Auto-sync: 2026-06-30 10:57:05 (part 3)

results/h16_typedprop_nooponly_sparsehead_prepend_margin0p05_summary.json

@@ -0,0 +1,421 @@
+{
+  "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
+  "objective": "near_miss_policy_bc5_typedprop_nooponly_p2",
+  "out_name": "policy_rollout_proposal_lattice_nooponly_sparsehead_prepend_margin0p05.json",
+  "num_completed": 3,
+  "baseline_h4_policy_success": 0.2967,
+  "baseline_h16_rank_checkpoint_success": 0.29739130434782607,
+  "mean_success": 0.3472463768115942,
+  "std_success": 0.03607735161159794,
+  "mean_progress": 0.5610268774401882,
+  "mean_action_mse_to_best": 0.45465255970136975,
+  "gain_vs_h4": 0.050546376811594185,
+  "gain_vs_h16_rank_checkpoint": 0.049855072463768135,
+  "rows": [
+    {
+      "seed": 0,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5_typedprop_nooponly_p2/seed_0/policy_rollout_proposal_lattice_nooponly_sparsehead_prepend_margin0p05.json",
+      "num_groups": 575,
+      "selection_mode": "proposal_lattice",
+      "num_candidates": 2,
+      "candidate_sigma": 0.0,
+      "proposal_types": [
+        "no_op"
+      ],
+      "proposal_lattice_types": [
+        "no_op"
+      ],
+      "selection_margin": 0.05,
+      "prepend_policy_candidate": true,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 0,
+      "retrieval_metric": "none",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_type_success_bonus_scale": 0.0,
+      "retrieval_residual_consensus_penalty_scale": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_min_source_advantage": -1000000000.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.0,
+      "retrieval_residual_source_advantage_bonus_scale": 0.0,
+      "retrieval_residual_composite_l2_penalty_scale": 0.0,
+      "retrieval_residual_scale": 0.0,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "none",
+      "retrieval_residual_direction": "none",
+      "retrieval_residual_reduce": "none",
+      "retrieval_residual_challenger_types": [],
+      "retrieval_residual_challenger_scales": [],
+      "retrieval_residual_challenger_margin": 0.0,
+      "lattice_exclude_types": [],
+      "candidate_type_bonuses": {},
+      "candidate_type_bonus_components": false,
+      "candidate_oracle_rollouts": 0,
+      "candidate_oracle_unique_tolerance": 1e-06,
+      "candidate_oracle_success_rate": null,
+      "candidate_oracle_progress": null,
+      "candidate_oracle_score": null,
+      "candidate_oracle_regret": null,
+      "candidate_oracle_score_gain_over_selected": null,
+      "candidate_oracle_unique_count": null,
+      "candidate_oracle_improvement_rate": null,
+      "candidate_oracle_selected_branch_success_rate": null,
+      "candidate_oracle_selected_branch_progress": null,
+      "candidate_oracle_type_counts": {},
+      "candidate_oracle_best_branch_rank": null,
+      "candidate_oracle_best_branch_rank_counts": {},
+      "candidate_oracle_branch_success_rates": [],
+      "candidate_oracle_branch_progress": [],
+      "candidate_oracle_branch_score_gains_over_selected": [],
+      "selected_residual_scale_counts": {},
+      "policy_rollout_success_rate": 0.31478260869565217,
+      "policy_rollout_progress": 0.5357955083963664,
+      "oracle_success_rate": 0.8573913043478261,
+      "action_mse_to_best": 0.4179149465454382,
+      "best_policy_val": {
+        "bc_loss": 0.1388491830892033,
+        "field_effect_loss": 0.009226797407286035,
+        "field_potential_loss": 0.2551382887694571,
+        "field_preference_loss": 0.516539043850369,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.1904102149936888,
+        "proposal_loss": 3.2502353709181384e-06,
+        "rank_acc": 0.8304381403658125,
+        "rank_loss": 0.516533311870363,
+        "regret_mae": 0.3661145286427604,
+        "success_accuracy": 0.8777371744314829,
+        "total_loss": 1.557318925857544
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.31925635191986396,
+          "expert_success_rate": 0.8865979381443299,
+          "num_groups": 97,
+          "oracle_success_rate": 0.9278350515463918,
+          "policy_expert_regret": 1.24966243293482,
+          "policy_oracle_regret": 1.2892768885364239,
+          "policy_rollout_progress": 0.47971898793559714,
+          "policy_rollout_success_rate": 0.13402061855670103,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.3708913918109333,
+          "expert_success_rate": 0.9375,
+          "num_groups": 208,
+          "oracle_success_rate": 0.9471153846153846,
+          "policy_expert_regret": 1.0504980465182319,
+          "policy_oracle_regret": 1.0571646205722713,
+          "policy_rollout_progress": 0.5796610939746293,
+          "policy_rollout_success_rate": 0.28846153846153844,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6472928293046548,
+          "expert_success_rate": 0.19480519480519481,
+          "num_groups": 77,
+          "oracle_success_rate": 0.36363636363636365,
+          "policy_expert_regret": 0.27947029582672306,
+          "policy_oracle_regret": 0.5762322676162441,
+          "policy_rollout_progress": 0.2556867856387194,
+          "policy_rollout_success_rate": 0.1038961038961039,
+          "restore_max_error": 2.3096799850463867e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.365135474804234,
+          "expert_success_rate": 0.8617021276595744,
+          "num_groups": 94,
+          "oracle_success_rate": 0.9893617021276596,
+          "policy_expert_regret": 0.38729391040954186,
+          "policy_oracle_regret": 0.48438419148008877,
+          "policy_rollout_progress": 0.7709349574560814,
+          "policy_rollout_success_rate": 0.7446808510638298,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.48508622366332954,
+          "expert_success_rate": 0.6767676767676768,
+          "num_groups": 99,
