Auto-sync: 2026-06-29 01:53:54 (part 2)

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025_summary.json

@@ -0,0 +1,322 @@
+{
+  "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
+  "objective": "near_miss_policy_bc5",
+  "out_name": "policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025.json",
+  "num_completed": 3,
+  "baseline_h4_policy_success": 0.2967,
+  "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
+  "mean_success": 0.35188405797101446,
+  "std_success": 0.01458521231931493,
+  "mean_progress": 0.566551425677729,
+  "mean_action_mse_to_best": 0.3954427905701965,
+  "gain_vs_h4": 0.055184057971014444,
+  "gain_vs_h16_rank_checkpoint": 0.054492753623188395,
+  "rows": [
+    {
+      "seed": 0,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 6,
+      "candidate_sigma": 0.0,
+      "selection_margin": 0.2,
+      "prepend_policy_candidate": false,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 4,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.025,
+      "retrieval_residual_scale": 0.4,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "expert",
+      "retrieval_residual_reduce": "mean_by_type",
+      "candidate_type_bonuses": {},
+      "selected_residual_scale_counts": {
+        "0.4": 575
+      },
+      "policy_rollout_success_rate": 0.3443478260869565,
+      "policy_rollout_progress": 0.5502321733202299,
+      "oracle_success_rate": 0.8573913043478261,
+      "action_mse_to_best": 0.38164905525743964,
+      "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": 0.31969772460726426,
+          "expert_success_rate": 0.8865979381443299,
+          "num_groups": 97,
+          "oracle_success_rate": 0.9278350515463918,
+          "policy_expert_regret": 1.0783334703175063,
+          "policy_oracle_regret": 1.1047489480259491,
+          "policy_rollout_progress": 0.5714634232914325,
+          "policy_rollout_success_rate": 0.2268041237113402,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.30271800868929577,
+          "expert_success_rate": 0.9375,
+          "num_groups": 208,
+          "oracle_success_rate": 0.9471153846153846,
+          "policy_expert_regret": 0.9680723937658163,
+          "policy_oracle_regret": 0.9814952611923218,
+          "policy_rollout_progress": 0.6140019423686541,
+          "policy_rollout_success_rate": 0.33653846153846156,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6002476956550177,
+          "expert_success_rate": 0.19480519480519481,
+          "num_groups": 77,
+          "oracle_success_rate": 0.36363636363636365,
+          "policy_expert_regret": 0.246578575468731,
+          "policy_oracle_regret": 0.4201346135625688,
+          "policy_rollout_progress": 0.3096777588649133,
+          "policy_rollout_success_rate": 0.2077922077922078,
+          "restore_max_error": 2.3096799850463867e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.36465501579198434,
+          "expert_success_rate": 0.8617021276595744,
+          "num_groups": 94,
+          "oracle_success_rate": 0.9893617021276596,
+          "policy_expert_regret": 0.5508309629686335,
+          "policy_oracle_regret": 0.647163599729538,
+          "policy_rollout_progress": 0.6932619321853557,
+          "policy_rollout_success_rate": 0.6595744680851063,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.45429835988754275,
+          "expert_success_rate": 0.6767676767676768,
+          "num_groups": 99,
+          "oracle_success_rate": 0.8585858585858586,
+          "policy_expert_regret": 0.9403071221378114,
+          "policy_oracle_regret": 1.10289698026397,
+          "policy_rollout_progress": 0.4467407528198127,
+          "policy_rollout_success_rate": 0.2828282828282828,
+          "restore_max_error": 2.384185791015625e-07
+        }
+      }
+    },
+    {
+      "seed": 1,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 6,
+      "candidate_sigma": 0.0,
+      "selection_margin": 0.2,
+      "prepend_policy_candidate": false,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 4,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.025,
+      "retrieval_residual_scale": 0.4,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "expert",
+      "retrieval_residual_reduce": "mean_by_type",
+      "candidate_type_bonuses": {},
+      "selected_residual_scale_counts": {
+        "0.4": 575
+      },
+      "policy_rollout_success_rate": 0.3426086956521739,
+      "policy_rollout_progress": 0.5665369803134513,
+      "oracle_success_rate": 0.8695652173913043,
+      "action_mse_to_best": 0.38807988138464483,
+      "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": 0.32591217734018524,
+          "expert_success_rate": 0.8584070796460177,
+          "num_groups": 113,
+          "oracle_success_rate": 0.9380530973451328,
+          "policy_expert_regret": 0.9873926745051831,
+          "policy_oracle_regret": 1.0665947045900126,
+          "policy_rollout_progress": 0.6355170211960784,
