds. Late SVD truncation CONSISTENTLY reduces gen gap. P1 (late SVD improves gap) SUPPORTED across all widths 32/64/128. STRONG CAUSAL EVIDENCE. 7. Finding #56 (WD x Optimizer): 216+ -> 234+ seeds. +18 wdoptim seeds. Growing, optimizer interaction study. 8. Finding #53 (MP Universality GPU): 83+ -> 95+ results. +12 mpuniv_gpu seeds. Departure at epoch 0 consistent. H1 NOT supported. 9. Finding #50 (WD Window): 288+ -> 300+ seeds. +12 wdwindow seeds. Diminishing returns SUPPORTED. 10. Finding #61 (Depth Generality): 616+ -> 625+ seeds — DEFINITIVELY CONFIRMED. +9 wddepth seeds. d2=100%, d3=98.3%, d4=82.9%, d5=75.1%, d6=77.3%. 11. Finding #54 (WD Timing Scale GPU): 69+ -> 78+ seeds, ALL FALSE. +9 wdtimscale_gpu seeds. Hyperparam artifact confirmed (WD=0.001 too weak). 12. Finding #47 (Rep Crystallization): 191+ -> 196+ seeds. +5 repcrystal seeds. P3 (crystallization before optimal WD onset): 100%. Timing confirmed; rank-at-onset NOT the mechanism. All findings continue 100% consistent with prior data. No surprises. PART 2: RESEARCH & DEPLOYMENT (Session s0303as, ~10:12 UTC) ============================================================ NO NEW EXPERIMENTS DESIGNED — weight rebalancing only. Mature experiments reduced in priority; queue replenished after depletion. Deployment Summary: 650 CPU WUs + 14 GPU (still unsent, skipped), 0 errors ---------------------------------------------------------------------- CPU Experiments (650 WUs, all untargeted BOINC-scheduled): | Experiment | WUs | Finding | Wt | Seeds | Status | |--------------------------------|------|---------|-----|--------|--------------------------------------| | wd_timing_extended_width | 108 | #62 | 2 | 447+ | 100% all widths — top priority | | wd_optimizer_interaction | 162 | #56 | 3 | 234+ | Growing, most scientifically active | | svd_rank_intervention | 54 | #51 | 1 | 870+ | Causal mechanism, very mature | | percolation_scaling | 54 | #52 | 1 | 631+ | Cross-disciplinary, robust | | wd_timing_depth | 54 | #61 | 1 | 625+ | Definitively confirmed | | wd_timing_task_sweep | 54 | #60 | 1 | 464+ | Strongly confirmed | | wd_lr_interaction | 54 | #49 | 1 | 469+ | LR schedule interaction | | wd_window_duration | 54 | #50 | 1 | 300+ | Window duration | | wd_width_transition | 56 | #59 | 1 | 476+ | Beyond definitive | Not deployed (weight 0): repcrystal, regmech, wdrebound, bottmech. GPU Experiments (14 WUs still unsent from prior batch, skipped this session): | Experiment | WUs | Finding | Notes | |--------------------------------|-----|---------|------------------------------------------- | | wd_timing_scale_gpu | 6 | #54 | 78+ seeds, ALL FALSE — hyperparam artifact | | mp_universality_test | 8 | #53 | 95+ results, H1 NOT supported | GPU Checkpoint: Active GPU hosts: 6 confirmed - Pyhelix (h1, PyHelix): RTX 3080 Laptop - thonon-meylan (h249, dthonon): RTX 4060 Ti - Dell-XPS-15-9560 (h319, ChelseaOilman): GTX 1050 x2 - W10-Home (h57, Henk Haneveld): GPU - dbgrensenh27 (h9, [DPC] hansR): Quadro M4000M - DESKTOP-P57624Q (h29, kotenok2000): GPU GPU workunits this session: 14 (6 wdtimscale_gpu + 8 mpuniv_gpu) — still unsent GPU scripts active: wd_timing_scale_gpu.py, mp_universality_test.py Retired GPU experiments: grad_subspace_wd_gpu.py (7 seeds, COMPLETE NEGATIVE) Known GPU issues: vega (h348) SSL cert failure, gigas (h322) SSL cert failure Weight Adjustments (from s0303as): ---------------------------------------------------------------------- wdextwidth REDUCED to weight 2 (was 3) — 447+ valid seeds, all 