Auto-sync: 2026-06-30 10:07:28
Browse files
scripts/build_paper_analysis.py
CHANGED
|
@@ -473,6 +473,30 @@ METHODS = [
|
|
| 473 |
"proposal_lattice_types6_prepend_margin0p05_summary.json"
|
| 474 |
),
|
| 475 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 476 |
MethodSpec(
|
| 477 |
key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",
|
| 478 |
label="K4 composed compatible tangents, no-op bonus 0.03, singleton near-miss bonus 0.01",
|
|
|
|
| 473 |
"proposal_lattice_types6_prepend_margin0p05_summary.json"
|
| 474 |
),
|
| 475 |
),
|
| 476 |
+
MethodSpec(
|
| 477 |
+
key="typed_proposal_lattice_types4safe_prepend_margin005",
|
| 478 |
+
label="Typed proposal lattice head, safe four families, policy-prepended margin 0.05",
|
| 479 |
+
summary_path=(
|
| 480 |
+
"h16_policy_ckpt_near_miss_policy_bc5_typedprop_p2_bestpt_"
|
| 481 |
+
"proposal_lattice_types4safe_prepend_margin0p05_summary.json"
|
| 482 |
+
),
|
| 483 |
+
),
|
| 484 |
+
MethodSpec(
|
| 485 |
+
key="typed_proposal_lattice_types2sparse_prepend_margin005",
|
| 486 |
+
label="Typed proposal lattice head, sparse no-op/wrong-gripper families, policy-prepended margin 0.05",
|
| 487 |
+
summary_path=(
|
| 488 |
+
"h16_policy_ckpt_near_miss_policy_bc5_typedprop_p2_bestpt_"
|
| 489 |
+
"proposal_lattice_types2sparse_prepend_margin0p05_summary.json"
|
| 490 |
+
),
|
| 491 |
+
),
|
| 492 |
+
MethodSpec(
|
| 493 |
+
key="typed_proposal_lattice_nooponly_prepend_margin005",
|
| 494 |
+
label="Typed proposal lattice head, no-op family only, policy-prepended margin 0.05",
|
| 495 |
+
summary_path=(
|
| 496 |
+
"h16_policy_ckpt_near_miss_policy_bc5_typedprop_p2_bestpt_"
|
| 497 |
+
"proposal_lattice_nooponly_prepend_margin0p05_summary.json"
|
| 498 |
+
),
|
| 499 |
+
),
|
| 500 |
MethodSpec(
|
| 501 |
key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",
|
| 502 |
label="K4 composed compatible tangents, no-op bonus 0.03, singleton near-miss bonus 0.01",
|
scripts/build_paper_table_status.py
CHANGED
|
@@ -845,6 +845,36 @@ SPECS = [
|
|
| 845 |
story_role="model-generated counterfactual proposal support test with a conservative policy-abstention margin",
|
| 846 |
pending_job="14962264/14962357/14962364",
|
| 847 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 848 |
ResultSpec(
|
| 849 |
key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",
|
| 850 |
label="K4 composed compatible residual retrieval, no-op bonus 0.03, singleton near-miss bonus 0.01",
|
|
|
|
| 845 |
story_role="model-generated counterfactual proposal support test with a conservative policy-abstention margin",
|
| 846 |
pending_job="14962264/14962357/14962364",
|
| 847 |
),
|
| 848 |
+
ResultSpec(
|
| 849 |
+
key="typed_proposal_lattice_types4safe_prepend_margin005",
|
| 850 |
+
label="Typed proposal lattice head, safe four families, policy-prepended margin 0.05",
|
| 851 |
+
path="h16_policy_ckpt_near_miss_policy_bc5_typedprop_p2_bestpt_proposal_lattice_types4safe_prepend_margin0p05_summary.json",
|
| 852 |
+
clean_deployment="yes",
|
| 853 |
+
same_state_proposals="no",
|
| 854 |
+
expert_proposal="no",
|
| 855 |
+
story_role="anti-goal proposal-family mask test for the model-generated typed lattice",
|
| 856 |
+
pending_job="14963030/14963044",
|
| 857 |
+
),
|
| 858 |
+
ResultSpec(
|
| 859 |
+
key="typed_proposal_lattice_types2sparse_prepend_margin005",
|
| 860 |
+
label="Typed proposal lattice head, sparse no-op/wrong-gripper families, policy-prepended margin 0.05",
|
| 861 |
+
path="h16_policy_ckpt_near_miss_policy_bc5_typedprop_p2_bestpt_proposal_lattice_types2sparse_prepend_margin0p05_summary.json",
|
| 862 |
+
clean_deployment="yes",
|
| 863 |
+
same_state_proposals="no",
|
| 864 |
+
expert_proposal="no",
|
| 865 |
+
story_role="sparse safe-support test for whether typed generated no-op/wrong-gripper proposals recover the clean support gap",
|
| 866 |
+
pending_job="14963031/14963042",
|
| 867 |
+
),
|
| 868 |
+
ResultSpec(
|
| 869 |
+
key="typed_proposal_lattice_nooponly_prepend_margin005",
|
| 870 |
+
label="Typed proposal lattice head, no-op family only, policy-prepended margin 0.05",
|
| 871 |
+
path="h16_policy_ckpt_near_miss_policy_bc5_typedprop_p2_bestpt_proposal_lattice_nooponly_prepend_margin0p05_summary.json",
|
| 872 |
+
clean_deployment="yes",
|
| 873 |
+
same_state_proposals="no",
|
| 874 |
+
expert_proposal="no",
|
| 875 |
+
story_role="minimal generated no-op proposal-family test after sparse typed lattice improved over the full six-family row",
|
| 876 |
+
pending_job="14963032/14963043",
|
| 877 |
+
),
|
| 878 |
ResultSpec(
|
| 879 |
key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",
|
| 880 |
label="K4 composed compatible residual retrieval, no-op bonus 0.03, singleton near-miss bonus 0.01",
|