+          "policy_rollout_success_rate": 0.23008849557522124,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.31969838518568355,
+          "expert_success_rate": 0.9402173913043478,
+          "num_groups": 184,
+          "oracle_success_rate": 0.9456521739130435,
+          "policy_expert_regret": 1.070886625169331,
+          "policy_oracle_regret": 1.0709106296442612,
+          "policy_rollout_progress": 0.5743152240554438,
+          "policy_rollout_success_rate": 0.29347826086956524,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.5715771112591028,
+          "expert_success_rate": 0.25,
+          "num_groups": 76,
+          "oracle_success_rate": 0.40789473684210525,
+          "policy_expert_regret": 0.30013844498286124,
+          "policy_oracle_regret": 0.5234715515061429,
+          "policy_rollout_progress": 0.28913191000097677,
+          "policy_rollout_success_rate": 0.18421052631578946,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.35765871029716356,
+          "expert_success_rate": 0.8198198198198198,
+          "num_groups": 111,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.32971009934270706,
+          "policy_oracle_regret": 0.38007361851296984,
+          "policy_rollout_progress": 0.8181245796852283,
+          "policy_rollout_success_rate": 0.8018018018018018,
+          "restore_max_error": 2.3795291781425476e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.48739973169106704,
+          "expert_success_rate": 0.6923076923076923,
+          "num_groups": 91,
+          "oracle_success_rate": 0.8571428571428571,
+          "policy_expert_regret": 1.1409930893025555,
+          "policy_oracle_regret": 1.3006549364590383,
+          "policy_rollout_progress": 0.38995039070045556,
+          "policy_rollout_success_rate": 0.15384615384615385,
+          "restore_max_error": 1.9371509552001953e-07
+        }
+      }
+    },
+    {
+      "seed": 2,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 6,
+      "candidate_sigma": 0.0,
+      "selection_margin": 0.2,
+      "prepend_policy_candidate": false,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 4,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.025,
+      "retrieval_residual_scale": 0.4,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "expert",
+      "retrieval_residual_reduce": "mean_by_type",
+      "candidate_type_bonuses": {},
+      "selected_residual_scale_counts": {
+        "0.4": 575
+      },
+      "policy_rollout_success_rate": 0.36869565217391304,
+      "policy_rollout_progress": 0.5828851233995058,
+      "oracle_success_rate": 0.8765217391304347,
+      "action_mse_to_best": 0.41659943506850494,
+      "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": 0.35488338287783944,
+          "expert_success_rate": 0.8229166666666666,
+          "num_groups": 96,
+          "oracle_success_rate": 0.9270833333333334,
+          "policy_expert_regret": 0.8476849501021206,
+          "policy_oracle_regret": 0.9539546399998168,
+          "policy_rollout_progress": 0.6463205499264101,
+          "policy_rollout_success_rate": 0.3333333333333333,
+          "restore_max_error": 1.955777406692505e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.335480088330429,
+          "expert_success_rate": 0.9444444444444444,
+          "num_groups": 198,
+          "oracle_success_rate": 0.9595959595959596,
+          "policy_expert_regret": 0.9807514474924767,
+          "policy_oracle_regret": 0.9878348980211851,
+          "policy_rollout_progress": 0.6357372662409989,
+          "policy_rollout_success_rate": 0.3383838383838384,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6283310380246904,
+          "expert_success_rate": 0.24444444444444444,
+          "num_groups": 90,
+          "oracle_success_rate": 0.4666666666666667,
+          "policy_expert_regret": 0.3140720418619821,
+          "policy_oracle_regret": 0.5426390113998827,
+          "policy_rollout_progress": 0.31892168570862445,
+          "policy_rollout_success_rate": 0.2111111111111111,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3851195577495169,
+          "expert_success_rate": 0.8514851485148515,
+          "num_groups": 101,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.3620445999768701,
+          "policy_oracle_regret": 0.41694685138098087,
+          "policy_rollout_progress": 0.8008749307972369,
+          "policy_rollout_success_rate": 0.7821782178217822,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.4844882684863276,
+          "expert_success_rate": 0.7666666666666667,
+          "num_groups": 90,
+          "oracle_success_rate": 0.9111111111111111,
+          "policy_expert_regret": 1.1607355892658233,
+          "policy_oracle_regret": 1.2970718792743152,
+          "policy_rollout_progress": 0.4182763857973946,
+          "policy_rollout_success_rate": 0.16666666666666666,
+          "restore_max_error": 2.2351741790771484e-07
+        }
+      }
+    }
+  ]
+}