+          "oracle_success_rate": 0.8585858585858586,
+          "policy_expert_regret": 0.8837137747593601,
+          "policy_oracle_regret": 1.0362644341557916,
+          "policy_rollout_progress": 0.49317575208466463,
+          "policy_rollout_success_rate": 0.30303030303030304,
+          "restore_max_error": 2.384185791015625e-07
+        }
+      }
+    },
+    {
+      "seed": 1,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5_typedprop_nooponly_p2/seed_1/policy_rollout_proposal_lattice_nooponly_sparsehead_prepend_margin0p05.json",
+      "num_groups": 575,
+      "selection_mode": "proposal_lattice",
+      "num_candidates": 2,
+      "candidate_sigma": 0.0,
+      "proposal_types": [
+        "no_op"
+      ],
+      "proposal_lattice_types": [
+        "no_op"
+      ],
+      "selection_margin": 0.05,
+      "prepend_policy_candidate": true,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 0,
+      "retrieval_metric": "none",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_type_success_bonus_scale": 0.0,
+      "retrieval_residual_consensus_penalty_scale": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_min_source_advantage": -1000000000.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.0,
+      "retrieval_residual_source_advantage_bonus_scale": 0.0,
+      "retrieval_residual_composite_l2_penalty_scale": 0.0,
+      "retrieval_residual_scale": 0.0,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "none",
+      "retrieval_residual_direction": "none",
+      "retrieval_residual_reduce": "none",
+      "retrieval_residual_challenger_types": [],
+      "retrieval_residual_challenger_scales": [],
+      "retrieval_residual_challenger_margin": 0.0,
+      "lattice_exclude_types": [],
+      "candidate_type_bonuses": {},
+      "candidate_type_bonus_components": false,
+      "candidate_oracle_rollouts": 0,
+      "candidate_oracle_unique_tolerance": 1e-06,
+      "candidate_oracle_success_rate": null,
+      "candidate_oracle_progress": null,
+      "candidate_oracle_score": null,
+      "candidate_oracle_regret": null,
+      "candidate_oracle_score_gain_over_selected": null,
+      "candidate_oracle_unique_count": null,
+      "candidate_oracle_improvement_rate": null,
+      "candidate_oracle_selected_branch_success_rate": null,
+      "candidate_oracle_selected_branch_progress": null,
+      "candidate_oracle_type_counts": {},
+      "candidate_oracle_best_branch_rank": null,
+      "candidate_oracle_best_branch_rank_counts": {},
+      "candidate_oracle_branch_success_rates": [],
+      "candidate_oracle_branch_progress": [],
+      "candidate_oracle_branch_score_gains_over_selected": [],
+      "selected_residual_scale_counts": {},
+      "policy_rollout_success_rate": 0.3408695652173913,
+      "policy_rollout_progress": 0.5624661266301637,
+      "oracle_success_rate": 0.8695652173913043,
+      "action_mse_to_best": 0.441326511521981,
+      "best_policy_val": {
+        "bc_loss": 0.1322652217414644,
+        "field_effect_loss": 0.010026000388380554,
+        "field_potential_loss": 0.2885795517100228,
+        "field_preference_loss": 0.5062332368559308,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.1950002759695053,
+        "proposal_loss": 3.3075696895442284e-06,
+        "rank_acc": 0.8296620117293464,
+        "rank_loss": 0.5062223523855209,
+        "regret_mae": 0.3615676628218757,
+        "success_accuracy": 0.8715242776605818,
+        "total_loss": 1.5516093638208177
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.32245841615347076,
+          "expert_success_rate": 0.8584070796460177,
+          "num_groups": 113,
+          "oracle_success_rate": 0.9380530973451328,
+          "policy_expert_regret": 0.9845319982387323,
+          "policy_oracle_regret": 1.054598491671866,
+          "policy_rollout_progress": 0.6185233694525947,
+          "policy_rollout_success_rate": 0.24778761061946902,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.45968855023787364,
+          "expert_success_rate": 0.9402173913043478,
+          "num_groups": 184,
+          "oracle_success_rate": 0.9456521739130435,
+          "policy_expert_regret": 1.0760140888463787,
+          "policy_oracle_regret": 1.0760140888463787,
+          "policy_rollout_progress": 0.585793553512422,
+          "policy_rollout_success_rate": 0.27717391304347827,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.5957562092137203,
+          "expert_success_rate": 0.25,
+          "num_groups": 76,
+          "oracle_success_rate": 0.40789473684210525,
+          "policy_expert_regret": 0.278456690231044,
+          "policy_oracle_regret": 0.4177438636849585,
+          "policy_rollout_progress": 0.3457642624546823,
+          "policy_rollout_success_rate": 0.21052631578947367,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.3573058122383045,
+          "expert_success_rate": 0.8198198198198198,
+          "num_groups": 111,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.46710044588591604,
+          "policy_oracle_regret": 0.5667114986761196,
+          "policy_rollout_progress": 0.7305857986211777,
+          "policy_rollout_success_rate": 0.7027027027027027,
+          "restore_max_error": 2.3795291781425476e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.5253168439267428,
+          "expert_success_rate": 0.6923076923076923,
+          "num_groups": 91,
+          "oracle_success_rate": 0.8571428571428571,
+          "policy_expert_regret": 0.9979727482730216,
+          "policy_oracle_regret": 1.1575737043723955,
+          "policy_rollout_progress": 0.42160176510339253,
+          "policy_rollout_success_rate": 0.25274725274725274,
+          "restore_max_error": 1.9371509552001953e-07
+        }
+      }
+    },
+    {
+      "seed": 2,
