Auto-sync: 2026-06-30 21:33:23

results/paper_analysis.json

@@ -1,6 +1,6 @@
 {
   "best_clean_key": "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001",
-  "generated_utc": "2026-07-01T00:31:55+00:00",
+  "generated_utc": "2026-07-01T01:39:11+00:00",
   "mechanism_gap": {
     "best_clean_vs_direct_same_ckpt": 0.0776811594202898,
     "best_clean_vs_h16": 0.06318840579710144,

results/paper_analysis.md

@@ -1,6 +1,6 @@
 # Paper Analysis
 
-Generated: `2026-07-01T00:31:55+00:00`
+Generated: `2026-07-01T01:39:11+00:00`
 
 ## Main Seed Statistics
 

results/paper_table_status.json

@@ -1647,7 +1647,7 @@
       "expert_proposal": "no",
       "story_role": "type-specific affordance margin: keep near-miss globally sensitive while requiring stronger wrong-gripper evidence only on Pick/Pull",
       "fallback_success": null,
-      "pending_job": "14989757/14989775",
+      "pending_job": "14991117/14991125",
       "path_exists": false,
       "status": "pending",
       "success": null,
@@ -1666,7 +1666,7 @@
       "expert_proposal": "no",
       "story_role": "upper setting for type-specific wrong-gripper evidence on Pick/Pull while preserving global near-miss calibration",
       "fallback_success": null,
-      "pending_job": "14989760/14989780",
+      "pending_job": "14991123/14991126",
       "path_exists": false,
       "status": "pending",
       "success": null,

results/paper_table_status.md

@@ -89,8 +89,8 @@ Baseline h=16 policy: 29.74%
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgmargin005_challenger001 | K4 compatible residual retrieval, near-miss challenger 0.01, wrong-gripper margin 0.05 | complete | 35.59% | +5.86 pp | yes | no | no | family-specific reliability margin upper setting for wrong-gripper challenger evidence |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgchallenger001_pickpull | K4 compatible residual retrieval, near-miss/wrong-gripper challenger gate 0.01 on Pick/Pull only | complete | 36.00% | +6.26 pp | yes | no | no | task-conditional challenger gate using branch trace evidence that wrong-gripper tangents help Pick/Pull but harm other tasks |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmglobal_wgpickpull_challenger001 | K4 compatible residual retrieval, near-miss global + wrong-gripper challenger on Pick/Pull, gate 0.01 | complete | 36.06% | +6.32 pp | yes | no | no | type-specific affordance gate: keep near-miss global while restricting wrong-gripper challenger to acquisition/contact tasks |
-| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmglobal_wgpickpull_wgmargin003_challenger001 | K4 compatible residual retrieval, near-miss global + wrong-gripper Pick/Pull margin 0.03 | pending 14989757/14989775 | pending | pending | yes | no | no | type-specific affordance margin: keep near-miss globally sensitive while requiring stronger wrong-gripper evidence only on Pick/Pull |
-| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmglobal_wgpickpull_wgmargin005_challenger001 | K4 compatible residual retrieval, near-miss global + wrong-gripper Pick/Pull margin 0.05 | pending 14989760/14989780 | pending | pending | yes | no | no | upper setting for type-specific wrong-gripper evidence on Pick/Pull while preserving global near-miss calibration |
+| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmglobal_wgpickpull_wgmargin003_challenger001 | K4 compatible residual retrieval, near-miss global + wrong-gripper Pick/Pull margin 0.03 | pending 14991117/14991125 | pending | pending | yes | no | no | type-specific affordance margin: keep near-miss globally sensitive while requiring stronger wrong-gripper evidence only on Pick/Pull |
+| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmglobal_wgpickpull_wgmargin005_challenger001 | K4 compatible residual retrieval, near-miss global + wrong-gripper Pick/Pull margin 0.05 | pending 14991123/14991126 | pending | pending | yes | no | no | upper setting for type-specific wrong-gripper evidence on Pick/Pull while preserving global near-miss calibration |
 | typed_proposal_lattice_types6_prepend_margin000 | Typed proposal lattice head, six families, policy-prepended margin 0.00 | complete | 31.30% | +1.57 pp | yes | no | no | model-generated counterfactual proposal support test; field scores typed proposals rather than retrieved train-state residuals |
 | typed_proposal_lattice_types6_prepend_margin005 | Typed proposal lattice head, six families, policy-prepended margin 0.05 | complete | 32.17% | +2.43 pp | yes | no | no | model-generated counterfactual proposal support test with a conservative policy-abstention margin |
 | typed_proposal_lattice_types4safe_prepend_margin005 | Typed proposal lattice head, safe four families, policy-prepended margin 0.05 | complete | 33.45% | +3.71 pp | yes | no | no | anti-goal proposal-family mask test for the model-generated typed lattice |

scripts/build_paper_table_status.py

@@ -883,7 +883,7 @@ SPECS = [
         same_state_proposals="no",
         expert_proposal="no",
         story_role="type-specific affordance margin: keep near-miss globally sensitive while requiring stronger wrong-gripper evidence only on Pick/Pull",
-        pending_job="14989757/14989775",
+        pending_job="14991117/14991125",
     ),
     ResultSpec(
         key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmglobal_wgpickpull_wgmargin005_challenger001",
@@ -893,7 +893,7 @@ SPECS = [
         same_state_proposals="no",
         expert_proposal="no",
         story_role="upper setting for type-specific wrong-gripper evidence on Pick/Pull while preserving global near-miss calibration",
-        pending_job="14989760/14989780",
+        pending_job="14991123/14991126",
     ),
     ResultSpec(
         key="typed_proposal_lattice_types6_prepend_margin000",