Auto-sync: 2026-06-28 02:25:24 (part 2)

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_type_success0p50_summary.json

@@ -0,0 +1,286 @@
+{
+  "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
+  "objective": "near_miss_policy_bc5",
+  "out_name": "policy_rollout_retrieval_residual_scale0p50_type_success0p50.json",
+  "num_completed": 3,
+  "baseline_h4_policy_success": 0.2967,
+  "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
+  "mean_success": 0.3333333333333333,
+  "std_success": 0.00821880978478714,
+  "mean_progress": 0.5528276873363749,
+  "mean_action_mse_to_best": 0.4331563813282528,
+  "gain_vs_h4": 0.036633333333333296,
+  "gain_vs_h16_rank_checkpoint": 0.035942028985507246,
+  "rows": [
+    {
+      "seed": 0,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_scale0p50_type_success0p50.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 16,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 1,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.5,
+      "retrieval_residual_scale": 0.5,
+      "policy_rollout_success_rate": 0.33043478260869563,
+      "policy_rollout_progress": 0.544471482436942,
+      "oracle_success_rate": 0.8573913043478261,
+      "action_mse_to_best": 0.4129274774707206,
+      "best_policy_val": {
+        "bc_loss": 0.13721593966086706,
+        "field_effect_loss": 0.009290305380192068,
+        "field_potential_loss": 0.2666468388504452,
+        "field_preference_loss": 0.5130573478009965,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.1933159919248687,
+        "rank_acc": 0.8265031774838766,
+        "rank_loss": 0.5130523675017886,
+        "regret_mae": 0.3756548762321472,
+        "success_accuracy": 0.8773836526605818,
+        "total_loss": 1.5581054819954767
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.33314846369159434,
+          "expert_success_rate": 0.8865979381443299,
+          "num_groups": 97,
+          "oracle_success_rate": 0.9278350515463918,
+          "policy_expert_regret": 1.0792845781009222,
+          "policy_oracle_regret": 1.105700055809365,
+          "policy_rollout_progress": 0.5705123155080166,
+          "policy_rollout_success_rate": 0.2268041237113402,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.3205748214744605,
+          "expert_success_rate": 0.9375,
+          "num_groups": 208,
+          "oracle_success_rate": 0.9471153846153846,
+          "policy_expert_regret": 1.0125951060147669,
+          "policy_oracle_regret": 1.0260179734412724,
+          "policy_rollout_progress": 0.5887099993504727,
+          "policy_rollout_success_rate": 0.3173076923076923,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6671761597518797,
+          "expert_success_rate": 0.19480519480519481,
+          "num_groups": 77,
+          "oracle_success_rate": 0.36363636363636365,
+          "policy_expert_regret": 0.24359342272018458,
+          "policy_oracle_regret": 0.4621977220204743,
+          "policy_rollout_progress": 0.3145493839371514,
+          "policy_rollout_success_rate": 0.2077922077922078,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3714648786773707,
+          "expert_success_rate": 0.8617021276595744,
+          "num_groups": 94,
+          "oracle_success_rate": 0.9893617021276596,
+          "policy_expert_regret": 0.5719942746961371,
+          "policy_oracle_regret": 0.6495450359392674,
+          "policy_rollout_progress": 0.6908804959756263,
+          "policy_rollout_success_rate": 0.6595744680851063,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5267482101446902,
+          "expert_success_rate": 0.6767676767676768,
+          "num_groups": 99,
+          "oracle_success_rate": 0.8585858585858586,
+          "policy_expert_regret": 0.9495171221217724,
+          "policy_oracle_regret": 1.112660668715082,
+          "policy_rollout_progress": 0.4658248358302646,
+          "policy_rollout_success_rate": 0.24242424242424243,
+          "restore_max_error": 3.948807716369629e-07
+        }
+      }
+    },
+    {
+      "seed": 1,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p50_type_success0p50.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 16,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 1,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.5,
+      "retrieval_residual_scale": 0.5,
+      "policy_rollout_success_rate": 0.3269565217391304,
+      "policy_rollout_progress": 0.5530163816378816,
+      "oracle_success_rate": 0.8695652173913043,
+      "action_mse_to_best": 0.42286321135642735,
+      "best_policy_val": {
+        "bc_loss": 0.1315989200439718,
+        "field_effect_loss": 0.010110370814800262,
+        "field_potential_loss": 0.31546536915832096,
+        "field_preference_loss": 0.4960637440284093,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.19023226698239645,
+        "rank_acc": 0.8255469501018524,
+        "rank_loss": 0.4960531195004781,
+        "regret_mae": 0.42156532737943864,
+        "success_accuracy": 0.8807473679383596,
+        "total_loss": 1.561984618504842
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.33098896499723196,
+          "expert_success_rate": 0.8584070796460177,
+          "num_groups": 113,
+          "oracle_success_rate": 0.9380530973451328,
+          "policy_expert_regret": 0.998058971973647,
+          "policy_oracle_regret": 1.0891584641901793,
+          "policy_rollout_progress": 0.619361627022777,
+          "policy_rollout_success_rate": 0.21238938053097345,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.3595517423692281,
+          "expert_success_rate": 0.9402173913043478,
+          "num_groups": 184,
+          "oracle_success_rate": 0.9456521739130435,
+          "policy_expert_regret": 1.1190804282618363,
+          "policy_oracle_regret": 1.1266173332291858,
+          "policy_rollout_progress": 0.5457061722864518,
+          "policy_rollout_success_rate": 0.266304347826087,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6149560018981758,
+          "expert_success_rate": 0.25,
+          "num_groups": 76,
+          "oracle_success_rate": 0.40789473684210525,
+          "policy_expert_regret": 0.27304727383154004,
+          "policy_oracle_regret": 0.5447679004680953,
+          "policy_rollout_progress": 0.27486991495090096,
+          "policy_rollout_success_rate": 0.15789473684210525,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.39232181858372045,
+          "expert_success_rate": 0.8198198198198198,
+          "num_groups": 111,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.34771036054636983,
+          "policy_oracle_regret": 0.38119086757436527,
+          "policy_rollout_progress": 0.8170073306238329,
+          "policy_rollout_success_rate": 0.8018018018018018,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5417878558490794,
+          "expert_success_rate": 0.6923076923076923,
+          "num_groups": 91,
+          "oracle_success_rate": 0.8571428571428571,
+          "policy_expert_regret": 1.122757653777416,
+          "policy_oracle_regret": 1.2949054079068887,
+          "policy_rollout_progress": 0.3956999192526052,
+          "policy_rollout_success_rate": 0.15384615384615385,
+          "restore_max_error": 3.948807716369629e-07
+        }
+      }
+    },
+    {
+      "seed": 2,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p50_type_success0p50.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 16,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 1,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.5,
+      "retrieval_residual_scale": 0.5,
+      "policy_rollout_success_rate": 0.3426086956521739,
+      "policy_rollout_progress": 0.560995197934301,
+      "oracle_success_rate": 0.8765217391304347,
+      "action_mse_to_best": 0.46367845515761036,
+      "best_policy_val": {
+        "bc_loss": 0.11367896075050037,
+        "field_effect_loss": 0.009670218582161598,
+        "field_potential_loss": 0.2641640139950646,
+        "field_preference_loss": 0.5130490180518892,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.2021729110015763,
+        "rank_acc": 0.8333857821093665,
+        "rank_loss": 0.5130119257503085,
+        "regret_mae": 0.3958987047274907,
+        "success_accuracy": 0.8680730561415354,
+        "total_loss": 1.4394984311527677
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.36153392906514153,
+          "expert_success_rate": 0.8229166666666666,
+          "num_groups": 96,
+          "oracle_success_rate": 0.9270833333333334,
+          "policy_expert_regret": 0.8829973929872116,
+          "policy_oracle_regret": 0.9760098507006963,
+          "policy_rollout_progress": 0.6346820058921973,
+          "policy_rollout_success_rate": 0.3229166666666667,
+          "restore_max_error": 3.5762786865234375e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.37652896986239487,
+          "expert_success_rate": 0.9444444444444444,
+          "num_groups": 198,
+          "oracle_success_rate": 0.9595959595959596,
+          "policy_expert_regret": 1.0838303450457376,
+          "policy_oracle_regret": 1.090913795574446,
+          "policy_rollout_progress": 0.5840688492688868,
+          "policy_rollout_success_rate": 0.2727272727272727,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.7301618857516183,
+          "expert_success_rate": 0.24444444444444444,
+          "num_groups": 90,
+          "oracle_success_rate": 0.4666666666666667,
+          "policy_expert_regret": 0.30010166329642135,
+          "policy_oracle_regret": 0.5302090961693062,
+          "policy_rollout_progress": 0.31182427391823797,
+          "policy_rollout_success_rate": 0.2111111111111111,
+          "restore_max_error": 4.0978193283081055e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.4178345319215614,
+          "expert_success_rate": 0.8514851485148515,
+          "num_groups": 101,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.38219682946063505,
+          "policy_oracle_regret": 0.4192075215943969,
+          "policy_rollout_progress": 0.7986142605838209,
+          "policy_rollout_success_rate": 0.7821782178217822,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5493251227877206,
+          "expert_success_rate": 0.7666666666666667,
+          "num_groups": 90,
+          "oracle_success_rate": 0.9111111111111111,
+          "policy_expert_regret": 1.15032865587208,
+          "policy_oracle_regret": 1.3123159415192074,
+          "policy_rollout_progress": 0.41414343466361364,
+          "policy_rollout_success_rate": 0.15555555555555556,
+          "restore_max_error": 4.76837158203125e-07
+        }
+      }
+    }
+  ]
+}

