Auto-sync: 2026-06-29 13:04:05 (part 2)

results/paper_analysis.json

@@ -1,11 +1,11 @@
 {
-  "best_clean_key": "residual_k4_composemasked_grid035040045_noopbonus003",
-  "generated_utc": "2026-06-29T16:57:54+00:00",
+  "best_clean_key": "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003",
+  "generated_utc": "2026-06-29T17:02:31+00:00",
   "mechanism_gap": {
-    "best_clean_vs_direct_same_ckpt": 0.07246376811594196,
-    "best_clean_vs_h16": 0.0579710144927536,
+    "best_clean_vs_direct_same_ckpt": 0.07304347826086954,
+    "best_clean_vs_h16": 0.058550724637681184,
     "same_state_full_vs_no_expert": 0.12347826086956515,
-    "same_state_no_expert_vs_best_clean": 0.21449275362318854,
+    "same_state_no_expert_vs_best_clean": 0.21391304347826096,
     "same_state_no_expert_vs_h16": 0.27246376811594214
   },
   "methods": {
@@ -948,9 +948,101 @@
       "std_success": 0.011579701075616346
     },
     "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003": {
+      "ci95_success": 0.024567904418045634,
       "label": "K4 composed type-consensus tangents, masked, drop near-miss+no-op composite",
-      "missing": true,
-      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_summary.json"
+      "mean_action_mse_to_best": 0.40635193244135687,
+      "mean_progress": 0.5706848979810625,
+      "mean_success": 0.35594202898550725,
+      "num_completed": 3,
+      "per_task_success": {
+        "LiftPegUpright-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.26340865087329823,
+          "std_success": 0.06057881385012102
+        },
+        "PickCube-v1": {
+          "mean_num_groups": 196.66666666666666,
+          "mean_success": 0.3278169093386485,
+          "std_success": 0.02533117351097172
+        },
+        "PullCube-v1": {
+          "mean_num_groups": 81.0,
+          "mean_success": 0.20474165211007314,
+          "std_success": 0.019188581862611215
+        },
+        "PushCube-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.757403601199683,
+          "std_success": 0.07781968814460603
+        },
+        "StackCube-v1": {
+          "mean_num_groups": 93.33333333333333,
+          "mean_success": 0.20044030044030045,
+          "std_success": 0.08896982358351117
+        }
+      },
+      "seed_action_mse_to_best": {
+        "0": 0.39301921862784933,
+        "1": 0.39824521353549286,
+        "2": 0.4277913651607283
+      },
+      "seed_progress": {
+        "0": 0.5583201415569562,
+        "1": 0.5714951352645522,
+        "2": 0.5822394171216793
+      },
+      "seed_success": {
+        "0": 0.35304347826086957,
+        "1": 0.34782608695652173,
+        "2": 0.36695652173913046
+      },
+      "selected_candidate_type_counts": {
+        "retrieval_residual_policy_residual": 1613,
+        "retrieval_residual_residual_near_miss": 8,
+        "retrieval_residual_residual_near_miss+residual_wrong_gripper": 5,
+        "retrieval_residual_residual_no_op": 56,
+        "retrieval_residual_residual_no_op+residual_wrong_gripper": 24,
+        "retrieval_residual_residual_wrong_gripper": 19
+      },
+      "selected_residual_scale_counts": {
+        "0.35": 1631,
+        "0.4": 11,
+        "0.45": 83
+      },
+      "selected_type_outcomes": {
+        "retrieval_residual_policy_residual": {
+          "count": 1613.0,
+          "mean_progress": 0.560971513695882,