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025_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_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025.json`
+Completed seeds: 3
+Baseline h=4 policy success: 29.67%
+Baseline h=16 rank-checkpoint success: 29.74%
+
+Mean success: 35.19% +/- 1.46%
+Gain vs h=16 rank checkpoint: +5.45%
+Mean progress: 56.66%
+Mean action MSE to best: 0.395
+
+| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual reduce | min type success | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
+|---:|---|---:|---|---:|---|---|---|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|
+| 0 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.025 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 34.43% | 55.02% | 85.74% | 0.382 |
+| 1 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.025 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 34.26% | 56.65% | 86.96% | 0.388 |
+| 2 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.025 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 36.87% | 58.29% | 87.65% | 0.417 |

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02_summary.json

@@ -0,0 +1,322 @@
+{
+  "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
+  "objective": "near_miss_policy_bc5",
+  "out_name": "policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02.json",
+  "num_completed": 3,
+  "baseline_h4_policy_success": 0.2967,
+  "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
+  "mean_success": 0.35246376811594204,
+  "std_success": 0.014164396200429698,
+  "mean_progress": 0.5668095894270369,
+  "mean_action_mse_to_best": 0.3954349665103507,
+  "gain_vs_h4": 0.055763768115942025,
+  "gain_vs_h16_rank_checkpoint": 0.055072463768115976,
+  "rows": [
+    {
+      "seed": 0,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 6,
+      "candidate_sigma": 0.0,
+      "selection_margin": 0.2,
+      "prepend_policy_candidate": false,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 4,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.02,
+      "retrieval_residual_scale": 0.4,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "expert",
+      "retrieval_residual_reduce": "mean_by_type",
+      "candidate_type_bonuses": {},
+      "selected_residual_scale_counts": {
+        "0.4": 575
+      },
+      "policy_rollout_success_rate": 0.34608695652173915,
+      "policy_rollout_progress": 0.5509739407888897,
+      "oracle_success_rate": 0.8573913043478261,
+      "action_mse_to_best": 0.3816288484017486,
+      "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": 0.31969772460726426,
+          "expert_success_rate": 0.8865979381443299,
+          "num_groups": 97,
+          "oracle_success_rate": 0.9278350515463918,
+          "policy_expert_regret": 1.0783334703175063,
+          "policy_oracle_regret": 1.1047489480259491,
+          "policy_rollout_progress": 0.5714634232914325,
+          "policy_rollout_success_rate": 0.2268041237113402,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.30266214839111155,
+          "expert_success_rate": 0.9375,
+          "num_groups": 208,
+          "oracle_success_rate": 0.9471153846153846,
+          "policy_expert_regret": 0.9612141423500501,
+          "policy_oracle_regret": 0.9746370097765555,
+          "policy_rollout_progress": 0.616052501476728,
+          "policy_rollout_success_rate": 0.34134615384615385,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6002476956550177,
+          "expert_success_rate": 0.19480519480519481,
+          "num_groups": 77,
+          "oracle_success_rate": 0.36363636363636365,
+          "policy_expert_regret": 0.246578575468731,
+          "policy_oracle_regret": 0.4201346135625688,
+          "policy_rollout_progress": 0.3096777588649133,
+          "policy_rollout_success_rate": 0.2077922077922078,
+          "restore_max_error": 2.3096799850463867e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.36465501579198434,
+          "expert_success_rate": 0.8617021276595744,
+          "num_groups": 94,
+          "oracle_success_rate": 0.9893617021276596,
+          "policy_expert_regret": 0.5508309629686335,
+          "policy_oracle_regret": 0.647163599729538,
+          "policy_rollout_progress": 0.6932619321853557,
+          "policy_rollout_success_rate": 0.6595744680851063,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.45429835988754275,
+          "expert_success_rate": 0.6767676767676768,
+          "num_groups": 99,
+          "oracle_success_rate": 0.8585858585858586,
+          "policy_expert_regret": 0.9403071221378114,
+          "policy_oracle_regret": 1.10289698026397,
+          "policy_rollout_progress": 0.4467407528198127,
+          "policy_rollout_success_rate": 0.2828282828282828,
+          "restore_max_error": 2.384185791015625e-07
+        }
+      }
+    },
+    {
+      "seed": 1,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 6,
+      "candidate_sigma": 0.0,
+      "selection_margin": 0.2,
+      "prepend_policy_candidate": false,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 4,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.02,
+      "retrieval_residual_scale": 0.4,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "expert",
+      "retrieval_residual_reduce": "mean_by_type",
+      "candidate_type_bonuses": {},
+      "selected_residual_scale_counts": {
+        "0.4": 575
+      },
+      "policy_rollout_success_rate": 0.3426086956521739,
+      "policy_rollout_progress": 0.5665376163727563,
+      "oracle_success_rate": 0.8695652173913043,