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_type_success0p50_summary.md

@@ -0,0 +1,19 @@
+# h=16 Best-Policy Checkpoint Rollout
+
+Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
+Objective: `near_miss_policy_bc5`
+Result file: `policy_rollout_retrieval_residual_scale0p50_type_success0p50.json`
+Completed seeds: 3
+Baseline h=4 policy success: 29.67%
+Baseline h=16 rank-checkpoint success: 29.74%
+
+Mean success: 33.33% +/- 0.82%
+Gain vs h=16 rank checkpoint: +3.59%
+Mean progress: 55.28%
+Mean action MSE to best: 0.433
+
+| seed | mode | k | retrieval K | retrieval metric | min type success | residual scale | sigma | opt steps | trust | success | progress | oracle | action MSE |
+|---:|---|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| 0 | retrieval_residual | 16 | 1 | raw | 0.50 | 0.50 | 0.00 | 0 | 0.00 | 33.04% | 54.45% | 85.74% | 0.413 |
+| 1 | retrieval_residual | 16 | 1 | raw | 0.50 | 0.50 | 0.00 | 0 | 0.00 | 32.70% | 55.30% | 86.96% | 0.423 |
+| 2 | retrieval_residual | 16 | 1 | raw | 0.50 | 0.50 | 0.00 | 0 | 0.00 | 34.26% | 56.10% | 87.65% | 0.464 |