+          "success_rate": 0.3453192808431494
+        },
+        "retrieval_residual_residual_near_miss": {
+          "count": 8.0,
+          "mean_progress": 0.8760369084775448,
+          "success_rate": 0.75
+        },
+        "retrieval_residual_residual_near_miss+residual_wrong_gripper": {
+          "count": 5.0,
+          "mean_progress": 0.7557853102684021,
+          "success_rate": 0.6
+        },
+        "retrieval_residual_residual_no_op": {
+          "count": 56.0,
+          "mean_progress": 0.7249722117225507,
+          "success_rate": 0.5
+        },
+        "retrieval_residual_residual_no_op+residual_wrong_gripper": {
+          "count": 24.0,
+          "mean_progress": 0.671148752172788,
+          "success_rate": 0.4583333333333333
+        },
+        "retrieval_residual_residual_wrong_gripper": {
+          "count": 19.0,
+          "mean_progress": 0.6363769314791027,
+          "success_rate": 0.47368421052631576
+        }
+      },
+      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_summary.json",
+      "std_success": 0.009889114266221456
     },
     "residual_k4_composemasked_grid035040045": {
       "ci95_success": 0.03024411731871433,
@@ -6321,12 +6413,12 @@
   },
   "paired_deltas": {
     "best_clean - canonical_h16": {
-      "ci95_delta": 0.042406376811594176,
-      "left": "residual_k4_composemasked_grid035040045_noopbonus003",
-      "mean_delta": 0.057971014492753624,
+      "ci95_delta": 0.04370704391798567,
+      "left": "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003",
+      "mean_delta": 0.05855072463768116,
       "right": "h16_policy_canonical",
       "seed_deltas": {
-        "0": 0.0678260869565217,
+        "0": 0.06956521739130433,
         "1": 0.038260869565217404,
         "2": 0.06782608695652176
       },
@@ -6335,15 +6427,15 @@
         1,
         2
       ],
-      "std_delta": 0.01706948621951936
+      "std_delta": 0.01759303293390552
     },
     "best_clean - direct_same_ckpt": {
-      "ci95_delta": 0.041064533365485906,
-      "left": "residual_k4_composemasked_grid035040045_noopbonus003",
-      "mean_delta": 0.07246376811594202,
+      "ci95_delta": 0.04121578446082111,
+      "left": "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003",
+      "mean_delta": 0.07304347826086956,
       "right": "near_miss_policy_bc5",
       "seed_deltas": {
-        "0": 0.07304347826086954,
+        "0": 0.07478260869565218,
         "1": 0.055652173913043446,
         "2": 0.08869565217391306
       },
@@ -6352,7 +6444,7 @@
         1,
         2
       ],
-      "std_delta": 0.01652936513551684
+      "std_delta": 0.0165902469811643
     },
     "full_lattice - no_expert_lattice": {
       "ci95_delta": 0.02628111194117426,
@@ -6408,14 +6500,14 @@
   },
   "per_task_deltas": {
     "best_clean_vs_h16": {
-      "LiftPegUpright-v1": 0.029344886922320512,
+      "LiftPegUpright-v1": 0.03278131303915899,
       "PickCube-v1": 0.1219595621769535,
       "PullCube-v1": 0.006887167413483192,
       "PushCube-v1": 0.006846429323783432,
       "StackCube-v1": 0.0594738594738595
     },
     "no_expert_vs_best_clean": {
-      "LiftPegUpright-v1": 0.35718494358787223,
+      "LiftPegUpright-v1": 0.35374851747103375,
       "PickCube-v1": 0.29366195398804096,
       "PullCube-v1": -0.006261866788182552,
       "PushCube-v1": 0.043639384064915965,