+      "action_mse_to_best": 0.38808534023716396,
+      "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": 0.32591217734018524,
+          "expert_success_rate": 0.8584070796460177,
+          "num_groups": 113,
+          "oracle_success_rate": 0.9380530973451328,
+          "policy_expert_regret": 0.9873926745051831,
+          "policy_oracle_regret": 1.0665947045900126,
+          "policy_rollout_progress": 0.6355170211960784,
+          "policy_rollout_success_rate": 0.23008849557522124,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.3197154440998059,
+          "expert_success_rate": 0.9402173913043478,
+          "num_groups": 184,
+          "oracle_success_rate": 0.9456521739130435,
+          "policy_expert_regret": 1.070884637484003,
+          "policy_oracle_regret": 1.070908641958933,
+          "policy_rollout_progress": 0.5743172117407717,
+          "policy_rollout_success_rate": 0.29347826086956524,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.5715771112591028,
+          "expert_success_rate": 0.25,
+          "num_groups": 76,
+          "oracle_success_rate": 0.40789473684210525,
+          "policy_expert_regret": 0.30013844498286124,
+          "policy_oracle_regret": 0.5234715515061429,
+          "policy_rollout_progress": 0.28913191000097677,
+          "policy_rollout_success_rate": 0.18421052631578946,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.35765871029716356,
+          "expert_success_rate": 0.8198198198198198,
+          "num_groups": 111,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.32971009934270706,
+          "policy_oracle_regret": 0.38007361851296984,
+          "policy_rollout_progress": 0.8181245796852283,
+          "policy_rollout_success_rate": 0.8018018018018018,
+          "restore_max_error": 2.3795291781425476e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.48739973169106704,
+          "expert_success_rate": 0.6923076923076923,
+          "num_groups": 91,
+          "oracle_success_rate": 0.8571428571428571,
+          "policy_expert_regret": 1.1409930893025555,
+          "policy_oracle_regret": 1.3006549364590383,
+          "policy_rollout_progress": 0.38995039070045556,
+          "policy_rollout_success_rate": 0.15384615384615385,
+          "restore_max_error": 1.9371509552001953e-07
+        }
+      }
+    },
+    {
+      "seed": 2,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02.json",
+      "num_groups": 575,
+      "selection_mode": "retrieval_residual",
+      "num_candidates": 6,
+      "candidate_sigma": 0.0,
+      "selection_margin": 0.2,
+      "prepend_policy_candidate": false,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 4,
+      "retrieval_metric": "raw",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.02,
+      "retrieval_residual_scale": 0.4,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "expert",
+      "retrieval_residual_reduce": "mean_by_type",
+      "candidate_type_bonuses": {},
+      "selected_residual_scale_counts": {
+        "0.4": 575
+      },
+      "policy_rollout_success_rate": 0.36869565217391304,
+      "policy_rollout_progress": 0.5829172111194645,
+      "oracle_success_rate": 0.8765217391304347,
+      "action_mse_to_best": 0.41659071089213956,
+      "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": 0.35488338287783944,
+          "expert_success_rate": 0.8229166666666666,
+          "num_groups": 96,
+          "oracle_success_rate": 0.9270833333333334,
+          "policy_expert_regret": 0.8476849501021206,
+          "policy_oracle_regret": 0.9539546399998168,
+          "policy_rollout_progress": 0.6463205499264101,
+          "policy_rollout_success_rate": 0.3333333333333333,
+          "restore_max_error": 1.955777406692505e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.33545475296977195,
+          "expert_success_rate": 0.9444444444444444,
+          "num_groups": 198,
+          "oracle_success_rate": 0.9595959595959596,
+          "policy_expert_regret": 0.9806582634572429,
+          "policy_oracle_regret": 0.9877417139859512,
+          "policy_rollout_progress": 0.6358304502762328,
+          "policy_rollout_success_rate": 0.3383838383838384,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6283310380246904,
+          "expert_success_rate": 0.24444444444444444,
+          "num_groups": 90,
+          "oracle_success_rate": 0.4666666666666667,
+          "policy_expert_regret": 0.3140720418619821,
+          "policy_oracle_regret": 0.5426390113998827,
+          "policy_rollout_progress": 0.31892168570862445,
+          "policy_rollout_success_rate": 0.2111111111111111,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3851195577495169,
+          "expert_success_rate": 0.8514851485148515,
+          "num_groups": 101,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.3620445999768701,
+          "policy_oracle_regret": 0.41694685138098087,
+          "policy_rollout_progress": 0.8008749307972369,
+          "policy_rollout_success_rate": 0.7821782178217822,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.4844882684863276,
+          "expert_success_rate": 0.7666666666666667,
+          "num_groups": 90,
+          "oracle_success_rate": 0.9111111111111111,
+          "policy_expert_regret": 1.1607355892658233,
+          "policy_oracle_regret": 1.2970718792743152,
+          "policy_rollout_progress": 0.4182763857973946,
+          "policy_rollout_success_rate": 0.16666666666666666,
+          "restore_max_error": 2.2351741790771484e-07
+        }
+      }
+    }
+  ]
+}

