Auto-sync: 2026-06-30 04:24:31
Browse files
scripts/build_paper_analysis.py
CHANGED
|
@@ -409,6 +409,22 @@ METHODS = [
|
|
| 409 |
"k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_summary.json"
|
| 410 |
),
|
| 411 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 412 |
MethodSpec(
|
| 413 |
key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
|
| 414 |
label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.03",
|
|
|
|
| 409 |
"k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_summary.json"
|
| 410 |
),
|
| 411 |
),
|
| 412 |
+
MethodSpec(
|
| 413 |
+
key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger0005",
|
| 414 |
+
label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.005",
|
| 415 |
+
summary_path=(
|
| 416 |
+
"h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
|
| 417 |
+
"k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p005_summary.json"
|
| 418 |
+
),
|
| 419 |
+
),
|
| 420 |
+
MethodSpec(
|
| 421 |
+
key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger0015",
|
| 422 |
+
label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.015",
|
| 423 |
+
summary_path=(
|
| 424 |
+
"h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
|
| 425 |
+
"k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p015_summary.json"
|
| 426 |
+
),
|
| 427 |
+
),
|
| 428 |
MethodSpec(
|
| 429 |
key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
|
| 430 |
label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.03",
|
scripts/build_paper_table_status.py
CHANGED
|
@@ -765,6 +765,26 @@ SPECS = [
|
|
| 765 |
story_role="lower-margin sensitivity for trace-motivated singleton near-miss challenger calibration",
|
| 766 |
pending_job="14954526/14954527",
|
| 767 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 768 |
ResultSpec(
|
| 769 |
key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
|
| 770 |
label="K4 compatible residual retrieval, near-miss challenger gate 0.03",
|
|
|
|
| 765 |
story_role="lower-margin sensitivity for trace-motivated singleton near-miss challenger calibration",
|
| 766 |
pending_job="14954526/14954527",
|
| 767 |
),
|
| 768 |
+
ResultSpec(
|
| 769 |
+
key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger0005",
|
| 770 |
+
label="K4 compatible residual retrieval, near-miss challenger gate 0.005",
|
| 771 |
+
path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p005_summary.json",
|
| 772 |
+
clean_deployment="yes",
|
| 773 |
+
same_state_proposals="no",
|
| 774 |
+
expert_proposal="no",
|
| 775 |
+
story_role="fine lower-margin calibration for the trace-motivated singleton near-miss challenger gate",
|
| 776 |
+
pending_job="14954890/14954893",
|
| 777 |
+
),
|
| 778 |
+
ResultSpec(
|
| 779 |
+
key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger0015",
|
| 780 |
+
label="K4 compatible residual retrieval, near-miss challenger gate 0.015",
|
| 781 |
+
path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p015_summary.json",
|
| 782 |
+
clean_deployment="yes",
|
| 783 |
+
same_state_proposals="no",
|
| 784 |
+
expert_proposal="no",
|
| 785 |
+
story_role="fine upper-margin calibration for the trace-motivated singleton near-miss challenger gate",
|
| 786 |
+
pending_job="14954894/14954895",
|
| 787 |
+
),
|
| 788 |
ResultSpec(
|
| 789 |
key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
|
| 790 |
label="K4 compatible residual retrieval, near-miss challenger gate 0.03",
|