+      "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5_typedprop_nooponly_p2/seed_2/policy_rollout_proposal_lattice_nooponly_sparsehead_prepend_margin0p05.json",
+      "num_groups": 575,
+      "selection_mode": "proposal_lattice",
+      "num_candidates": 2,
+      "candidate_sigma": 0.0,
+      "proposal_types": [
+        "no_op"
+      ],
+      "proposal_lattice_types": [
+        "no_op"
+      ],
+      "selection_margin": 0.05,
+      "prepend_policy_candidate": true,
+      "field_optim_steps": 0,
+      "field_optim_step_size": 0.0,
+      "field_optim_trust_radius": 0.0,
+      "field_optim_l2_penalty": 0.0,
+      "retrieval_neighbors": 0,
+      "retrieval_metric": "none",
+      "retrieval_type_min_success": 0.0,
+      "retrieval_type_success_bonus_scale": 0.0,
+      "retrieval_residual_consensus_penalty_scale": 0.0,
+      "retrieval_residual_min_source_progress": 0.0,
+      "retrieval_residual_min_source_advantage": -1000000000.0,
+      "retrieval_residual_source_progress_bonus_scale": 0.0,
+      "retrieval_residual_source_score_bonus_scale": 0.0,
+      "retrieval_residual_source_advantage_bonus_scale": 0.0,
+      "retrieval_residual_composite_l2_penalty_scale": 0.0,
+      "retrieval_residual_scale": 0.0,
+      "retrieval_residual_scales": [],
+      "retrieval_residual_anchor": "none",
+      "retrieval_residual_direction": "none",
+      "retrieval_residual_reduce": "none",
+      "retrieval_residual_challenger_types": [],
+      "retrieval_residual_challenger_scales": [],
+      "retrieval_residual_challenger_margin": 0.0,
+      "lattice_exclude_types": [],
+      "candidate_type_bonuses": {},
+      "candidate_type_bonus_components": false,
+      "candidate_oracle_rollouts": 0,
+      "candidate_oracle_unique_tolerance": 1e-06,
+      "candidate_oracle_success_rate": null,
+      "candidate_oracle_progress": null,
+      "candidate_oracle_score": null,
+      "candidate_oracle_regret": null,
+      "candidate_oracle_score_gain_over_selected": null,
+      "candidate_oracle_unique_count": null,
+      "candidate_oracle_improvement_rate": null,
+      "candidate_oracle_selected_branch_success_rate": null,
+      "candidate_oracle_selected_branch_progress": null,
+      "candidate_oracle_type_counts": {},
+      "candidate_oracle_best_branch_rank": null,
+      "candidate_oracle_best_branch_rank_counts": {},
+      "candidate_oracle_branch_success_rates": [],
+      "candidate_oracle_branch_progress": [],
+      "candidate_oracle_branch_score_gains_over_selected": [],
+      "selected_residual_scale_counts": {},
+      "policy_rollout_success_rate": 0.38608695652173913,
+      "policy_rollout_progress": 0.5848189972940346,
+      "oracle_success_rate": 0.8765217391304347,
+      "action_mse_to_best": 0.50471622103669,
+      "best_policy_val": {
+        "bc_loss": 0.11538442307048374,
+        "field_effect_loss": 0.00973979427686168,
+        "field_potential_loss": 0.2751605494154824,
+        "field_preference_loss": 0.5092001971271303,
+        "lattice_edges": 3833.3333333333335,
+        "progress_mae": 0.198434103694227,
+        "proposal_loss": 3.109688893548284e-06,
+        "rank_acc": 0.833484646346834,
+        "rank_loss": 0.5091651446289487,
+        "regret_mae": 0.3825343433353636,
+        "success_accuracy": 0.8670964936415354,
+        "total_loss": 1.4539120992024739
+      },
+      "per_task": {
+        "LiftPegUpright-v1": {
+          "action_mse_to_best": 0.35275944230670575,
+          "expert_success_rate": 0.8229166666666666,
+          "num_groups": 96,
+          "oracle_success_rate": 0.9270833333333334,
+          "policy_expert_regret": 0.9295680510501066,
+          "policy_oracle_regret": 1.049254375199477,
+          "policy_rollout_progress": 0.605966795856754,
+          "policy_rollout_success_rate": 0.2916666666666667,
+          "restore_max_error": 1.955777406692505e-07
+        },
+        "PickCube-v1": {
+          "action_mse_to_best": 0.5511499861808821,
+          "expert_success_rate": 0.9444444444444444,
+          "num_groups": 198,
+          "oracle_success_rate": 0.9595959595959596,
+          "policy_expert_regret": 0.918729902691951,
+          "policy_oracle_regret": 0.9187821659560532,
+          "policy_rollout_progress": 0.6430067968404278,
+          "policy_rollout_success_rate": 0.3787878787878788,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PullCube-v1": {
+          "action_mse_to_best": 0.6161547317073099,
+          "expert_success_rate": 0.24444444444444444,
+          "num_groups": 90,
+          "oracle_success_rate": 0.4666666666666667,
+          "policy_expert_regret": 0.2370640807888574,
+          "policy_oracle_regret": 0.5298742365712921,
+          "policy_rollout_progress": 0.33887701496067973,
+          "policy_rollout_success_rate": 0.2222222222222222,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "PushCube-v1": {
+          "action_mse_to_best": 0.37325401663190066,
+          "expert_success_rate": 0.8514851485148515,
+          "num_groups": 101,
+          "oracle_success_rate": 1.0,
+          "policy_expert_regret": 0.30472644248811326,
+          "policy_oracle_regret": 0.323294205240684,
+          "policy_rollout_progress": 0.8450226264424844,
+          "policy_rollout_success_rate": 0.8316831683168316,
+          "restore_max_error": 2.384185791015625e-07
+        },
+        "StackCube-v1": {
+          "action_mse_to_best": 0.6007404648595386,
+          "expert_success_rate": 0.7666666666666667,
+          "num_groups": 90,
+          "oracle_success_rate": 0.9111111111111111,
+          "policy_expert_regret": 1.20392378386524,
+          "policy_oracle_regret": 1.3271643911798796,
+          "policy_rollout_progress": 0.38818387389183046,
+          "policy_rollout_success_rate": 0.16666666666666666,
+          "restore_max_error": 2.2351741790771484e-07
+        }
+      }
+    }
+  ]
+}