results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02_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_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02.json`
+Completed seeds: 3
+Baseline h=4 policy success: 29.67%
+Baseline h=16 rank-checkpoint success: 29.74%
+
+Mean success: 35.25% +/- 1.42%
+Gain vs h=16 rank checkpoint: +5.51%
+Mean progress: 56.68%
+Mean action MSE to best: 0.395
+
+| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual reduce | min type success | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | action MSE |
+|---:|---|---:|---|---:|---|---|---|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|
+| 0 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.020 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 34.61% | 55.10% | 85.74% | 0.382 |
+| 1 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.020 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 34.26% | 56.65% | 86.96% | 0.388 |
+| 2 | retrieval_residual | 6 | no | 4 | raw | expert | mean_by_type | 0.00 | 0.00 | 0.000 | 0.020 | 0.40 | none | 0.200 | 0.00 | 0 | 0.00 | 36.87% | 58.29% | 87.65% | 0.417 |

results/paper_analysis.json

@@ -1,6 +1,6 @@
 {
   "best_clean_key": "residual_k4_consensus_noopbonus003",
-  "generated_utc": "2026-06-29T05:34:13+00:00",
+  "generated_utc": "2026-06-29T05:52:54+00:00",
   "mechanism_gap": {
     "best_clean_vs_direct_same_ckpt": 0.06956521739130428,
     "best_clean_vs_h16": 0.05507246376811592,
@@ -1505,19 +1505,235 @@
       "std_success": 0.01358304291462029
     },
     "residual_k4_consensus_srcscorebonus0015": {
+      "ci95_success": 0.03518915067855206,
       "label": "K4 mean-by-type tangent consensus, source-score bonus 0.015",
-      "missing": true,
-      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015_summary.json"
+      "mean_action_mse_to_best": 0.3954103358281587,
+      "mean_progress": 0.5668091635216157,
+      "mean_success": 0.352463768115942,
+      "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": 0.3244027510331859,
+          "std_success": 0.026822320875242476
+        },
+        "PullCube-v1": {
+          "mean_num_groups": 81.0,
+          "mean_success": 0.20103794840636946,
+          "std_success": 0.014667153114049147
+        },
+        "PushCube-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.7478514959028968,
+          "std_success": 0.07707721123855243
+        },
+        "StackCube-v1": {
+          "mean_num_groups": 93.33333333333333,
+          "mean_success": 0.2011137011137011,
+          "std_success": 0.07105663963231741
+        }
+      },
+      "seed_action_mse_to_best": {
+        "0": 0.3816270524878865,
+        "1": 0.38801324410445015,
+        "2": 0.41659071089213956
+      },
+      "seed_progress": {
+        "0": 0.5509786044636175,
+        "1": 0.5665316749817652,
+        "2": 0.5829172111194645
+      },
+      "seed_success": {
+        "0": 0.34608695652173915,
+        "1": 0.3426086956521739,
+        "2": 0.36869565217391304
+      },
+      "selected_candidate_type_counts": {
+        "retrieval_residual_policy_residual": 1654,
+        "retrieval_residual_residual_no_op": 44,
+        "retrieval_residual_residual_wrong_gripper": 27
+      },
+      "selected_residual_scale_counts": {
+        "0.4": 1725
+      },
+      "selected_type_outcomes": {
+        "retrieval_residual_policy_residual": {
+          "count": 1654.0,
+          "mean_progress": 0.5609882382687458,
+          "success_rate": 0.34461910519951633
+        },
+        "retrieval_residual_residual_no_op": {
+          "count": 44.0,
+          "mean_progress": 0.7778678314590995,
+          "success_rate": 0.5909090909090909
+        },
+        "retrieval_residual_residual_wrong_gripper": {
+          "count": 27.0,
+          "mean_progress": 0.5794472738548562,
+          "success_rate": 0.4444444444444444
+        }
+      },
+      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p015_summary.json",
+      "std_success": 0.0141643962004297
     },
     "residual_k4_consensus_srcscorebonus002": {
+      "ci95_success": 0.03518915067855206,
       "label": "K4 mean-by-type tangent consensus, source-score bonus 0.02",
-      "missing": true,
-      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02_summary.json"
+      "mean_action_mse_to_best": 0.3954349665103507,
+      "mean_progress": 0.5668095894270369,
+      "mean_success": 0.352463768115942,
+      "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": 0.3244027510331859,
+          "std_success": 0.026822320875242476
+        },
+        "PullCube-v1": {
+          "mean_num_groups": 81.0,
+          "mean_success": 0.20103794840636946,
+          "std_success": 0.014667153114049147
+        },
+        "PushCube-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.7478514959028968,
+          "std_success": 0.07707721123855243
+        },
+        "StackCube-v1": {
+          "mean_num_groups": 93.33333333333333,
+          "mean_success": 0.2011137011137011,
+          "std_success": 0.07105663963231741
+        }
+      },
+      "seed_action_mse_to_best": {
+        "0": 0.3816288484017486,
+        "1": 0.38808534023716396,
+        "2": 0.41659071089213956
+      },
+      "seed_progress": {
+        "0": 0.5509739407888897,
+        "1": 0.5665376163727563,
+        "2": 0.5829172111194645
+      },
+      "seed_success": {
+        "0": 0.34608695652173915,
+        "1": 0.3426086956521739,
+        "2": 0.36869565217391304
+      },
+      "selected_candidate_type_counts": {
+        "retrieval_residual_policy_residual": 1650,
+        "retrieval_residual_residual_no_op": 47,
+        "retrieval_residual_residual_wrong_gripper": 28
+      },
+      "selected_residual_scale_counts": {
+        "0.4": 1725
+      },
+      "selected_type_outcomes": {
+        "retrieval_residual_policy_residual": {
+          "count": 1650.0,
+          "mean_progress": 0.5604633099658212,
+          "success_rate": 0.34424242424242424
+        },
+        "retrieval_residual_residual_no_op": {
+          "count": 47.0,
+          "mean_progress": 0.7731277430628208,
+          "success_rate": 0.574468085106383
+        },
+        "retrieval_residual_residual_wrong_gripper": {
+          "count": 28.0,
+          "mean_progress": 0.5944670140743256,
+          "success_rate": 0.4642857142857143
+        }
+      },
+      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p02_summary.json",
+      "std_success": 0.0141643962004297
     },
     "residual_k4_consensus_srcscorebonus0025": {
+      "ci95_success": 0.03623460024137681,
       "label": "K4 mean-by-type tangent consensus, source-score bonus 0.025",
-      "missing": true,
-      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025_summary.json"
+      "mean_action_mse_to_best": 0.3954427905701965,
+      "mean_progress": 0.566551425677729,
+      "mean_success": 0.3518840579710145,
+      "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": 0.3228001869306218,
+          "std_success": 0.025410290531060663
+        },
+        "PullCube-v1": {
+          "mean_num_groups": 81.0,
+          "mean_success": 0.20103794840636946,
+          "std_success": 0.014667153114049147
+        },
+        "PushCube-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.7478514959028968,
+          "std_success": 0.07707721123855243
+        },
+        "StackCube-v1": {
+          "mean_num_groups": 93.33333333333333,
+          "mean_success": 0.2011137011137011,
+          "std_success": 0.07105663963231741
+        }
+      },
+      "seed_action_mse_to_best": {
+        "0": 0.38164905525743964,
+        "1": 0.38807988138464483,
+        "2": 0.41659943506850494
+      },
+      "seed_progress": {
+        "0": 0.5502321733202299,
+        "1": 0.5665369803134513,
+        "2": 0.5828851233995058
+      },
+      "seed_success": {
+        "0": 0.3443478260869565,
+        "1": 0.3426086956521739,
+        "2": 0.36869565217391304
+      },
+      "selected_candidate_type_counts": {
+        "retrieval_residual_policy_residual": 1644,
+        "retrieval_residual_residual_no_op": 53,
+        "retrieval_residual_residual_wrong_gripper": 28
+      },
+      "selected_residual_scale_counts": {
+        "0.4": 1725
+      },
+      "selected_type_outcomes": {
+        "retrieval_residual_policy_residual": {
+          "count": 1644.0,
+          "mean_progress": 0.559960644420736,
+          "success_rate": 0.3442822384428224
+        },
+        "retrieval_residual_residual_no_op": {
+          "count": 53.0,
+          "mean_progress": 0.7562421409870094,
+          "success_rate": 0.5283018867924528
+        },
+        "retrieval_residual_residual_wrong_gripper": {
+          "count": 28.0,
+          "mean_progress": 0.5944670140743256,
+          "success_rate": 0.4642857142857143
+        }
+      },
+      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4s040_safe_margin0p20_mean_by_type_srcscorebonus0p025_summary.json",
+      "std_success": 0.01458521231931493
     },
     "residual_k4_consensus_wgbonus003": {
       "ci95_success": 0.03271496576240179,