results/paper_analysis.md

@@ -1,6 +1,6 @@
 # Paper Analysis
 
-Generated: `2026-06-29T16:57:54+00:00`
+Generated: `2026-06-29T17:02:31+00:00`
 
 ## Main Seed Statistics
 
@@ -47,7 +47,7 @@ Generated: `2026-06-29T16:57:54+00:00`
 | residual_k4_compose_grid035040045_noopbonus003 | K4 composed type-consensus tangents, scales 0.35/0.40/0.45, no-op bonus 0.03 | 3 | 34.14% +/- 1.58 | +/- 3.92 | 56.00% | 0.482 | +4.41 pp |
 | residual_k4_composemasked_grid035040045 | K4 composed type-consensus tangents, masked, scales 0.35/0.40/0.45 | 3 | 35.30% +/- 1.22 | +/- 3.02 | 56.91% | 0.410 | +5.57 pp |
 | residual_k4_composemasked_grid035040045_noopbonus003 | K4 composed type-consensus tangents, masked, scales 0.35/0.40/0.45, no-op bonus 0.03 | 3 | 35.54% +/- 1.02 | +/- 2.53 | 57.02% | 0.411 | +5.80 pp |
-| residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003 | K4 composed type-consensus tangents, masked, drop near-miss+no-op composite | 0 | missing | missing | missing | missing | missing |
+| residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003 | K4 composed type-consensus tangents, masked, drop near-miss+no-op composite | 3 | 35.59% +/- 0.99 | +/- 2.46 | 57.07% | 0.406 | +5.86 pp |
 | residual_k4_composemasked_compbonus_grid035040045_noopbonus003 | K4 composed type-consensus tangents, masked, component no-op bonus 0.03 | 3 | 35.36% +/- 1.16 | +/- 2.88 | 56.98% | 0.413 | +5.62 pp |
 | residual_k4_composemasked_l2comp002_grid035040045_noopbonus003 | K4 composed type-consensus tangents, masked, composite L2 penalty 0.02 | 3 | 35.54% +/- 1.06 | +/- 2.64 | 57.03% | 0.408 | +5.80 pp |
 | residual_k4_composemasked_l2comp005_grid035040045_noopbonus003 | K4 composed type-consensus tangents, masked, composite L2 penalty 0.05 | 3 | 35.48% +/- 0.97 | +/- 2.41 | 56.94% | 0.405 | +5.74 pp |
@@ -90,8 +90,8 @@ Generated: `2026-06-29T16:57:54+00:00`
 
 | comparison | seeds | mean delta | 95% CI | seed deltas |
 |---|---:|---:|---:|---|
-| best_clean - canonical_h16 | 3 | +5.80 pp | +/- 4.24 | 0:+6.78, 1:+3.83, 2:+6.78 |
-| best_clean - direct_same_ckpt | 3 | +7.25 pp | +/- 4.11 | 0:+7.30, 1:+5.57, 2:+8.87 |
+| best_clean - canonical_h16 | 3 | +5.86 pp | +/- 4.37 | 0:+6.96, 1:+3.83, 2:+6.78 |
+| best_clean - direct_same_ckpt | 3 | +7.30 pp | +/- 4.12 | 0:+7.48, 1:+5.57, 2:+8.87 |
 | no_expert_lattice - canonical_h16 | 3 | +27.25 pp | +/- 8.58 | 0:+23.30, 1:+28.70, 2:+29.74 |
 | full_lattice - no_expert_lattice | 3 | +12.35 pp | +/- 2.63 | 0:+13.57, 1:+11.83, 2:+11.65 |
 | policy_candidate_lattice - no_expert_lattice | 3 | -16.29 pp | +/- 7.55 | 0:-15.48, 1:-13.74, 2:-19.65 |
@@ -100,7 +100,7 @@ Generated: `2026-06-29T16:57:54+00:00`
 
 | task | h16 policy | best clean | near-miss lattice | no-expert lattice | full lattice | clean-h16 delta | noexpert-clean gap |
 |---|---:|---:|---:|---:|---:|---:|---:|
-| LiftPegUpright-v1 | 23.06% | 26.00% | 62.90% | 61.72% | 73.19% | +2.93 pp | +35.72 pp |
+| LiftPegUpright-v1 | 23.06% | 26.34% | 62.90% | 61.72% | 73.19% | +3.28 pp | +35.37 pp |
 | PickCube-v1 | 20.59% | 32.78% | 58.13% | 62.15% | 84.19% | +12.20 pp | +29.37 pp |
 | PullCube-v1 | 19.79% | 20.47% | 15.02% | 19.85% | 22.41% | +0.69 pp | -0.63 pp |
 | PushCube-v1 | 75.06% | 75.74% | 82.54% | 80.10% | 81.92% | +0.68 pp | +4.36 pp |
@@ -108,9 +108,9 @@ Generated: `2026-06-29T16:57:54+00:00`
 
 ## Mechanism Gap
 
-- Best clean residual transport improves over canonical h16 by +5.80 pp.
+- Best clean residual transport improves over canonical h16 by +5.86 pp.
 - Same-state no-expert lattice improves over canonical h16 by +27.25 pp.
-- Remaining clean-to-same-state proposal gap is +21.45 pp.
+- Remaining clean-to-same-state proposal gap is +21.39 pp.
 - Full lattice adds expert proposals and reaches 69.33%, a +12.35 pp gain over no-expert.
 