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_type_success0p75_summary.json

@@ -0,0 +1,286 @@
+{
+  "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
+  "objective": "near_miss_policy_bc5",
+  "out_name": "policy_rollout_retrieval_residual_scale0p50_type_success0p75.json",
+  "num_completed": 3,
+  "baseline_h4_policy_success": 0.2967,
+  "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
+  "mean_success": 0.3333333333333333,
+  "std_success": 0.00821880978478714,
+  "mean_progress": 0.5528276873363749,
+  "mean_action_mse_to_best": 0.4331563813282528,
+  "gain_vs_h4": 0.036633333333333296,
+  "gain_vs_h16_rank_checkpoint": 0.035942028985507246,
+  "rows": [
+    {
+      "seed": 0,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_scale0p50_type_success0p75.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 16,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 1,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.75,
+      "retrieval_residual_scale": 0.5,
+      "policy_rollout_success_rate": 0.33043478260869563,
+      "policy_rollout_progress": 0.544471482436942,
+      "oracle_success_rate": 0.8573913043478261,
+      "action_mse_to_best": 0.4129274774707206,
+      "best_policy_val": {
+        "bc_loss": 0.13721593966086706,
+        "field_effect_loss": 0.009290305380192068,
+        "field_potential_loss": 0.2666468388504452,
+        "field_preference_loss": 0.5130573478009965,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.1933159919248687,
+        "rank_acc": 0.8265031774838766,
+        "rank_loss": 0.5130523675017886,
+        "regret_mae": 0.3756548762321472,
+        "success_accuracy": 0.8773836526605818,
+        "total_loss": 1.5581054819954767
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.33314846369159434,
+          "expert_success_rate": 0.8865979381443299,
+          "num_groups": 97,
+          "oracle_success_rate": 0.9278350515463918,
+          "policy_expert_regret": 1.0792845781009222,
+          "policy_oracle_regret": 1.105700055809365,
+          "policy_rollout_progress": 0.5705123155080166,
+          "policy_rollout_success_rate": 0.2268041237113402,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.3205748214744605,
+          "expert_success_rate": 0.9375,
+          "num_groups": 208,
+          "oracle_success_rate": 0.9471153846153846,
+          "policy_expert_regret": 1.0125951060147669,
+          "policy_oracle_regret": 1.0260179734412724,
+          "policy_rollout_progress": 0.5887099993504727,
+          "policy_rollout_success_rate": 0.3173076923076923,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6671761597518797,
+          "expert_success_rate": 0.19480519480519481,
+          "num_groups": 77,
+          "oracle_success_rate": 0.36363636363636365,
+          "policy_expert_regret": 0.24359342272018458,
+          "policy_oracle_regret": 0.4621977220204743,
+          "policy_rollout_progress": 0.3145493839371514,
+          "policy_rollout_success_rate": 0.2077922077922078,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3714648786773707,
+          "expert_success_rate": 0.8617021276595744,
+          "num_groups": 94,
+          "oracle_success_rate": 0.9893617021276596,
+          "policy_expert_regret": 0.5719942746961371,
+          "policy_oracle_regret": 0.6495450359392674,
+          "policy_rollout_progress": 0.6908804959756263,
+          "policy_rollout_success_rate": 0.6595744680851063,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5267482101446902,
+          "expert_success_rate": 0.6767676767676768,
+          "num_groups": 99,
+          "oracle_success_rate": 0.8585858585858586,
+          "policy_expert_regret": 0.9495171221217724,
+          "policy_oracle_regret": 1.112660668715082,
+          "policy_rollout_progress": 0.4658248358302646,
+          "policy_rollout_success_rate": 0.24242424242424243,
+          "restore_max_error": 3.948807716369629e-07
+        }
+      }
+    },
+    {
+      "seed": 1,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_scale0p50_type_success0p75.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 16,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 1,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.75,
+      "retrieval_residual_scale": 0.5,
+      "policy_rollout_success_rate": 0.3269565217391304,
+      "policy_rollout_progress": 0.5530163816378816,
+      "oracle_success_rate": 0.8695652173913043,
+      "action_mse_to_best": 0.42286321135642735,
+      "best_policy_val": {
+        "bc_loss": 0.1315989200439718,
+        "field_effect_loss": 0.010110370814800262,
+        "field_potential_loss": 0.31546536915832096,
+        "field_preference_loss": 0.4960637440284093,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.19023226698239645,
+        "rank_acc": 0.8255469501018524,
+        "rank_loss": 0.4960531195004781,
+        "regret_mae": 0.42156532737943864,
+        "success_accuracy": 0.8807473679383596,
+        "total_loss": 1.561984618504842
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.33098896499723196,
+          "expert_success_rate": 0.8584070796460177,
+          "num_groups": 113,
+          "oracle_success_rate": 0.9380530973451328,
+          "policy_expert_regret": 0.998058971973647,
+          "policy_oracle_regret": 1.0891584641901793,
+          "policy_rollout_progress": 0.619361627022777,
+          "policy_rollout_success_rate": 0.21238938053097345,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.3595517423692281,
+          "expert_success_rate": 0.9402173913043478,
+          "num_groups": 184,
+          "oracle_success_rate": 0.9456521739130435,
+          "policy_expert_regret": 1.1190804282618363,
+          "policy_oracle_regret": 1.1266173332291858,
+          "policy_rollout_progress": 0.5457061722864518,
+          "policy_rollout_success_rate": 0.266304347826087,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6149560018981758,
+          "expert_success_rate": 0.25,
+          "num_groups": 76,
+          "oracle_success_rate": 0.40789473684210525,
+          "policy_expert_regret": 0.27304727383154004,
+          "policy_oracle_regret": 0.5447679004680953,
+          "policy_rollout_progress": 0.27486991495090096,
+          "policy_rollout_success_rate": 0.15789473684210525,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.39232181858372045,
+          "expert_success_rate": 0.8198198198198198,
+          "num_groups": 111,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.34771036054636983,
+          "policy_oracle_regret": 0.38119086757436527,
+          "policy_rollout_progress": 0.8170073306238329,
+          "policy_rollout_success_rate": 0.8018018018018018,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5417878558490794,
+          "expert_success_rate": 0.6923076923076923,
+          "num_groups": 91,
+          "oracle_success_rate": 0.8571428571428571,
+          "policy_expert_regret": 1.122757653777416,
+          "policy_oracle_regret": 1.2949054079068887,
+          "policy_rollout_progress": 0.3956999192526052,
+          "policy_rollout_success_rate": 0.15384615384615385,
+          "restore_max_error": 3.948807716369629e-07
+        }
+      }
+    },
+    {
+      "seed": 2,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_scale0p50_type_success0p75.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 16,
+      "candidate_sigma": 0.0,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 1,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.75,
+      "retrieval_residual_scale": 0.5,
+      "policy_rollout_success_rate": 0.3426086956521739,
+      "policy_rollout_progress": 0.560995197934301,
+      "oracle_success_rate": 0.8765217391304347,
+      "action_mse_to_best": 0.46367845515761036,
+      "best_policy_val": {
+        "bc_loss": 0.11367896075050037,
+        "field_effect_loss": 0.009670218582161598,
+        "field_potential_loss": 0.2641640139950646,
+        "field_preference_loss": 0.5130490180518892,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.2021729110015763,
+        "rank_acc": 0.8333857821093665,
+        "rank_loss": 0.5130119257503085,
+        "regret_mae": 0.3958987047274907,
+        "success_accuracy": 0.8680730561415354,
+        "total_loss": 1.4394984311527677
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.36153392906514153,
+          "expert_success_rate": 0.8229166666666666,
+          "num_groups": 96,
+          "oracle_success_rate": 0.9270833333333334,
+          "policy_expert_regret": 0.8829973929872116,
+          "policy_oracle_regret": 0.9760098507006963,
+          "policy_rollout_progress": 0.6346820058921973,
+          "policy_rollout_success_rate": 0.3229166666666667,
+          "restore_max_error": 3.5762786865234375e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.37652896986239487,
+          "expert_success_rate": 0.9444444444444444,
+          "num_groups": 198,
+          "oracle_success_rate": 0.9595959595959596,
+          "policy_expert_regret": 1.0838303450457376,
+          "policy_oracle_regret": 1.090913795574446,
+          "policy_rollout_progress": 0.5840688492688868,
+          "policy_rollout_success_rate": 0.2727272727272727,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.7301618857516183,
+          "expert_success_rate": 0.24444444444444444,
+          "num_groups": 90,
+          "oracle_success_rate": 0.4666666666666667,
+          "policy_expert_regret": 0.30010166329642135,
+          "policy_oracle_regret": 0.5302090961693062,
+          "policy_rollout_progress": 0.31182427391823797,
+          "policy_rollout_success_rate": 0.2111111111111111,
+          "restore_max_error": 4.0978193283081055e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.4178345319215614,
+          "expert_success_rate": 0.8514851485148515,
+          "num_groups": 101,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.38219682946063505,
+          "policy_oracle_regret": 0.4192075215943969,
+          "policy_rollout_progress": 0.7986142605838209,
+          "policy_rollout_success_rate": 0.7821782178217822,
+          "restore_max_error": 4.76837158203125e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5493251227877206,
+          "expert_success_rate": 0.7666666666666667,
+          "num_groups": 90,
+          "oracle_success_rate": 0.9111111111111111,
+          "policy_expert_regret": 1.15032865587208,
+          "policy_oracle_regret": 1.3123159415192074,
+          "policy_rollout_progress": 0.41414343466361364,
+          "policy_rollout_success_rate": 0.15555555555555556,
+          "restore_max_error": 4.76837158203125e-07
+        }
+      }
+    }
+  ]
+}