results/paper_analysis.md

@@ -1,6 +1,6 @@
 # Paper Analysis
 
-Generated: `2026-06-29T05:34:13+00:00`
+Generated: `2026-06-29T05:52:54+00:00`
 
 ## Main Seed Statistics
 
@@ -28,9 +28,9 @@ Generated: `2026-06-29T05:34:13+00:00`
 | residual_k4_consensus_noopbonus003_srcprog075 | K4 mean-by-type tangent consensus, no-op bonus 0.03, source progress >= 0.75 | 3 | 34.72% +/- 1.42 | +/- 3.52 | 56.44% | 0.396 | +4.99 pp |
 | residual_k4_consensus_srcprogbonus003 | K4 mean-by-type tangent consensus, source-progress bonus 0.03 | 3 | 35.25% +/- 1.42 | +/- 3.52 | 56.68% | 0.395 | +5.51 pp |
 | residual_k4_consensus_srcprogbonus005 | K4 mean-by-type tangent consensus, source-progress bonus 0.05 | 3 | 35.13% +/- 1.36 | +/- 3.37 | 56.65% | 0.396 | +5.39 pp |
-| residual_k4_consensus_srcscorebonus0015 | K4 mean-by-type tangent consensus, source-score bonus 0.015 | 0 | missing | missing | missing | missing | missing |
-| residual_k4_consensus_srcscorebonus002 | K4 mean-by-type tangent consensus, source-score bonus 0.02 | 0 | missing | missing | missing | missing | missing |
-| residual_k4_consensus_srcscorebonus0025 | K4 mean-by-type tangent consensus, source-score bonus 0.025 | 0 | missing | missing | missing | missing | missing |
+| residual_k4_consensus_srcscorebonus0015 | K4 mean-by-type tangent consensus, source-score bonus 0.015 | 3 | 35.25% +/- 1.42 | +/- 3.52 | 56.68% | 0.395 | +5.51 pp |
+| residual_k4_consensus_srcscorebonus002 | K4 mean-by-type tangent consensus, source-score bonus 0.02 | 3 | 35.25% +/- 1.42 | +/- 3.52 | 56.68% | 0.395 | +5.51 pp |
+| residual_k4_consensus_srcscorebonus0025 | K4 mean-by-type tangent consensus, source-score bonus 0.025 | 3 | 35.19% +/- 1.46 | +/- 3.62 | 56.66% | 0.395 | +5.45 pp |
 | residual_taskrelative_k4_consensus_noopbonus003 | K4 task-relative tangent consensus, no-op bonus 0.03 | 3 | 34.43% +/- 1.22 | +/- 3.02 | 56.19% | 0.399 | +4.70 pp |
 | residual_k4_consensus_noopbonus001 | K4 mean-by-type tangent consensus, no-op bonus 0.01 | 3 | 35.19% +/- 1.32 | +/- 3.27 | 56.63% | 0.395 | +5.45 pp |
 | residual_k4_consensus_noopbonus002 | K4 mean-by-type tangent consensus, no-op bonus 0.02 | 3 | 35.19% +/- 1.32 | +/- 3.27 | 56.64% | 0.395 | +5.45 pp |