results/h16_typedprop_nooponly_sparsehead_prepend_margin0p05_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_typedprop_nooponly_p2`
+Result file: `policy_rollout_proposal_lattice_nooponly_sparsehead_prepend_margin0p05.json`
+Completed seeds: 3
+Baseline h=4 policy success: 29.67%
+Baseline h=16 rank-checkpoint success: 29.74%
+
+Mean success: 34.72% +/- 3.61%
+Gain vs h=16 rank checkpoint: +4.99%
+Mean progress: 56.10%
+Mean action MSE to best: 0.455
+
+| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
+|---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
+| 0 | proposal_lattice | 2 | yes | 0 | none | none | none | none | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 0.00 | none | 0.050 | 0.00 | 0 | 0.00 | 31.48% | 53.58% | 85.74% | n/a | n/a | 0.418 |
+| 1 | proposal_lattice | 2 | yes | 0 | none | none | none | none | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 0.00 | none | 0.050 | 0.00 | 0 | 0.00 | 34.09% | 56.25% | 86.96% | n/a | n/a | 0.441 |
+| 2 | proposal_lattice | 2 | yes | 0 | none | none | none | none | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 0.00 | none | 0.050 | 0.00 | 0 | 0.00 | 38.61% | 58.48% | 87.65% | n/a | n/a | 0.505 |

