Auto-sync: 2026-06-28 22:57:58
Browse files
results/paper_analysis.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"best_clean_key": "residual_k4_consensus_noopbonus003",
|
| 3 |
-
"generated_utc": "2026-06-29T02:
|
| 4 |
"mechanism_gap": {
|
| 5 |
"best_clean_vs_direct_same_ckpt": 0.06956521739130428,
|
| 6 |
"best_clean_vs_h16": 0.05507246376811592,
|
|
@@ -965,6 +965,16 @@
|
|
| 965 |
"source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_noopbonus0p035_summary.json",
|
| 966 |
"std_success": 0.01281933007970784
|
| 967 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 968 |
"residual_k4_consensus_noopbonus005": {
|
| 969 |
"ci95_success": 0.03271496576240179,
|
| 970 |
"label": "K4 mean-by-type tangent consensus, no-op bonus 0.05",
|
|
|
|
| 1 |
{
|
| 2 |
"best_clean_key": "residual_k4_consensus_noopbonus003",
|
| 3 |
+
"generated_utc": "2026-06-29T02:53:17+00:00",
|
| 4 |
"mechanism_gap": {
|
| 5 |
"best_clean_vs_direct_same_ckpt": 0.06956521739130428,
|
| 6 |
"best_clean_vs_h16": 0.05507246376811592,
|
|
|
|
| 965 |
"source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_noopbonus0p035_summary.json",
|
| 966 |
"std_success": 0.01281933007970784
|
| 967 |
},
|
| 968 |
+
"residual_k4_consensus_noopbonus003_srcprog050": {
|
| 969 |
+
"label": "K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.50",
|
| 970 |
+
"missing": true,
|
| 971 |
+
"source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_srcprog0p50_summary.json"
|
| 972 |
+
},
|
| 973 |
+
"residual_k4_consensus_noopbonus003_srcprog075": {
|
| 974 |
+
"label": "K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.75",
|
| 975 |
+
"missing": true,
|
| 976 |
+
"source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_srcprog0p75_summary.json"
|
| 977 |
+
},
|
| 978 |
"residual_k4_consensus_noopbonus005": {
|
| 979 |
"ci95_success": 0.03271496576240179,
|
| 980 |
"label": "K4 mean-by-type tangent consensus, no-op bonus 0.05",
|
results/paper_analysis.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
# Paper Analysis
|
| 2 |
|
| 3 |
-
Generated: `2026-06-29T02:
|
| 4 |
|
| 5 |
## Main Seed Statistics
|
| 6 |
|
|
@@ -23,6 +23,8 @@ Generated: `2026-06-29T02:40:20+00:00`
|
|
| 23 |
| residual_k4_fieldsoftmax_grid_margin000_noopbonus003 | K4 field-softmax tangent transport, margin 0.00, no-op bonus 0.03 | 3 | 34.84% +/- 0.70 | +/- 1.75 | 56.57% | 0.417 | +5.10 pp |
|
| 24 |
| residual_k8_fieldsoftmax_grid_noopbonus003 | K8 field-softmax tangent transport, scales 0.35/0.40/0.45, no-op bonus 0.03 | 3 | 34.84% +/- 1.35 | +/- 3.36 | 56.55% | 0.397 | +5.10 pp |
|
| 25 |
| residual_k4_consensus_noopbonus003 | K4 mean-by-type tangent consensus, no-op bonus 0.03 | 3 | 35.25% +/- 1.28 | +/- 3.18 | 56.68% | 0.395 | +5.51 pp |
|
|
|
|
|
|
|
| 26 |
| residual_taskrelative_k4_consensus_noopbonus003 | K4 task-relative tangent consensus, no-op bonus 0.03 | 3 | 34.43% +/- 1.22 | +/- 3.02 | 56.19% | 0.399 | +4.70 pp |
|
| 27 |
| residual_k4_consensus_noopbonus001 | K4 mean-by-type tangent consensus, no-op bonus 0.01 | 3 | 35.19% +/- 1.32 | +/- 3.27 | 56.63% | 0.395 | +5.45 pp |
|
| 28 |
| residual_k4_consensus_noopbonus002 | K4 mean-by-type tangent consensus, no-op bonus 0.02 | 3 | 35.19% +/- 1.32 | +/- 3.27 | 56.64% | 0.395 | +5.45 pp |
|
|
|
|
| 1 |
# Paper Analysis
|
| 2 |
|
| 3 |
+
Generated: `2026-06-29T02:53:17+00:00`
|
| 4 |
|
| 5 |
## Main Seed Statistics
|
| 6 |
|
|
|
