Axiom BOINC Session Log — s0303z/aa (~14:00-14:17 UTC, March 3, 2026) ===================================================================== PART 1: VALIDATION & CREDIT (Session s0303z, ~14:00 UTC) ========================================================= Credit Awarded: 685 results, ~1,827 credit, 13 hosts, 9 users ---------------------------------------------------------------------- | User | Credit | Notes | |-------------------|--------|-----------------------------------------------| | Steve Dodd | +898 | Top earner — h87 Dad-Workstation, h123 Dads-PC | | WTBroughton | +598 | h159 achernar, h347 sirius | | kotenok2000 | +172 | h29 DESKTOP-P57624Q | | Amapola | ~60 | h287 DESKTOP-N5RAJSE | | Anandbhat | ~40 | h222 DESKTOP-11MAEMP | | ChelseaOilman | ~30 | h319 Dell-XPS | | [DPC] hansR | ~15 | h9 dbgrensenh27 | | PyHelix | ~10 | h1 Pyhelix | | Vato | ~4 | h6 iand-r7-5800h | Total: ~1,827 credit (well under 10K safety cap) Counters updated: credited_count.txt: 4021 -> 4706 (+685) total_results.txt: -> 161,112 Cleanup: System Healthy ---------------------------------------------------------------------- - 0 stuck tasks - 0 tasks >48 hours - 0 broken experiments — all active scripts healthy - No experiments needed aborting (all failure rates <20%) Host Notes: - vega (h348): SSL certificate issue — ALL 166 results are instant CERTIFICATE_VERIFY_FAILED errors producing zero science. Needs ca-certificates installed or Python certifi fix. Deployment skipping this host until resolved. KEY SCIENCE ADVANCEMENTS FROM VALIDATED RESULTS ---------------------------------------------------------------------- *** Finding #61 (Depth Generality): 66 -> 181 seeds — NOW DEFINITIVE! *** - d2=100% | d3=98.3% | d4=82.9% | d5=75.1% | d6=77.3% - Inverse CP generalizes across ALL tested depths (d2-d6). - Massive +115 new seeds this session. Previously strongly confirmed, now DEFINITIVE. *** Finding #59 (Width Generality): 182 -> 265 seeds — BEYOND ANY DOUBT *** - 100% inverse CP at ALL widths w24-w64. - +83 new seeds this session. Paradigm shift beyond definitive. *** Finding #60 (Task Generality): 100 -> 111 seeds *** - compositional=100% | spiral=100% | teacher-student=~95% | hidden_manifold=~86% - +11 new seeds. Inverse CP confirmed across 4 distinct tasks. *** Finding #51 (SVD Rank Intervention): 293+ -> 398+ seeds *** - +105 new seeds. Causal mechanism test growing — late SVD truncation reliably reduces generalization gap across widths 32/64/128. *** Finding #52 (Percolation): 199+ -> 243+ seeds *** - +44 new seeds. ER p_c robustly 0.337-0.343 across all seeds. *** Finding #49 (WD x LR Interaction): 183+ -> 231+ seeds *** - +48 new seeds. LR schedule changes HOW MUCH WD timing matters, but NOT which timing is optimal. *** Finding #62 (Extended Width): 0 valid seeds *** - 36 completed results ALL from vega — ALL SSL certificate errors. - 72 in-progress on sirius + Dad-Workstation should deliver first real data soon. PART 2: RESEARCH & DEPLOYMENT (Session s0303aa, ~14:10 UTC) ============================================================ NO NEW EXPERIMENTS DESIGNED — priority reweighting only. Finding #62 (wd_timing_extended_width) was designed last session (s0303y) and is being boosted to top priority since it still has 0 valid seeds. Deployment Summary: 663 total WUs (649 CPU + 14 GPU), 0 errors ---------------------------------------------------------------------- CPU Experiments (649 WUs, all untargeted BOINC-scheduled): | Experiment | WUs | Finding | Wt | Seeds | Status | |--------------------------------|------|---------|-----|-------|---------------------------------| | wd_timing_extended_width | 212 | #62 | 30 | 0 | BOOSTED — publication-critical! | | wd_timing_depth | 127 | #61 | 18 | 181 | Definitively confirmed | | svd_rank_intervention | 71 | #51 | 10 | 398+ | Very mature | | wd_timing_task_sweep | 57 | #60 | 8 | 111 | Strongly confirmed | | percolation_scaling | 49 | #52 | 7 | 243+ | Mature | | wd_lr_interaction | 35 | #49 | 5 | 231+ | Mature | | wd_width_transition | 28 | #59 | 4 | 265 | Beyond definitive | | wd_window_duration | 28 | #50 | 4 | 143+ | Mature | | wd_optimizer_interaction | 21 | #56 | 3 | 70+ | Growing | | representation_crystallization | 21 | #47 | 3 | 93+ | Growing | Note: regmech/wdrebound/bottmech skipped — 3,480 orphaned WUs already in DB from prior sessions. GPU Experiments (14 