 ## Selection Histograms
@@ -119,8 +119,8 @@ Generated: `2026-06-29T16:57:54+00:00`
 - `same_state_no_expert`: lattice_near_miss=1263 (73.2%), lattice_no_op=222 (12.9%), lattice_random_negative=144 (8.3%), lattice_wrong_gripper=62 (3.6%), lattice_wrong_direction=34 (2.0%)
 - `same_state_policy_baseline`: policy_continuous=1022 (59.2%), lattice_near_miss=448 (26.0%), lattice_no_op=119 (6.9%), lattice_random_negative=75 (4.3%), lattice_wrong_gripper=45 (2.6%), lattice_wrong_direction=16 (0.9%)
 - `same_state_full`: lattice_expert=977 (56.6%), lattice_near_miss=348 (20.2%), lattice_no_op=177 (10.3%), lattice_random_negative=138 (8.0%), lattice_wrong_gripper=55 (3.2%), lattice_wrong_direction=30 (1.7%)
-- `residual_k4_composemasked_grid035040045_noopbonus003`: retrieval_residual_policy_residual=1608 (93.2%), retrieval_residual_residual_no_op=55 (3.2%), retrieval_residual_residual_no_op+residual_wrong_gripper=21 (1.2%), retrieval_residual_residual_wrong_gripper=19 (1.1%), retrieval_residual_residual_near_miss+residual_no_op=10 (0.6%), retrieval_residual_residual_near_miss=7 (0.4%), retrieval_residual_residual_near_miss+residual_wrong_gripper=5 (0.3%)
-- `residual_k4_composemasked_grid035040045_noopbonus003` residual scale counts: {'0.35': 1625, '0.4': 11, '0.45': 89}
+- `residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003`: retrieval_residual_policy_residual=1613 (93.5%), retrieval_residual_residual_no_op=56 (3.2%), retrieval_residual_residual_no_op+residual_wrong_gripper=24 (1.4%), retrieval_residual_residual_wrong_gripper=19 (1.1%), retrieval_residual_residual_near_miss=8 (0.5%), retrieval_residual_residual_near_miss+residual_wrong_gripper=5 (0.3%)
+- `residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003` residual scale counts: {'0.35': 1631, '0.4': 11, '0.45': 83}
 
 ## Selected-Type Outcomes
 

results/paper_table_status.json

@@ -1306,14 +1306,14 @@
       "story_role": "exact incompatibility mask for one weak composed tangent pair",
       "fallback_success": null,
       "pending_job": "14915009/14915013",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.35594202898550725,
+      "std_success": 0.009889114266221456,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.058550724637681184
     },
     {
       "key": "retrieval_residual_k4_composemasked_compbonus_grid035040045_noopbonus003",
@@ -2419,23 +2419,23 @@
     }
   ],
   "best_clean": {
-    "key": "retrieval_residual_k4_composemasked_grid035040045_noopbonus003",
-    "label": "K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03",
-    "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_grid035040045_safe_margin0p20_noopbonus0p03_summary.json",
+    "key": "retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003",
+    "label": "K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite",
+    "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_summary.json",
     "clean_deployment": "yes",
     "same_state_proposals": "no",
     "expert_proposal": "no",
-    "story_role": "local tangent composition with anti-goal composite masks on the current best typed prior",
+    "story_role": "exact incompatibility mask for one weak composed tangent pair",
     "fallback_success": null,
-    "pending_job": "14911980/14911983",
+    "pending_job": "14915009/14915013",
     "path_exists": true,
     "status": "complete",
-    "success": 0.3553623188405797,
-    "std_success": 0.010190374394925787,
+    "success": 0.35594202898550725,
+    "std_success": 0.009889114266221456,
     "completed_seeds": null,
     "num_completed": 3,
     "best_config": null,
-    "gain_vs_h16_policy": 0.05797101449275366
+    "gain_vs_h16_policy": 0.058550724637681184
   },
   "best_mechanism_no_expert": {
     "key": "no_expert_lattice",
@@ -2460,7 +2460,7 @@
     "Use no-expert same-state lattice as the conservative mechanism result, not as deployment-clean inference.",
     "Use full lattice only as an upper result because it includes expert proposals.",
     "Do not claim external SOTA from this table alone; add current external baselines separately.",
-    "Current best clean deployment row is K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03 at 35.54%.",
+    "Current best clean deployment row is K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite at 35.59%.",
     "Trust-region field optimization should be framed as a negative/diagnostic ablation.",
     "Train-state counterfactual residual retrieval is a positive clean bridge but remains below the current clean best.",
     "KNN counterfactual residual retrieval is a positive clean bridge but remains below the current clean best.",