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p50_type_success0p75_summary.md

@@ -0,0 +1,19 @@
+# h=16 Best-Policy Checkpoint Rollout
+
+Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
+Objective: `near_miss_policy_bc5`
+Result file: `policy_rollout_retrieval_residual_scale0p50_type_success0p75.json`
+Completed seeds: 3
+Baseline h=4 policy success: 29.67%
+Baseline h=16 rank-checkpoint success: 29.74%
+
+Mean success: 33.33% +/- 0.82%
+Gain vs h=16 rank checkpoint: +3.59%
+Mean progress: 55.28%
+Mean action MSE to best: 0.433
+
+| seed | mode | k | retrieval K | retrieval metric | min type success | residual scale | sigma | opt steps | trust | success | progress | oracle | action MSE |
+|---:|---|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| 0 | retrieval_residual | 16 | 1 | raw | 0.75 | 0.50 | 0.00 | 0 | 0.00 | 33.04% | 54.45% | 85.74% | 0.413 |
+| 1 | retrieval_residual | 16 | 1 | raw | 0.75 | 0.50 | 0.00 | 0 | 0.00 | 32.70% | 55.30% | 86.96% | 0.423 |
+| 2 | retrieval_residual | 16 | 1 | raw | 0.75 | 0.50 | 0.00 | 0 | 0.00 | 34.26% | 56.10% | 87.65% | 0.464 |

