Upload experiments/RESULTS_FINAL.md with huggingface_hub

Upload experiments/synth_train_results.json with huggingface_hub

Upload experiments/RESULTS_SUMMARY.md with huggingface_hub

Upload experiments/idea6_run.log with huggingface_hub

Upload experiments/idea4_run.log with huggingface_hub

Upload experiments/idea3_v3b.log with huggingface_hub

Upload experiments/idea3_v3.log with huggingface_hub

Upload experiments/idea3_v2e.log with huggingface_hub

Upload experiments/idea3_v2d.log with huggingface_hub

Upload experiments/idea3_v2c.log with huggingface_hub

Upload experiments/idea3_v2b.log with huggingface_hub

Upload experiments/idea3_v2.log with huggingface_hub

Upload experiments/idea3_t50.log with huggingface_hub

Upload experiments/idea3_t400.log with huggingface_hub

Upload experiments/idea3_t100.log with huggingface_hub

Upload experiments/idea3_run.log with huggingface_hub

Upload experiments/idea2_run.log with huggingface_hub

Upload experiments/idea1_tau0.5.log with huggingface_hub

Upload experiments/idea3_t0_sweep.md with huggingface_hub

Upload experiments/idea3_sweep/idea2_run.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_t100.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_t50.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_t400.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v3b.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v3.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v2e.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v2d.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v2c.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v2b.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_v2.log with huggingface_hub

Upload experiments/idea3_sweep/idea3_run.log with huggingface_hub

Few-shot optimization closeout: prototype+combo heads; WM gain confirmed K<=5; honest K>=10 tree recovery; docs+logs updated

Full autonomous run: 8 proposals with real results (few-shot WM wins K<=10; zero-shot via multi-LoD 0.67; noisy/cross-LoD solved by baseline) + all code, results, derived data

Few-shot baseline landscape: off-the-shelf methods cap at F1~0.48@K5, 0.53@K10 (ceiling 0.95) -> real headroom for a WM method

Headroom diagnostic: few-shot has huge headroom (F1 0.41@K5 -> 0.95 full); long-tail/clean-noisy near-solved

M2: reproduce official 3dSAGER matching (Bagging F1=0.946); add crawler snapshot