|
| 23 |
| residual_k4_fieldsoftmax_grid_margin000_noopbonus003 | K4 field-softmax tangent transport, margin 0.00, no-op bonus 0.03 | 3 | 34.84% +/- 0.70 | +/- 1.75 | 56.57% | 0.417 | +5.10 pp |
|
| 24 |
| residual_k8_fieldsoftmax_grid_noopbonus003 | K8 field-softmax tangent transport, scales 0.35/0.40/0.45, no-op bonus 0.03 | 3 | 34.84% +/- 1.35 | +/- 3.36 | 56.55% | 0.397 | +5.10 pp |
|
| 25 |
| residual_k4_consensus_noopbonus003 | K4 mean-by-type tangent consensus, no-op bonus 0.03 | 3 | 35.25% +/- 1.28 | +/- 3.18 | 56.68% | 0.395 | +5.51 pp |
|
| 26 |
+
| residual_k4_consensus_noopbonus003_srcprog050 | K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.50 | 0 | missing | missing | missing | missing | missing |
|
| 27 |
+
| residual_k4_consensus_noopbonus003_srcprog075 | K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.75 | 0 | missing | missing | missing | missing | missing |
|
| 28 |
| residual_taskrelative_k4_consensus_noopbonus003 | K4 task-relative tangent consensus, no-op bonus 0.03 | 3 | 34.43% +/- 1.22 | +/- 3.02 | 56.19% | 0.399 | +4.70 pp |
|
| 29 |
| residual_k4_consensus_noopbonus001 | K4 mean-by-type tangent consensus, no-op bonus 0.01 | 3 | 35.19% +/- 1.32 | +/- 3.27 | 56.63% | 0.395 | +5.45 pp |
|
| 30 |
| residual_k4_consensus_noopbonus002 | K4 mean-by-type tangent consensus, no-op bonus 0.02 | 3 | 35.19% +/- 1.32 | +/- 3.27 | 56.64% | 0.395 | +5.45 pp |
|
results/paper_table_status.json
CHANGED
|
@@ -840,6 +840,44 @@
|
|
| 840 |
"best_config": null,
|
| 841 |
"gain_vs_h16_policy": 0.055072463768115976
|
| 842 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 843 |
{
|
| 844 |
"key": "retrieval_residual_taskrelative_k4_mean_noopbonus003",
|
| 845 |
"label": "K4 task-relative mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.03",
|
|
|
|
| 840 |
"best_config": null,
|
| 841 |
"gain_vs_h16_policy": 0.055072463768115976
|
| 842 |
},
|
| 843 |
+
{
|
| 844 |
+
"key": "retrieval_residual_k4_mean_noopbonus003_srcprog050",
|
| 845 |
+
"label": "K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op bonus 0.03, source progress >= 0.50",
|
| 846 |
+
"path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_srcprog0p50_summary.json",
|
| 847 |
+
"clean_deployment": "yes",
|
| 848 |
+
"same_state_proposals": "no",
|
| 849 |
+
"expert_proposal": "no",
|
| 850 |
+
"story_role": "train-source viability gate for sparse residual transport",
|
| 851 |
+
"fallback_success": null,
|
| 852 |
+
"pending_job": "14894211/14894212",
|
| 853 |
+
"path_exists": false,
|
| 854 |
+
"status": "pending",
|
| 855 |
+
"success": null,
|
| 856 |
+
"std_success": null,
|
| 857 |
+
"completed_seeds": null,
|
| 858 |
+
"num_completed": null,
|
| 859 |
+
"best_config": null,
|
| 860 |
+
"gain_vs_h16_policy": null
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"key": "retrieval_residual_k4_mean_noopbonus003_srcprog075",
|
| 864 |
+
"label": "K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op bonus 0.03, source progress >= 0.75",
|
| 865 |
+
"path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_srcprog0p75_summary.json",
|
| 866 |
+
"clean_deployment": "yes",
|
| 867 |
+
"same_state_proposals": "no",
|
| 868 |
+
"expert_proposal": "no",
|
| 869 |
+
"story_role": "strict train-source viability gate for sparse residual transport",
|
| 870 |
+
"fallback_success": null,
|
| 871 |
+
"pending_job": "14894213/14894214",
|
| 872 |
+