results/paper_core_results.md

@@ -24,7 +24,13 @@ baseline is the h=16 rank-checkpoint online rollout (`29.74%`).
 | Field-selected no-expert policy + field, aligned allmap | No | No | 26.49% | -3.25 pp | Field scoring around the aligned student remains below baseline |
 | Train-state residual retrieval | No | No | 32.12% | +2.38 pp | Transferred counterfactual residuals are a positive clean bridge |
 | Train-state residual retrieval, scale 0.25 | No | No | 32.93% | +3.19 pp | Smaller tangent step ties the previous clean best |
-| Train-state residual retrieval, scale 0.50 | No | No | 33.33% | +3.59 pp | Current best deployment-clean bridge; calibrated local tangent transport |
+| Train-state residual retrieval, scale 0.50 | No | No | 33.33% | +3.59 pp | Calibrated local tangent transport |
+| Train-state residual retrieval, no random residuals | No | No | 33.45% | +3.71 pp | Removing anti-goal random residuals helps slightly |
+| Train-state residual retrieval, no random/wrong-direction residuals | No | No | 33.57% | +3.83 pp | Anti-goal family masking improves the clean bridge |
+| Train-state residual retrieval, policy/no-op/wrong-gripper residuals | No | No | 33.68% | +3.94 pp | Current best deployment-clean diagnostic |
+| Train-state residual retrieval, z-score metric | No | No | 32.23% | +2.49 pp | State normalization hurts nearest tangent retrieval here |
+| Train-state residual retrieval, z-score metric + anti-goal mask | No | No | 32.75% | +3.01 pp | Masking helps z-score but remains below raw |
+| Train-state residual retrieval, train family reliability prior | No | No | 33.33% | +3.59 pp | Train terminal-success thresholds through 0.75 do not filter enough |
 | Train-state residual retrieval, scale 0.75 | No | No | 32.70% | +2.96 pp | Larger tangent steps begin to lose success |
 | Train-state residual retrieval, scale 1.25 | No | No | 32.52% | +2.78 pp | Further scale increase does not help |
 | Residual+Gaussian hybrid, K32 sigma0.35 | No | No | 31.30% | +1.57 pp | Adding policy-centered Gaussian proposals dilutes residual transport |
@@ -48,17 +54,19 @@ Suggested main-table rows:
 7. Field-selected no-expert policy + field, seed-0 train map
 8. Field-selected no-expert policy + field, aligned allmap
 9. Train-state residual retrieval, scale 0.50
-10. Residual+Gaussian hybrid, K32 sigma0.35
-11. Lattice, near-miss only
-12. Lattice, no expert
-13. Lattice, full
-14. Oracle ceiling
+10. Train-state residual retrieval, typed safe families
+11. Residual+Gaussian hybrid, K32 sigma0.35
+12. Lattice, near-miss only
+13. Lattice, no expert
+14. Lattice, full
+15. Oracle ceiling
 
 Suggested claim:
 
 > DoVLA-CIL is not a better behavior-cloning policy; it is a local counterfactual action
-> selection rule. Deployment-clean calibrated counterfactual residual transport gives the
-> strongest clean gain so far, while field-gradient ascent, KNN residual retrieval, broader
-> non-expert BC targets, field-teacher distillation, and residual+Gaussian hybrids fail. The
-> large effect appears only when the field is queried on same-state intervention proposals,
-> and the mechanism is isolated to local near-miss counterfactual geometry.
+> selection rule. Deployment-clean typed counterfactual residual transport gives the strongest
+> clean gain so far, while field-gradient ascent, KNN residual retrieval, broader non-expert BC
+> targets, field-teacher distillation, z-score retrieval, train-family reliability priors, and
+> residual+Gaussian hybrids fail. The large effect appears only when the field is queried on
+> same-state intervention proposals, and the mechanism is isolated to local near-miss
+> counterfactual geometry.

results/paper_story_memo.md

@@ -16,7 +16,7 @@ when queried on proposal geometry that matches those local counterfactuals.
 | Same-state local counterfactual proposals are the mechanism | near-miss-only lattice is 55.94%; removing expert+near_miss drops to 25.57% | Strongly supported |
 | Conservative same-state result is large | no-expert lattice is 56.99% vs 29.74% policy | Main result |
 | Full lattice gives upper result | full lattice is 69.33%, oracle is 86.78% | Strong but label expert proposal clearly |
-| Deployment-clean proposal is currently a bottleneck | best clean residual transport is 33.33%, far below 56.99% | Supported |
+| Deployment-clean proposal is currently a bottleneck | best clean residual transport is 33.68%, far below 56.99% | Supported |
 | Gradient-based field optimization does not solve the clean proposal gap | `field_optim` best observed result is 25.39% | Negative diagnostic |
 | A broader non-expert proposal target does not reduce the proposal gap | direct broad non-expert policy is 27.88%; with field scoring it is 26.49% | Negative diagnostic |
 | Counterfactual residuals transfer better than absolute retrieved actions | nearest residual retrieval is 32.12% vs absolute retrieval 28.93%; KNN4 residual drops to 29.91% | Supported as a clean bridge |
@@ -24,9 +24,9 @@ when queried on proposal geometry that matches those local counterfactuals.
 | Residual transport and Gaussian local proposals are not complementary here | hybrid K32/K64 reach 31.30%/30.90%, below residual-only | Negative diagnostic |
 | Seed-0 train-split field-teacher distillation does not solve the proposal gap | direct student is 26.84%; with field scoring it is 27.65% | Negative diagnostic |
 | All-split field-teacher distillation does not fix checkpointing/coverage | allmap direct is 28.00%; field-guided best is 26.49% despite 100% target coverage | Negative diagnostic |
-| Residual family consistency is the next hypothesis | field-selected random/wrong-direction residuals have low rollout success; masked residual jobs are active | Active |
-| Retrieval metric locality is the next hypothesis | z-score train-bank retrieval jobs are active after unit smoke `14859165` passed | Active |
-| Train-split residual family reliability is the next paper-safe hypothesis | reliability-prior jobs are active and use only train-split candidate family outcomes | Active |
+| Residual family consistency improves clean transport | policy/no-op/wrong-gripper typed residuals reach 33.68%, above raw 33.33% | Supported as diagnostic |
+| Z-score retrieval metric does not help | z-score rows reach 32.23-32.81%, below raw retrieval | Negative diagnostic |
+| Train-split residual family reliability does not recover the typed mask | thresholds through 0.75 do not filter the bad families; rows stay at 33.33% | Negative diagnostic |
 