results/paper_core_results.md

@@ -43,6 +43,8 @@ and the remaining clean-to-same-state proposal gap is `+21.74 pp`.
 | K4 mean-by-type residual retrieval + no-op prior 0.03 | No | No | 35.25% | +5.51 pp | Current best clean diagnostic; 0.025-0.035 forms a small plateau that nudges high-value no-op residuals without changing the core proposal family |
 | K4 mean-by-type residual retrieval + source-progress prior 0.03 | No | No | 35.25% | +5.51 pp | Ties the current best exactly without a hand typed no-op prior; train-measured source progress can replace but not improve the typed prior |
 | K4 mean-by-type residual retrieval + source-progress prior 0.05 | No | No | 35.13% | +5.39 pp | A stronger measured-progress prior over-selects nonzero residuals and drops below the plateau |
+| K4 mean-by-type residual retrieval + source-score prior 0.015/0.020 | No | No | 35.25% | +5.51 pp | Full train reward score, including terminal success, also replaces the typed prior without improving it |
+| K4 mean-by-type residual retrieval + source-score prior 0.025 | No | No | 35.19% | +5.45 pp | A stronger reward-score prior drops below the plateau |
 | K4 mean-by-type residual retrieval + no-op prior + source-progress gate | No | No | 34.72-35.19% | +4.99-5.45 pp | Train-source progress viability is a near-tie/negative gate; soft threshold 0.25 is closest, while stricter thresholds over-abstain below the no-op plateau |
 | K4 task-relative mean-by-type residual retrieval + no-op prior 0.03 | No | No | 34.43% | +4.70 pp | Task-relative target/reference pose retrieval underperforms the raw-metric no-op plateau |
 | K4 kernel-weighted residual consensus + no-op prior 0.03 | No | No | 35.13-35.19% | +5.39-5.45 pp | Distance-weighted tangent interpolation is plausible but does not beat equal mean-consensus no-op plateau |
@@ -84,7 +86,7 @@ Suggested main-table rows:
 11. K2 train-state residual retrieval, typed safe families + advantage margin 0.20
 12. K4 train-state residual retrieval, mean-by-type tangent consensus
 13. K4 mean-by-type residual retrieval + no-op prior plateau, canonical 0.03
-14. K4 mean-by-type residual retrieval + source-progress prior diagnostics
+14. K4 mean-by-type residual retrieval + source-progress/source-score prior diagnostics
 15. Source-progress viability gate diagnostics
 16. K2/K4 task-relative retrieval metric diagnostics
 17. K4 kernel-weighted residual consensus + no-op prior diagnostics
@@ -104,10 +106,10 @@ Suggested claim:
 > DoVLA-CIL is not a better behavior-cloning policy; it is a local counterfactual action
 > selection rule. Deployment-clean K4 consensus residual transport with advantage
 > abstention and either a small typed no-op prior or an equivalent train-source
-> progress prior gives the strongest clean gain so far, while ungated KNN residual
+> progress/reward-score prior gives the strongest clean gain so far, while ungated KNN residual
 > retrieval, field-gradient ascent, broader non-expert BC targets, field-teacher/tangent distillation, z-score/task-relative retrieval,
 > train-family reliability priors, policy-relative anchoring, residual+Gaussian hybrids,
-> source-progress viability gates, overly strong progress priors, tangent consensus, kernel-weighted tangent interpolation, field-softmax tangent barycenters, tangent ray-search, wrong-gripper typed priors, and same-state policy-baseline fallback fail to improve the main rows.
+> source-progress viability gates, overly strong train-outcome priors, tangent consensus, kernel-weighted tangent interpolation, field-softmax tangent barycenters, tangent ray-search, wrong-gripper typed priors, and same-state policy-baseline fallback fail to improve the main rows.
 > 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

@@ -39,6 +39,7 @@ when queried on proposal geometry that matches those local counterfactuals.
 | Task-relative actor-pose retrieval metric does not improve tangent transfer | K2 task-relative residual retrieval reaches 34.26% vs raw K2 35.01%; K4 task-relative mean-by-type + no-op reaches 34.43% vs raw K4 35.25% | Negative diagnostic |
 | Train-source progress viability is too blunt a residual gate | source-progress thresholds 0.25/0.50/0.75 reach 35.19%/34.96%/34.72%, below the unfiltered no-op plateau at 35.25% | Negative/near-tie diagnostic |
 | Continuous train-source progress prior can replace the typed no-op prior but not improve it | source-progress bonus 0.03 ties the 35.25% best exactly; bonus 0.05 drops to 35.13% | Cleaner tie diagnostic |
+| Full train-source reward-score prior also ties but does not improve the clean best | source-score bonuses 0.015/0.020 tie 35.25%; 0.025 drops to 35.19% | Cleaner tie diagnostic |
 | Train-split residual family reliability does not recover the typed mask | after fixing threshold pass-through, scale-0.35 thresholds 0.10/0.25 reach 33.33%/33.28%, below typed safe residuals | Negative diagnostic |
 | Residual-tangent distillation does not solve clean proposal generation | aligned allmap tangent student reaches 28.87% despite low pseudo-target BC loss | Negative diagnostic |
 | Policy-relative residual anchoring does not improve the bridge | policy-anchor safe residual transport ties 33.74% rather than improving expert-anchor residuals | Negative diagnostic |
@@ -65,23 +66,24 @@ clean proposal result, the intended main rows are:
 14. K4 mean-by-type tangent consensus: 34.96%
 15. K4 mean-by-type tangent consensus + typed no-op prior 0.025-0.035: 35.25%
 16. K4 mean-by-type tangent consensus + train-source progress prior: 35.25% at bonus 0.03; 35.13% at bonus 0.05
-17. Source-progress viability gates: 35.19% / 34.96% / 34.72% for thresholds 0.25 / 0.50 / 0.75
-18. K4 kernel-weighted tangent consensus / + no-op prior: 34.96% / 35.19%
-19. K4 field-softmax tangent transport / best margin sweep: 34.96% / 35.19%
-20. Wrong-gripper prior / no-op+wrong-gripper prior: 35.19% / 35.25%
-21. K2 broad tangent ray-search: 34.96%
-22. K1/K2 tight tangent ray-search: 34.84% / 34.84%
-23. K4 tight tangent ray-search: 34.55%
-24. Residual-tangent distillation policy: 28.87%
-25. Z-score residual retrieval: 32.23-32.81%
-26. Task-relative residual retrieval metric: 34.26-34.43%
-27. Train-family reliability prior: 33.28-33.33%
-28. Residual+Gaussian hybrid K32/K64: 31.30% / 30.90%
-29. Lattice, near-miss only: 55.94%
-30. Lattice, no expert: 56.99%
-31. Lattice, no expert + policy baseline candidate: 40.70%
-32. Lattice, full: 69.33%
-33. Oracle ceiling: 86.78%
+17. K4 mean-by-type tangent consensus + train-source reward-score prior: 35.25% at bonuses 0.015/0.020; 35.19% at 0.025
+18. Source-progress viability gates: 35.19% / 34.96% / 34.72% for thresholds 0.25 / 0.50 / 0.75
+19. K4 kernel-weighted tangent consensus / + no-op prior: 34.96% / 35.19%
+20. K4 field-softmax tangent transport / best margin sweep: 34.96% / 35.19%
+21. Wrong-gripper prior / no-op+wrong-gripper prior: 35.19% / 35.25%
+22. K2 broad tangent ray-search: 34.96%
+23. K1/K2 tight tangent ray-search: 34.84% / 34.84%
+24. K4 tight tangent ray-search: 34.55%
+25. Residual-tangent distillation policy: 28.87%
+26. Z-score residual retrieval: 32.23-32.81%
+27. Task-relative residual retrieval metric: 34.26-34.43%
+28. Train-family reliability prior: 33.28-33.33%
+29. Residual+Gaussian hybrid K32/K64: 31.30% / 30.90%
+30. Lattice, near-miss only: 55.94%
+31. Lattice, no expert: 56.99%
+32. Lattice, no expert + policy baseline candidate: 40.70%
+33. Lattice, full: 69.33%
+34. Oracle ceiling: 86.78%
 
