round-10: physics-factorial + direct-2048 chain (scripts/hf_gpu_round10_job.py)

round-10: physics-factorial + direct-2048 chain (scripts/evaluate_frozen_benchmarks.py)

round-8 fix: base64 chain cmd (HF mangles embedded quote mixes)

round-7 follow-through: disjoint_256 arm (scripts/hf_gpu_ab_multiseed_job.py)

round-7 follow-through: disjoint_256 arm (scripts/evaluate_frozen_benchmarks.py)

round-7: disjoint encoder pipeline (scripts/hf_disjoint_encoder_job.py)

round-7: disjoint encoder pipeline (scripts/train_disjoint_encoder.py)

round-7: morgan_2048_rp parity arm (scripts/evaluate_frozen_benchmarks.py)

Factorial channel ablation: --physics-mode (gamma_only/headspace_only) in train.py + evaluator matching + factorial job script

Factorial channel ablation: --physics-mode (gamma_only/headspace_only) in train.py + evaluator matching + factorial job script

Upload scripts/hf_idea3_drydown_physics_job.py with huggingface_hub

Upload scripts/evaluate_drydown_physics.py with huggingface_hub

Upload scripts/hf_e1_physics_gamma_job.py with huggingface_hub

Upload scripts/evaluate_substantivity.py with huggingface_hub

Upload scripts/physics_redundancy_probe.py with huggingface_hub

Upload scripts/augment_gamma_channel.py with huggingface_hub

t4-medium flavor

multiseed job: resume + self-chaining

review-response: upload scripts/hf_gpu_ab_multiseed_job.py

review-response: upload scripts/multiseed_tost_analysis.py

review-response: upload scripts/openpom_overlap_audit.py

review-response: upload scripts/split_leakage_analysis.py

review-response: upload scripts/evaluate_frozen_benchmarks.py

Upload scripts/hf_gpu_ab_job.py with huggingface_hub

Upload scripts/hf_gpu_ab_job.py with huggingface_hub

Upload scripts/run_ab_job.py with huggingface_hub

feat: 3-arm GPU job on corrected labels — morgan/openpom_256/pom_alltags (v2)

feat: evaluators support pom_alltags arm (objective_dim-aware, prospective N/A for 575-dim)

fix: regenerate degenerate pyramid_targets as fold(objective_targets); add all-tags targets

fix: paper_data exporter prefers GPU readout; stamp val_total into GPU evals; stats bundle

feat: GPU readout figures + stats scripts + val-total macros

feat: HF stats-collector job — consolidated paper_stats_bundle.json

feat: paper figures from existing artifacts (ablation bars, POM UMAP, trajectory)

feat: paper_data.yaml exporter for autopaper (49 data_macros)

feat: v11.6 repair/enrichment scripts + independent-candidate prep + integrity tests

feat: low-evidence gating in substitution_v11 + wisemoor description provenance

fix: GPU job install step — don't shadow PyPI with torch index-url

feat: expand substitution-triplet benchmark 8 -> 20

feat: HF GPU two-arm A/B training job

feat: preregistered frozen-benchmark evaluator (substitution triplets + prospective)

feat: --structural-source flag in eval_v6 for arm-matched evaluation

feat: gate flips READY_FOR_TRAINING (all 5 checks pass)

feat: full molequles multi-tag target space (575-dim ablation arm)

feat: prospective formula benchmark (40 in-silico formulas, blinded labels)

feat: complete representation ablation + discordant substitution triplets

feat: link molequles material-profiles odor text into training descriptor matrix

feat: freeze representation validation gate

release: freeze blinded genre challenge v2

feat: add leakage-resistant genre baseline benchmark

v11.5 powered re-run: effective n=14→28, retriever=0/28 [0.000,0.121], physchem=0/28 [0.000,0.121], random=0/28 [0.000,0.121], verdict=underpowered