 ## Main Table Candidate
 
@@ -44,11 +44,14 @@ clean proposal result, the intended main rows are:
 8. Field-selected no-expert proposal + field, aligned allmap: 26.49%
 9. Train-state residual retrieval: 32.12%
 10. Train-state residual retrieval, scale 0.50: 33.33%
-11. Residual+Gaussian hybrid K32/K64: 31.30% / 30.90%
-12. Lattice, near-miss only: 55.94%
-13. Lattice, no expert: 56.99%
-14. Lattice, full: 69.33%
-15. Oracle ceiling: 86.78%
+11. Train-state residual retrieval, typed safe families: 33.68%
+12. Z-score residual retrieval: 32.23-32.81%
+13. Train-family reliability prior: 32.93-33.33%
+14. Residual+Gaussian hybrid K32/K64: 31.30% / 30.90%
+15. Lattice, near-miss only: 55.94%
+16. Lattice, no expert: 56.99%
+17. Lattice, full: 69.33%
+18. Oracle ceiling: 86.78%
 
 ## Novelty Framing
 
@@ -74,9 +77,9 @@ test-time search. The cleaner novelty is:
 | Method is just a bundle of tricks | use mechanism ablations to show one central idea: local counterfactual field | avoid presenting unrelated variants as core |
 | Not SOTA enough | current clean deploy result is modest | need external baselines and stronger proposal generator before claiming SOTA |
 
-## Active Jobs
+## Job Status
 
-Last checked: `2026-06-28 05:47 UTC`.
+Last checked: `2026-06-28 06:24 UTC`. No DoVLA jobs are currently queued.
 