"path_exists": false,
|
| 873 |
+
"status": "pending",
|
| 874 |
+
"success": null,
|
| 875 |
+
"std_success": null,
|
| 876 |
+
"completed_seeds": null,
|
| 877 |
+
"num_completed": null,
|
| 878 |
+
"best_config": null,
|
| 879 |
+
"gain_vs_h16_policy": null
|
| 880 |
+
},
|
| 881 |
{
|
| 882 |
"key": "retrieval_residual_taskrelative_k4_mean_noopbonus003",
|
| 883 |
"label": "K4 task-relative mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.03",
|
results/paper_table_status.md
CHANGED
|
@@ -47,6 +47,8 @@ Baseline h=16 policy: 29.74%
|
|
| 47 |
| retrieval_residual_k4_fieldsoftmax_grid_margin000_noopbonus003 | K4 field-softmax residual transport, safe residuals, scales 0.35/0.40/0.45, margin 0.00, no-op residual bonus 0.03 | complete | 34.84% | +5.10 pp | yes | no | no | field-conditioned tangent transport no-abstention diagnostic |
|
| 48 |
| retrieval_residual_k8_fieldsoftmax_grid_noopbonus003 | K8 field-softmax residual transport, safe residuals, scales 0.35/0.40/0.45, margin 0.20, no-op residual bonus 0.03 | complete | 34.84% | +5.10 pp | yes | no | no | field-conditioned tangent transport neighborhood scaling |
|
| 49 |
| retrieval_residual_k4_mean_noopbonus003 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.03 | complete | 35.25% | +5.51 pp | yes | no | no | current best clean typed sparse-intervention prior |
|
|
|
|
|
|
|
| 50 |
| retrieval_residual_taskrelative_k4_mean_noopbonus003 | K4 task-relative mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.03 | complete | 34.43% | +4.70 pp | yes | no | no | task-relative state metric for typed sparse-intervention prior |
|
| 51 |
| retrieval_residual_k4_mean_noopbonus001 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.01 | complete | 35.19% | +5.45 pp | yes | no | no | typed sparse-intervention prior fine sweep |
|
| 52 |
| retrieval_residual_k4_mean_noopbonus002 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.02 | complete | 35.19% | +5.45 pp | yes | no | no | typed sparse-intervention prior fine sweep |
|
|
|
|
| 47 |
| retrieval_residual_k4_fieldsoftmax_grid_margin000_noopbonus003 | K4 field-softmax residual transport, safe residuals, scales 0.35/0.40/0.45, margin 0.00, no-op residual bonus 0.03 | complete | 34.84% | +5.10 pp | yes | no | no | field-conditioned tangent transport no-abstention diagnostic |
|
| 48 |
| retrieval_residual_k8_fieldsoftmax_grid_noopbonus003 | K8 field-softmax residual transport, safe residuals, scales 0.35/0.40/0.45, margin 0.20, no-op residual bonus 0.03 | complete | 34.84% | +5.10 pp | yes | no | no | field-conditioned tangent transport neighborhood scaling |
|
| 49 |
| retrieval_residual_k4_mean_noopbonus003 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.03 | complete | 35.25% | +5.51 pp | yes | no | no | current best clean typed sparse-intervention prior |
|
| 50 |
+
| retrieval_residual_k4_mean_noopbonus003_srcprog050 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op bonus 0.03, source progress >= 0.50 | pending 14894211/14894212 | pending | pending | yes | no | no | train-source viability gate for sparse residual transport |
|
| 51 |
+
| retrieval_residual_k4_mean_noopbonus003_srcprog075 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op bonus 0.03, source progress >= 0.75 | pending 14894213/14894214 | pending | pending | yes | no | no | strict train-source viability gate for sparse residual transport |
|
| 52 |
