Axiom BOINC Session Results
Session file: results_2026-03-04_0312
Date: March 4, 2026

SOURCES
- Part 1 validation log: validate_2026-03-04_0309.txt
- Part 2 run log: run_2026-03-03_1352.log
- Findings context: findings_summary.txt

PART 1: RESULTS REVIEWED AND CREDIT AWARDED
- Reviewed uncredited completed exp_* rows at credit time: 112
- Credited this session: 112 results, 129.10 total credit (within 10,000 cap)
- Outcome mix in credited batch:
  - outcome=1: 31 rows (121.0 credit total)
  - outcome=5: 73 rows (7.3 credit total)
  - outcome=6: 8 rows (0.8 credit total)
- Per-user credit additions:
  - Orange Kid +84.40
  - makracz +21.00
  - marmot +8.00
  - _Scandinavian_ +4.10
  - kotenok2000 +4.10
  - PyHelix +4.10
  - Steve Dodd +1.40
  - dthonon +1.00
  - [DPC] hansR +0.70
  - rilian +0.20
  - Henk Haneveld +0.10

STUCK/BROKEN TASK CLEANUP
- Stuck cleanup (>12h running + dead host >6h): 0 aborted
- Hard ceiling cleanup (>48h running): 0 aborted
- Broken-prefix cleanup (unsent + in-progress only): 73 aborted total
  - abx_cycle_hgt_delay_resonance 13
  - lotka_delay_adaptive_harvest 6
  - allee_delay_climate_harvest 5
  - fisher_delay_adaptive_culling 5
  - grad_subspace_wd_gpu_gpu 7
  - heavytail_blockclip_spectrum 5
  - kuramoto_delay_noisy_control 5
  - sandpile_delay_adaptive_dissipation 5
  - wildfire_delay_adaptive_suppression 5
  - lorenz96_delay_assimilation_regime_shift 4
  - seir_delay_fatigue_control 4
  - toggle_delay_adaptive_control 4
  - tritrophic_delay_harvest_resilience 4
  - battery_pulse_delay_plating_resonance 1

PART 2: DEPLOYMENT AND RESEARCH SUMMARY
CPU deployment
- Host-targeted CPU queue fill to ~3x CPU per host
- CPU hosts seen: 81
- Hosts skipped for low RAM (<6 GB): 2
- CPU workunits created: 2937
- CPU scripts used:
  - wd_batchnoise_interaction.py
  - wd_labelsmooth_interaction.py

GPU deployment checkpoint
- Run log shows GPU deployment started with GPU scripts:
  - wd_curvature_trigger_gpu.py
  - wd_timing_scale_gpu.py
- Newest run log ended during GPU deployment (interrupted), so final created count was not printed there.
- Live checkpoint after run:
  - Active GPU-capable hosts (last 72h): 83
  - Queued GPU exp_* results (appid=2, states 1/2/4): 46
  - Queued with prefix exp_wd_curvature_trigger_gpu_gpu_h*: 7
  - Queued with prefix exp_wd_timing_scale_gpu_gpu_h*: 9

NEW/ACTIVE EXPERIMENT DESIGN NOTES + NOVELTY CHECK
- New CPU experiment script deployed in Part 2:
  - wd_batchnoise_interaction.py
  - Hypothesis: late weight decay gain interacts with gradient-noise regime (small vs large batch), not just main-effect WD timing.
- Existing active mechanism/interaction lines continued:
  - wd_labelsmooth_interaction.py
  - wd_curvature_trigger_gpu.py
  - wd_timing_scale_gpu.py
- Novelty check documentation recorded in run log via literature/web searches before deployment, including:
  - "site:arxiv.org weight decay label smoothing interaction"
  - "site:arxiv.org adaptive weight decay deep neural networks"
  - "site:arxiv.org batch size weight decay generalization"
  - "\"weight decay\" \"batch size\" \"schedule\" neural networks"
  - "arxiv 1711.05101 decoupled weight decay regularization"
- Novel angle emphasized: interaction and mechanism testing in BOINC-distributed, time-bounded repeated-trial format, rather than repeating settled main-effect WD confirmations.

KEY SCIENTIFIC FINDINGS
1. No new payload-derived scientific reversal was established in this session; the prior inverse critical-period/WD timing signal remains not reversed.
2. Validation QC still found missing upload artifacts as the dominant reliability issue (sampled 31/31 missing in the reviewed pre-credit sample), constraining direct content-level interpretation.
3. Operationally, successful compute-bearing outcomes continue to arrive (31 outcome=1 rows credited this session), indicating partial pipeline health despite artifact-loss failures.
4. Failure-heavy prefixes were actively pruned (73 tasks aborted) to reduce wasted volunteer compute and redirect capacity to novel interaction/mechanism experiments.
5. New and active deployment focus is now interaction/mechanism testing (batch-noise x WD timing, label smoothing x WD timing, and GPU curvature-trigger timing variants), which is the highest-value path for next interpretable findings.

END STATE
- Part 1 credit queue status after run: remaining uncredited completed exp_* rows = 0.
- Website counters in Part 1 were updated (credited_count and total_results_count) per validation log.