 - `14858328`-`14858333`: completed train-split `field_selected_noexpert_bc5`;
   direct rollout is 26.84%, field-guided best is 27.65%.
@@ -87,66 +90,26 @@ Last checked: `2026-06-28 05:47 UTC`.
   Earlier smoke jobs `14858889`/`14858894` caught and fixed two scale wiring bugs
   before rollout jobs started.
 - `14858875`-`14858883`: completed nearest residual scale sweep. Scale `0.50`
-  is the current best clean deployment bridge at 33.33%; scale `0.25` ties the
-  previous 32.93% clean best; larger scales are weaker.
+  reaches 33.33%; scale `0.25` ties the previous 32.93% clean best; larger
+  scales are weaker.
 - `14859041`: completed CPU Apptainer unit smoke for hybrid residual+Gaussian selection.
 - `14859042`-`14859046`: completed hybrid residual+Gaussian jobs; K32 reaches
   31.30% and K64 reaches 30.90%, both below residual-only transport.
-- `14859188`/`14859189`: active masked residual eval/summary, scale `0.50`,
-  excluding `residual_random_negative`.
-- `14859191`/`14859192`: active masked residual eval/summary, scale `0.50`,
-  excluding `residual_random_negative` and `residual_wrong_direction`.
-- `14859193`/`14859194`: active masked residual eval/summary, scale `0.25`,
-  excluding `residual_random_negative` and `residual_wrong_direction`.
-- `14859195`/`14859196`: active typed residual eval/summary, scale `0.50`,
-  keeping policy/no-op/wrong-gripper residual families.
-- `14859203`: rebuild `paper_table_status.*` after all masked and z-score summaries.
+- `14859188`-`14859203`: completed masked/z-score residual retrieval batch.
+  Best row is typed safe residual transport at 33.68%; z-score retrieval is
+  negative.
 - `14859165`: completed Apptainer unit smoke for z-score retrieval metric.
-- `14859197`/`14859198`: active z-score retrieval eval/summary, scale `0.50`.
-- `14859199`/`14859200`: active z-score retrieval eval/summary, scale `0.50`,
-  excluding `residual_random_negative` and `residual_wrong_direction`.
-- `14859201`/`14859202`: active z-score retrieval eval/summary, scale `0.25`,
-  excluding `residual_random_negative` and `residual_wrong_direction`.
-- `14859293`/`14859294`: active train-family reliability eval/summary, scale
-  `0.50`, minimum train success `0.10`.
-- `14859295`/`14859296`: active train-family reliability eval/summary, scale
-  `0.50`, minimum train success `0.25`.
-- `14859297`/`14859298`: active train-family reliability eval/summary, scale
-  `0.25`, minimum train success `0.25`.
-- `14859299`: rebuild `paper_table_status.*` after reliability-prior summaries.
-- `14859398`/`14859399`: active train-family reliability eval/summary, scale
-  `0.50`, minimum train success `0.50`.
-- `14859400`/`14859401`: active train-family reliability eval/summary, scale
-  `0.50`, minimum train success `0.75`.
-- `14859402`: rebuild `paper_table_status.*` after high-threshold reliability
-  summaries.
-
-## Decision Rule For Masked Residual Jobs
-
-- If a masked row beats 33.33%, promote it as evidence that transferable
-  counterfactual residuals need family-consistent local tangent proposals, not
-  anti-goal residuals.
-- If masks land near 33.33% but do not beat it, keep scale `0.50` as the clean
-  residual result and present masking as a diagnostic of field over-selection.
-- If masks fail, keep the story focused on residual scale calibration and the
-  larger same-state counterfactual mechanism.
-
-## Decision Rule For Z-Score Retrieval Jobs
-
-- If z-score retrieval beats 33.33%, promote state-normalized tangent retrieval
-  as the best deployment-clean bridge.
-- If z-score masks only help with the anti-goal residual exclusions, frame
-  retrieval locality and residual family consistency as two sides of the same
-  tangent-transport bottleneck.
-- If z-score retrieval fails, keep the raw nearest-state residual result as the
-  clean bridge and treat metric normalization as a negative ablation.
-
-## Decision Rule For Reliability-Prior Jobs
-
-- If train-family reliability beats the validation-diagnostic safe-types mask,
-  promote it as the paper-safe clean bridge because it uses only training
-  counterfactual outcomes.
-- If it improves over raw scale `0.50` but stays below safe-types, present it as
-  the deployable analogue of typed family masking.
-- If it fails, keep the best clean row as a diagnostic and avoid overclaiming
-  deployment-clean performance.
+- `14859293`-`14859402`: completed train-family reliability-prior batch.
+  Thresholds `0.10`, `0.25`, `0.50`, and `0.75` do not filter the bad residual
+  families and remain at the raw scale-0.50 result (33.33%), except scale-0.25
+  threshold `0.25` at 32.93%.
+
+## Decision Notes
+
+- Promote same-state no-expert lattice (56.99%) as the conservative mechanism
+  result.
+- Use typed safe residual transport (33.68%) only as the current best clean
+  deployment diagnostic, not as a SOTA claim.
+- Treat z-score retrieval, train-family reliability priors, Gaussian hybrids,
+  field optimization, and field-teacher distillation as negative diagnostics
+  that sharpen the story around local counterfactual proposal geometry.

results/paper_table_status.json

@@ -422,6 +422,82 @@
       "best_config": null,
       "gain_vs_h16_policy": 0.03942028985507251
     },
+    {
+      "key": "retrieval_residual_scale035_safe_types",
+      "label": "Train-state residual retrieval, scale 0.35, policy/no-op/wrong-gripper residuals",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "typed tangent scale fine sweep",
+      "fallback_success": null,
+      "pending_job": "14859503/14859504",
+      "path_exists": false,
+      "status": "pending",
+      "success": null,
+      "std_success": null,
+      "completed_seeds": null,
+      "num_completed": null,
+      "best_config": null,
+      "gain_vs_h16_policy": null
+    },
+    {
+      "key": "retrieval_residual_scale045_safe_types",
+      "label": "Train-state residual retrieval, scale 0.45, policy/no-op/wrong-gripper residuals",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p45_safe_types_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "typed tangent scale fine sweep",
+      "fallback_success": null,
+      "pending_job": "14859505/14859506",
+      "path_exists": false,
+      "status": "pending",
+      "success": null,
+      "std_success": null,
+      "completed_seeds": null,
+      "num_completed": null,
+      "best_config": null,
+      "gain_vs_h16_policy": null
+    },
+    {
+      "key": "retrieval_residual_scale060_safe_types",
+      "label": "Train-state residual retrieval, scale 0.60, policy/no-op/wrong-gripper residuals",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p60_safe_types_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "typed tangent scale fine sweep",
+      "fallback_success": null,
+      "pending_job": "14859507/14859508",
+      "path_exists": false,
+      "status": "pending",
+      "success": null,
+      "std_success": null,
+      "completed_seeds": null,
+      "num_completed": null,
+      "best_config": null,
+      "gain_vs_h16_policy": null
+    },
+    {
+      "key": "retrieval_residual_scale070_safe_types",
+      "label": "Train-state residual retrieval, scale 0.70, policy/no-op/wrong-gripper residuals",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p70_safe_types_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "typed tangent scale fine sweep",
+      "fallback_success": null,
+      "pending_job": "14859509/14859510",
+      "path_exists": false,
+      "status": "pending",
+      "success": null,
+      "std_success": null,
+      "completed_seeds": null,
+      "num_completed": null,
+      "best_config": null,
+      "gain_vs_h16_policy": null
+    },
     {
       "key": "retrieval_residual_scale050_type_success010",
       "label": "Train-state residual retrieval, scale 0.50, train family success >= 0.10",
@@ -470,14 +546,14 @@
       "story_role": "train-split residual family reliability prior",
       "fallback_success": null,
       "pending_job": "14859398/14859399",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.3333333333333333,
+      "std_success": 0.00821880978478714,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.035942028985507246
     },
     {
       "key": "retrieval_residual_scale050_type_success075",
@@ -489,14 +565,14 @@
       "story_role": "train-split residual family reliability prior",
       "fallback_success": null,
       "pending_job": "14859400/14859401",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.3333333333333333,
+      "std_success": 0.00821880978478714,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.035942028985507246
     },
     {
       "key": "retrieval_residual_scale025_type_success025",