100% svdrank REDUCED to weight 1 (was 2) — 870+ seeds, very mature percolation REDUCED to weight 1 (was 2) — 631+ seeds, very mature wdoptim KEPT at weight 3 — still most scientifically interesting wddepth/wdtasksweep/wdlr/wdwindow/wdwidthtrans KEPT at weight 1 Immediate Uptake: ---------------------------------------------------------------------- 99+ WUs grabbed within seconds after deployment: achernar (h159, WTBroughton): first to connect Dell-9520 (h320, ChelseaOilman): fast uptake Foxtrot-3 (h340, ChelseaOilman): fast uptake Active unsent pool: 748 total after deployment Skipped hosts: Latitude (h63, 3.9GB), Athlon (h118, 3.5GB), vega (h348, SSL broken) Novelty Check: No new experiments designed this session. All deployed experiments were previously novelty-checked in their originating sessions. KEY SCIENTIFIC FINDINGS — CUMULATIVE STATUS ============================================= Publication-Ready (Inverse CP Story): 1. Finding #44 (Inverse CP for WD): 34+ seeds, 100% at w32. BASELINE. 2. Finding #45 (Reg Timing Universality): 140 seeds. WD+Dropout. COMPLETE. 3. Finding #59 (Width w24-w64): 476+ seeds, 100% all widths. DEFINITIVE. 4. Finding #60 (Task Generality): 464+ seeds, 4 tasks. STRONG. 5. Finding #61 (Depth d2-d6): 625+ seeds. DEFINITIVE. 6. Finding #62 (Extended Width w96-w512): 447+ valid seeds, 100% all widths. BREAKTHROUGH. 7. Finding #51 (SVD Rank — Causal): 870+ seeds. Mechanism confirmed. STRONG. *** INVERSE CP IS NOW A GENERAL PHENOMENON *** Width-general: 100% at w24-w512 (#59 + #62, 923+ combined seeds) Task-general: 100%/100%/~95%/~86% across 4 tasks (#60, 464+ seeds) Depth-general: 100% d2-d3, 75-83% d4-d6 (#61, 625+ seeds) Growing / Active: - Finding #49 (WD x LR): 469+ seeds. Interaction confirmed. - Finding #50 (WD Window): 300+ seeds. Diminishing returns supported. - Finding #52 (Percolation): 631+ seeds. ER p_c robustly 0.337-0.343. - Finding #53 (MP Universality GPU): 95+ results. H1 NOT supported. - Finding #54 (WD Timing Scale GPU): 78+ seeds, ALL FALSE (hyperparam artifact). - Finding #56 (WD x Optimizer): 234+ seeds. - Finding #47 (Rep Crystallization): 196+ seeds. Timing confirmed. - Finding #46 (WD Rebound): Preliminary negative. Complete/Retired: - Finding #55 (Grad Subspace): 7 seeds, COMPLETE NEGATIVE. - Finding #57 (Width Mechanism): 83 seeds, INCONCLUSIVE — RETIRED. - Finding #58 (Critical Width): 50 seeds, NEGATIVE — RETIRED. KNOWN ISSUES ============= - vega (h348): SSL cert failure — ALL results are errors. Deployment SKIPS this host. - gigas (h322): SSL cert failure — GPU error results (low priority). - Andre-WEBK (h345): Mixed success rate (outcome=3/exit=202, host-specific). - GPU CUDA arch mismatch on some hosts (Blackwell sm_120+). NEXT SESSION PRIORITIES ======================== 1. Continue wd_timing_extended_width (#62) — 447+ seeds, all 100%, keep growing 2. Continue wd_optimizer_interaction (#56) — 234+ seeds, most scientifically active 3. Continue svd_rank_intervention (#51) — 870+ seeds, causal mechanism 4. Continue percolation_scaling (#52) — 631+ seeds, cross-disciplinary 5. Monitor GPU experiments (#54 robustness, #53 continuing) 6. Consider new experiment: inverse CP at w256-2048 with WD=0.05 (GPU) 7. Monitor weight adjustments — mature experiments continue trending toward weight 1 Session duration: ~10:05-10:19 UTC, March 3, 2026 Parts completed: Validation (s0303ar) + Research/Deployment (s0303as) + Save (this log) Cumulative credited: 8,108 results | 173,152 total results in database