results/paper_analysis.json

@@ -1,6 +1,6 @@
 {
   "best_clean_key": "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001",
-  "generated_utc": "2026-06-30T14:36:44+00:00",
+  "generated_utc": "2026-06-30T14:52:31+00:00",
   "mechanism_gap": {
     "best_clean_vs_direct_same_ckpt": 0.0776811594202898,
     "best_clean_vs_h16": 0.06318840579710144,
@@ -8269,9 +8269,73 @@
       "std_success": 0.04558379972897795
     },
     "typed_proposal_lattice_nooponly_sparsehead_prepend_margin005": {
+      "ci95_success": 0.08962834306379469,
       "label": "Typed proposal lattice head, no-op-only head retrain, policy-prepended margin 0.05",
-      "missing": true,
-      "source": "results/h16_typedprop_nooponly_sparsehead_prepend_margin0p05_summary.json"
+      "mean_action_mse_to_best": 0.45465255970136975,
+      "mean_progress": 0.5610268774401882,
+      "mean_success": 0.34724637681159415,
+      "num_completed": 3,
+      "per_task_success": {
+        "LiftPegUpright-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.22449163194761224,
+          "std_success": 0.08136397291112392
+        },
+        "PickCube-v1": {
+          "mean_num_groups": 196.66666666666666,
+          "mean_success": 0.3148077767642985,
+          "std_success": 0.05569508697457448
+        },
+        "PullCube-v1": {
+          "mean_num_groups": 81.0,
+          "mean_success": 0.17888154730259995,
+          "std_success": 0.06520207894670627
+        },
+        "PushCube-v1": {
+          "mean_num_groups": 102.0,
+          "mean_success": 0.7596889073611214,
+          "std_success": 0.06578693975916784
+        },
+        "StackCube-v1": {
+          "mean_num_groups": 93.33333333333333,
+          "mean_success": 0.24081474081474077,
+          "std_success": 0.0689604884070811
+        }
+      },
+      "seed_action_mse_to_best": {
+        "0": 0.4179149465454382,
+        "1": 0.441326511521981,
+        "2": 0.50471622103669
+      },
+      "seed_progress": {
+        "0": 0.5357955083963664,
+        "1": 0.5624661266301637,
+        "2": 0.5848189972940346
+      },
+      "seed_success": {
+        "0": 0.31478260869565217,
+        "1": 0.3408695652173913,
+        "2": 0.38608695652173913
+      },
+      "selected_candidate_type_counts": {
+        "policy_continuous": 1456,
+        "proposal_no_op": 269
+      },
+      "selected_residual_scale_counts": {},
+      "selected_type_outcomes": {
+        "policy_continuous": {
+          "count": 1456.0,
+          "mean_progress": 0.5603371559486182,
+          "success_rate": 0.33447802197802196
+        },
+        "proposal_no_op": {
+          "count": 269.0,
+          "mean_progress": 0.5647600911640767,
+          "success_rate": 0.4163568773234201
+        }
+      },
+      "source": "results/h16_typedprop_nooponly_sparsehead_prepend_margin0p05_summary.json",
+      "std_success": 0.036077351611597946
     },
     "typed_proposal_lattice_types2sparse_bestpolicy_prepend_margin005": {
       "ci95_success": 0.09107461025155232,