| retrieval_residual_taskrelative_k4_mean_noopbonus003 | K4 task-relative mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.03 | complete | 34.43% | +4.70 pp | yes | no | no | task-relative state metric for typed sparse-intervention prior |
|
| 53 |
| retrieval_residual_k4_mean_noopbonus001 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.01 | complete | 35.19% | +5.45 pp | yes | no | no | typed sparse-intervention prior fine sweep |
|
| 54 |
| retrieval_residual_k4_mean_noopbonus002 | K4 mean-by-type residual retrieval, scale 0.40, margin 0.20, no-op residual bonus 0.02 | complete | 35.19% | +5.45 pp | yes | no | no | typed sparse-intervention prior fine sweep |
|
scripts/build_paper_analysis.py
CHANGED
|
@@ -155,6 +155,22 @@ METHODS = [
|
|
| 155 |
"k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_summary.json"
|
| 156 |
),
|
| 157 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 158 |
MethodSpec(
|
| 159 |
key="residual_taskrelative_k4_consensus_noopbonus003",
|
| 160 |
label="K4 task-relative tangent consensus, no-op bonus 0.03",
|
|
|
|
| 155 |
"k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_summary.json"
|
| 156 |
),
|
| 157 |
),
|
| 158 |
+
MethodSpec(
|
| 159 |
+
key="residual_k4_consensus_noopbonus003_srcprog050",
|
| 160 |
+
label="K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.50",
|
| 161 |
+
summary_path=(
|
| 162 |
+
"h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
|
| 163 |
+
"k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_srcprog0p50_summary.json"
|
| 164 |
+
),
|
| 165 |
+
),
|
| 166 |
+
MethodSpec(
|
| 167 |
+
key="residual_k4_consensus_noopbonus003_srcprog075",
|
| 168 |
+
label="K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.75",
|
| 169 |
+
summary_path=(
|
| 170 |
+
"h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
|
| 171 |
+
"k4s040_safe_margin0p20_mean_by_type_noopbonus0p03_srcprog0p75_summary.json"
|
| 172 |
+
),
|
| 173 |
+
),
|
| 174 |
MethodSpec(
|
| 175 |
key="residual_taskrelative_k4_consensus_noopbonus003",
|
| 176 |
label="K4 task-relative tangent consensus, no-op bonus 0.03",
|
scripts/build_paper_table_status.py
CHANGED
|
@@ -463,7 +463,7 @@ SPECS = [
|
|
| 463 |
same_state_proposals="no",
|
| 464 |
expert_proposal="no",
|
| 465 |
story_role="train-source viability gate for sparse residual transport",
|
| 466 |
-
pending_job="
|
| 467 |
),
|
| 468 |
ResultSpec(
|
| 469 |
key="retrieval_residual_k4_mean_noopbonus003_srcprog075",
|
|
@@ -473,7 +473,7 @@ SPECS = [
|
|
| 473 |
same_state_proposals="no",
|
| 474 |
expert_proposal="no",
|
| 475 |
story_role="strict train-source viability gate for sparse residual transport",
|
| 476 |
-
pending_job="
|
| 477 |
),
|
| 478 |
ResultSpec(
|
| 479 |
key="retrieval_residual_taskrelative_k4_mean_noopbonus003",
|
|
|
|
| 463 |
same_state_proposals="no",
|
| 464 |
expert_proposal="no",
|
| 465 |
story_role="train-source viability gate for sparse residual transport",
|
| 466 |
+
pending_job="14894211/14894212",
|
| 467 |
),
|
| 468 |
ResultSpec(
|
| 469 |
key="retrieval_residual_k4_mean_noopbonus003_srcprog075",
|
|
|
|
| 473 |
same_state_proposals="no",
|
| 474 |
expert_proposal="no",
|
| 475 |
story_role="strict train-source viability gate for sparse residual transport",
|
| 476 |
+
pending_job="14894213/14894214",
|
| 477 |
),
|
| 478 |
ResultSpec(
|
| 479 |
key="retrieval_residual_taskrelative_k4_mean_noopbonus003",
|