 ## Novelty Framing
 
@@ -109,9 +111,9 @@ test-time search. The cleaner novelty is:
 
 ## Job Status
 
-Last checked: `2026-06-29 03:58 UTC`. The source-progress bonus-prior batch
+Last checked: `2026-06-29 05:54 UTC`. The source-score bonus-prior batch
 completed after passing CPU/unit smokes, and the paper table/paired analysis now
-include both threshold gates and continuous train-source progress priors.
+include progress and reward-score train-outcome priors.
 
 - `14858328`-`14858333`: completed train-split `field_selected_noexpert_bc5`;
   direct rollout is 26.84%, field-guided best is 27.65%.
@@ -244,6 +246,13 @@ include both threshold gates and continuous train-source progress priors.
   no-op prior. Bonus `0.03` ties the current best at 35.25%; bonus `0.05`
   reaches 35.13%. Summary jobs `14894676`/`14894677` completed; rebuild job
   `14894678` was queued after them.
+- `14897121`/`14897122`: completed unit and CPU rollout smokes for the
+  train-source reward-score bonus path. The unit smoke validates that terminal
+  success contributes to the candidate prior.
+- `14897123`/`14897124`/`14897125`: completed source-score bonus arrays.
+  Bonuses `0.015` and `0.020` tie the current best at 35.25%; bonus `0.025`
+  reaches 35.19%. Summary jobs `14897126`/`14897127`/`14897128` and rebuild job
+  `14897129` completed.
 - `14869627`: completed CPU Apptainer smoke for the new residual scale-grid
   selector. It selected index `3` on a two-residual/two-scale toy case and
   returned the expected action `0.20`, validating the candidate expansion and
@@ -267,9 +276,10 @@ include both threshold gates and continuous train-source progress priors.
   typed no-op prior plateau (0.025-0.035, canonical midpoint 0.03, 35.25%) as
   the current best clean deployment diagnostic, not as a SOTA claim. The
   continuous train-source progress prior at bonus 0.03 ties this result without
-  a hand typed no-op prior, which is a cleaner story hook but still not a higher
-  SOTA number. The completed K2/ray-search rows are near-ties that support the
-  sparse-intervention story.
+  a hand typed no-op prior; train-source reward-score priors at 0.015/0.020 also
+  tie it. This is a cleaner story hook but still not a higher SOTA number. The
+  completed K2/ray-search rows are near-ties that support the sparse-intervention
+  story.
 - Use `results/paper_analysis.md` for paired seed deltas, per-task gaps, and
   selection histograms when writing reviewer-facing tables.
 - Treat z-score and task-relative retrieval metrics, source-progress viability gates,

results/paper_table_status.json

@@ -945,14 +945,14 @@
       "story_role": "train-source reward-score prior for sparse residual transport",
       "fallback_success": null,
       "pending_job": "14897123/14897126",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.35246376811594204,
+      "std_success": 0.014164396200429698,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.055072463768115976
     },
     {
       "key": "retrieval_residual_k4_mean_srcscorebonus002",
@@ -964,14 +964,14 @@
       "story_role": "train-source reward-score prior for sparse residual transport",
       "fallback_success": null,
       "pending_job": "14897124/14897127",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.35246376811594204,
+      "std_success": 0.014164396200429698,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.055072463768115976
     },
     {
       "key": "retrieval_residual_k4_mean_srcscorebonus0025",
@@ -983,14 +983,14 @@
       "story_role": "train-source reward-score prior for sparse residual transport",
       "fallback_success": null,
       "pending_job": "14897125/14897128",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.35188405797101446,
+      "std_success": 0.01458521231931493,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.054492753623188395
     },
     {
       "key": "retrieval_residual_taskrelative_k4_mean_noopbonus003",