results/paper_analysis.md

@@ -1,6 +1,6 @@
 # Paper Analysis
 
-Generated: `2026-06-30T14:36:44+00:00`
+Generated: `2026-06-30T14:52:31+00:00`
 
 ## Main Seed Statistics
 
@@ -67,7 +67,7 @@ Generated: `2026-06-30T14:36:44+00:00`
 | typed_proposal_lattice_nooponly_prepend_margin005 | Typed proposal lattice head, no-op family only, policy-prepended margin 0.05 | 3 | 34.43% +/- 4.56 | +/- 11.32 | 56.23% | 0.450 | +4.70 pp |
 | typed_proposal_lattice_types2sparse_bestpolicy_prepend_margin005 | Typed proposal lattice head, sparse families, best-policy checkpoint, policy-prepended margin 0.05 | 3 | 32.29% +/- 3.67 | +/- 9.11 | 55.71% | 0.451 | +2.55 pp |
 | typed_proposal_lattice_nooponly_bestpolicy_prepend_margin005 | Typed proposal lattice head, no-op family only, best-policy checkpoint, policy-prepended margin 0.05 | 3 | 33.16% +/- 3.72 | +/- 9.24 | 56.54% | 0.432 | +3.42 pp |
-| typed_proposal_lattice_nooponly_sparsehead_prepend_margin005 | Typed proposal lattice head, no-op-only head retrain, policy-prepended margin 0.05 | 0 | missing | missing | missing | missing | missing |
+| typed_proposal_lattice_nooponly_sparsehead_prepend_margin005 | Typed proposal lattice head, no-op-only head retrain, policy-prepended margin 0.05 | 3 | 34.72% +/- 3.61 | +/- 8.96 | 56.10% | 0.455 | +4.99 pp |
 | residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001 | K4 composed compatible tangents, no-op bonus 0.03, singleton near-miss bonus 0.01 | 3 | 35.59% +/- 0.99 | +/- 2.46 | 57.10% | 0.406 | +5.86 pp |
 | residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus002 | K4 composed compatible tangents, no-op bonus 0.03, singleton near-miss bonus 0.02 | 3 | 35.59% +/- 0.99 | +/- 2.46 | 57.10% | 0.406 | +5.86 pp |
 | residual_k4_composemasked_dropnmnoop_grid035040045_srcscorebonus002 | K4 composed type-consensus tangents, masked, drop near-miss+no-op composite, source-score bonus 0.02 | 3 | 35.48% +/- 1.22 | +/- 3.02 | 57.02% | 0.408 | +5.74 pp |