WUs, all untargeted BOINC-scheduled): | Experiment | WUs | Finding | Notes | |--------------------------------|-----|---------|------------------------------------------- | | mp_universality_test | 8 | #53 | 40+ results, H1 NOT supported | | wd_timing_scale_gpu | 6 | #54 | 37+ seeds, ALL FALSE — hyperparam artifact | 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 (wd_timing_scale_gpu + mp_universality_test) 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 issue: vega (h348) SSL cert failure, gigas (h322) SSL cert failure Key Deployment Decisions: ---------------------------------------------------------------------- - Finding #62 (wdextwidth) BOOSTED to weight 30 (top priority) — still 0 valid seeds after vega's SSL failures consumed 36 WUs. 72 in-progress on sirius + Dad-Workstation should deliver first real data soon. - All queue-depleted high-priority experiments replenished - Mature experiments (wdwidthtrans 265 seeds, wdlr 231+, wdwindow 143+) reduced proportionally to make room for wdextwidth boost - Skipped vega (h348) — SSL broken, wastes WUs - Skipped regmech/wdrebound/bottmech (weight 0, 3,480 orphaned WUs in DB) Dispatch Status: ---------------------------------------------------------------------- 154 dispatched to hosts immediately: Dad-Workstation (h87, Steve Dodd): 124 WUs grabbed DESKTOP-P57624Q (h29, kotenok2000): 30 WUs grabbed 509 unsent waiting for 13+ additional active hosts to poll 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 (100%) + Dropout (98.6%). COMPLETE. 3. Finding #59 (Width Generality): 265 seeds, 100% at w24-w64. DEFINITIVE. 4. Finding #60 (Task Generality): 111 seeds, 4 tasks (100%/100%/~95%/~86%). STRONG. 5. Finding #61 (Depth Generality): 181 seeds, d2-d6 (100%/98.3%/82.9%/75.1%/77.3%). DEFINITIVE. 6. Finding #51 (SVD Rank Intervention): 398+ seeds. Causal mechanism. STRONG. Awaiting First Valid Data: 7. Finding #62 (Extended Width): 0 valid seeds — Tests w96-w512 with WD=0.05. KEY remaining publication question: Does inverse CP hold beyond w64? 212 WUs deployed this session (boosted to top priority). Growing / Active: - Finding #49 (WD x LR): 231+ seeds. Interaction confirmed. - Finding #50 (WD Window): 143+ seeds. Diminishing returns supported. - Finding #52 (Percolation): 243+ seeds. ER p_c robustly 0.337-0.343. - Finding #53 (MP Universality GPU): 40+ results. H1 NOT supported. - Finding #54 (WD Timing Scale GPU): 37+ seeds, ALL FALSE (hyperparam artifact). - Finding #56 (WD x Optimizer): 70+ seeds. Consistent with #54. - Finding #47 (Rep Crystallization): 93+ results. 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. PUBLICATION STATUS — INVERSE CP NOW APPROACHING SUBMISSION ============================================================ The inverse critical period for weight decay story now has THREE DEFINITIVE and ONE STRONG generalization: 1. WIDTH (#59): 100% at w24-w64, 265 seeds — DEFINITIVE 2. DEPTH (#61): 100% d2-d3, 75-83% d4-d6, 181 seeds — DEFINITIVE (was 66 last session!) 3. TASK (#60): 100%/100%/~95%/~86% across 4 tasks, 111 seeds — STRONG 4. EXTENDED WIDTH (#62): w96-w512, WD=0.05 — 0 valid seeds, awaiting first data Key remaining question: Does inverse CP hold at w256+ with matched WD? Finding #62 will answer this. If YES: truly universal. If NO: genuine width boundary. KNOWN ISSUES ============= - vega (h348): SSL cert failure — ALL 166 results are errors. Needs ca-certificates fix. Deployment now SKIPS this host. Volunteer notification recommended. - gigas (h322): SSL cert failure — 24 GPU error results (low priority) - Andre-WEBK (h345): Mixed success rate (outcome=3/exit=202, host-specific) - Orphaned WUs (~3,480) from prior sessions (low priority cleanup) - GPU CUDA arch mismatch on some hosts (Blackwell sm_120+) NEXT SESSION PRIORITIES ======================== 1. Analyze incoming wd_timing_extended_width (#62) results — KEY publication question 2. Continue wd_timing_depth (#61) — 181 seeds, now definitively confirmed 3. Continue SVD rank intervention (#51) — 398+ seeds, causal mechanism 4. Monitor vega (h348) — notify volunteer about SSL cert fix 5. Monitor GPU experiments (#54 robustness, #53 continuing) Session duration: ~14:00-14:17 UTC, March 3, 2026 Parts completed: Validation (s0303z) + Research/Deployment (s0303aa) + Save (this log) Cumulative credited: 4,706 results | 161,112 total results in database