results/paper_table_status.md

@@ -71,7 +71,7 @@ Baseline h=16 policy: 29.74%
 | retrieval_residual_k4_compose_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03 | complete | 34.14% | +4.41 pp | yes | no | no | local tangent composition on the current best typed prior |
 | retrieval_residual_k4_composemasked_grid035040045 | K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20 | complete | 35.30% | +5.57 pp | yes | no | no | local tangent composition with anti-goal composite masks |
 | retrieval_residual_k4_composemasked_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03 | complete | 35.54% | +5.80 pp | yes | no | no | local tangent composition with anti-goal composite masks on the current best typed prior |
-| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite | pending 14915009/14915013 | pending | pending | yes | no | no | exact incompatibility mask for one weak composed tangent pair |
+| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite | complete | 35.59% | +5.86 pp | yes | no | no | exact incompatibility mask for one weak composed tangent pair |
 | retrieval_residual_k4_composemasked_compbonus_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, masked, component no-op bonus 0.03 | complete | 35.36% | +5.62 pp | yes | no | no | component-wise sparse prior on the masked local tangent composition chart |
 | retrieval_residual_k4_composemasked_l2comp002_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, masked, composite L2 penalty 0.02 | complete | 35.54% | +5.80 pp | yes | no | no | trust-radius penalty on composed local tangent candidates |
 | retrieval_residual_k4_composemasked_l2comp005_grid035040045_noopbonus003 | K4 composed type-consensus residual retrieval, masked, composite L2 penalty 0.05 | complete | 35.48% | +5.74 pp | yes | no | no | stronger trust-radius penalty on composed local tangent candidates |
@@ -136,7 +136,7 @@ Baseline h=16 policy: 29.74%
 - Use no-expert same-state lattice as the conservative mechanism result, not as deployment-clean inference.
 - Use full lattice only as an upper result because it includes expert proposals.
 - Do not claim external SOTA from this table alone; add current external baselines separately.
-- Current best clean deployment row is K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03 at 35.54%.
+- Current best clean deployment row is K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite at 35.59%.
 - Trust-region field optimization should be framed as a negative/diagnostic ablation.
 - Train-state counterfactual residual retrieval is a positive clean bridge but remains below the current clean best.
 - KNN counterfactual residual retrieval is a positive clean bridge but remains below the current clean best.

scripts/build_paper_analysis.py

@@ -361,6 +361,14 @@ METHODS = [
             "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_summary.json"
         ),
     ),
+    MethodSpec(
+        key="residual_k4_composemasked_dropnmnoop_l2comp002_grid035040045_noopbonus003",
+        label="K4 composed type-consensus tangents, masked, drop near-miss+no-op composite, composite L2 penalty 0.02",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_dropnmnoop_l2comp002_grid035040045_safe_margin0p20_noopbonus0p03_summary.json"
+        ),
+    ),
     MethodSpec(
         key="residual_k4_composemasked_compbonus_grid035040045_noopbonus003",
         label="K4 composed type-consensus tangents, masked, component no-op bonus 0.03",

scripts/build_paper_table_status.py

@@ -705,6 +705,15 @@ SPECS = [
         story_role="exact incompatibility mask for one weak composed tangent pair",
         pending_job="14915009/14915013",
     ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_dropnmnoop_l2comp002_grid035040045_noopbonus003",
+        label="K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite, composite L2 penalty 0.02",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_l2comp002_grid035040045_safe_margin0p20_noopbonus0p03_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="compatible local tangent composition with trust-radius regularization",
+    ),
     ResultSpec(
         key="retrieval_residual_k4_composemasked_compbonus_grid035040045_noopbonus003",
         label="K4 composed type-consensus residual retrieval, masked, component no-op bonus 0.03",