results/paper_table_status.md

@@ -25,10 +25,14 @@ Baseline h=16 policy: 29.74%
 | retrieval_residual_scale050_no_random_wrongdir | Train-state residual retrieval, scale 0.50, no random/wrong-direction residuals | complete | 33.57% | +3.83 pp | yes | no | no | anti-goal residual family mask ablation |
 | retrieval_residual_scale025_no_random_wrongdir | Train-state residual retrieval, scale 0.25, no random/wrong-direction residuals | complete | 33.45% | +3.71 pp | yes | no | no | anti-goal residual family mask ablation |
 | retrieval_residual_scale050_safe_types | Train-state residual retrieval, scale 0.50, policy/no-op/wrong-gripper residuals | complete | 33.68% | +3.94 pp | yes | no | no | typed tangent-family mask ablation |
+| retrieval_residual_scale035_safe_types | Train-state residual retrieval, scale 0.35, policy/no-op/wrong-gripper residuals | pending 14859503/14859504 | pending | pending | yes | no | no | typed tangent scale fine sweep |
+| retrieval_residual_scale045_safe_types | Train-state residual retrieval, scale 0.45, policy/no-op/wrong-gripper residuals | pending 14859505/14859506 | pending | pending | yes | no | no | typed tangent scale fine sweep |
+| retrieval_residual_scale060_safe_types | Train-state residual retrieval, scale 0.60, policy/no-op/wrong-gripper residuals | pending 14859507/14859508 | pending | pending | yes | no | no | typed tangent scale fine sweep |
+| retrieval_residual_scale070_safe_types | Train-state residual retrieval, scale 0.70, policy/no-op/wrong-gripper residuals | pending 14859509/14859510 | pending | pending | yes | no | no | typed tangent scale fine sweep |
 | retrieval_residual_scale050_type_success010 | Train-state residual retrieval, scale 0.50, train family success >= 0.10 | complete | 33.33% | +3.59 pp | yes | no | no | train-split residual family reliability prior |
 | retrieval_residual_scale050_type_success025 | Train-state residual retrieval, scale 0.50, train family success >= 0.25 | complete | 33.33% | +3.59 pp | yes | no | no | train-split residual family reliability prior |
-| retrieval_residual_scale050_type_success050 | Train-state residual retrieval, scale 0.50, train family success >= 0.50 | pending 14859398/14859399 | pending | pending | yes | no | no | train-split residual family reliability prior |
-| retrieval_residual_scale050_type_success075 | Train-state residual retrieval, scale 0.50, train family success >= 0.75 | pending 14859400/14859401 | pending | pending | yes | no | no | train-split residual family reliability prior |
+| retrieval_residual_scale050_type_success050 | Train-state residual retrieval, scale 0.50, train family success >= 0.50 | complete | 33.33% | +3.59 pp | yes | no | no | train-split residual family reliability prior |
+| retrieval_residual_scale050_type_success075 | Train-state residual retrieval, scale 0.50, train family success >= 0.75 | complete | 33.33% | +3.59 pp | yes | no | no | train-split residual family reliability prior |
 | retrieval_residual_scale025_type_success025 | Train-state residual retrieval, scale 0.25, train family success >= 0.25 | complete | 32.93% | +3.19 pp | yes | no | no | train-split residual family reliability prior |
 | retrieval_residual_scale075 | Train-state residual retrieval, scale 0.75 | complete | 32.70% | +2.96 pp | yes | no | no | tangent transport scale ablation |
 | retrieval_residual_scale125 | Train-state residual retrieval, scale 1.25 | complete | 32.52% | +2.78 pp | yes | no | no | tangent transport scale ablation |

scripts/build_paper_table_status.py

@@ -235,6 +235,46 @@ SPECS = [
         story_role="typed tangent-family mask ablation",
         pending_job="14859195/14859196",
     ),
+    ResultSpec(
+        key="retrieval_residual_scale035_safe_types",
+        label="Train-state residual retrieval, scale 0.35, policy/no-op/wrong-gripper residuals",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p35_safe_types_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="typed tangent scale fine sweep",
+        pending_job="14859503/14859504",
+    ),
+    ResultSpec(
+        key="retrieval_residual_scale045_safe_types",
+        label="Train-state residual retrieval, scale 0.45, policy/no-op/wrong-gripper residuals",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p45_safe_types_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="typed tangent scale fine sweep",
+        pending_job="14859505/14859506",
+    ),
+    ResultSpec(
+        key="retrieval_residual_scale060_safe_types",
+        label="Train-state residual retrieval, scale 0.60, policy/no-op/wrong-gripper residuals",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p60_safe_types_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="typed tangent scale fine sweep",
+        pending_job="14859507/14859508",
+    ),
+    ResultSpec(
+        key="retrieval_residual_scale070_safe_types",
+        label="Train-state residual retrieval, scale 0.70, policy/no-op/wrong-gripper residuals",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_scale0p70_safe_types_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="typed tangent scale fine sweep",
+        pending_job="14859509/14859510",
+    ),
     ResultSpec(
         key="retrieval_residual_scale050_type_success010",
         label="Train-state residual retrieval, scale 0.50, train family success >= 0.10",