results/paper_core_results.md

@@ -61,7 +61,7 @@ lattice is `+27.25 pp`, and the remaining clean-to-same-state proposal gap is
 | K4 compatible residual retrieval, unique candidate-oracle prefix K=8 | No | No | 43.07% diagnostic | +13.33 pp diagnostic | Diagnostic-only measured oracle over generated clean candidate prefix; unique-action trace shows real selector headroom, not duplicate-action inflation |
 | K4 compatible residual retrieval + near-miss challenger gate 0.01 | No | No | 36.06% | +6.32 pp | Current best clean deployment row; a two-stage selector keeps the robust compatible-chart anchor and only lets singleton near-miss tangents override under a tightly calibrated positive field margin |
 | K4 compatible residual retrieval + near-miss challenger fine calibration | No | No | 36.00-36.00% | +6.26 pp | Margins 0.005/0.015/0.02 form a near-tie plateau around the 0.01 optimum; scale-gating the challenger to 0.35 or 0.35/0.40 lowers MSE slightly but does not recover the lost success; 0.03 and near-miss+wrong-gripper are lower or higher-MSE, so the effect is calibrated singleton near-miss geometry rather than broad residual mixing |
-| Typed proposal lattice head, generated primitive families | No | No | 31.30-34.43% | +1.57-4.70 pp | Direct generated support improves as anti-goal families are masked: six-family is 31.30%/32.17%, safe4 is 33.45%, sparse no-op/wrong-gripper is 34.26%, and no-op-only is 34.43%; best-policy checkpoint ablations drop to 32.29%/33.16%, and a no-op-only sparse-head retrain is pending |
+| Typed proposal lattice head, generated primitive families | No | No | 31.30-34.72% | +1.57-4.99 pp | Direct generated support improves as anti-goal families are masked: six-family is 31.30%/32.17%, safe4 is 33.45%, sparse no-op/wrong-gripper is 34.26%, no-op-only is 34.43%, and a no-op-only sparse-head retrain reaches 34.72%; best-policy checkpoint ablations drop to 32.29%/33.16%, so learned proposal locality remains the bottleneck |
 | K4 compatible residual retrieval, margin 0.10 | No | No | 34.67% | +4.93 pp | Naively lowering abstention selects too many bad nonzero tangents and falls below the 36.06% top row |
 | K4 composed type-consensus residual retrieval, exact compatibility mask + source-score prior | No | No | 35.48% | +5.74 pp | Measured train-source reward confidence does not replace the sparse typed prior on the compatible chart; useful negative calibration |
 | K4 composed type-consensus residual retrieval, exact compatibility mask + no-op/source-score priors | No | No | 35.54% | +5.80 pp | Adding measured train-source reward confidence to the current typed-prior chart ties near-best but does not beat exact compatibility masking alone |

results/paper_story_memo.md

@@ -34,7 +34,7 @@ when queried on proposal geometry that matches those local counterfactuals.
 | Singleton near-miss priors do not open a new clean route | adding a small exact `residual_near_miss` singleton prior 0.01 or 0.02 on the compatible chart ties the previous 35.59% compatibility row and slightly raises progress to 57.10% | Tie diagnostic: revived near-miss tangents are high-precision but too sparse to close the clean-to-same-state proposal gap |
 | Candidate-prefix oracle needs unique-action hygiene | the first K=8 prefix diagnostic was archived as `_nonunique`; the deduplicated unique-action trace reaches 43.07% candidate-oracle success with mean best branch rank 2.85 | Supported diagnostic: proposal headroom is real, but branch success falls with rank, so selector calibration must be conditional |
 | Near-miss challenger calibration improves clean deployment | the trace-motivated two-stage selector keeps the compatible-chart anchor and lets singleton near-miss tangents override only under a tightly calibrated margin; margin 0.01 reaches 36.06%, while 0.005/0.015/0.02 and scale-gated 0.35 or 0.35/0.40 variants tie at 36.00% | Current best clean row; small but story-aligned selector-calibration gain, with scale-gating explaining MSE but not improving success and wrong-gripper challengers rejected by lower success or higher MSE |
-| Typed proposal generation confirms support geometry, not just support count, is the remaining clean bottleneck | six-family generated typed proposal lattices reach 31.30%/32.17%; masking anti-goal families raises safe4 to 33.45%, sparse no-op/wrong-gripper to 34.26%, and no-op-only to 34.43%; best-policy checkpoint ablations drop to 32.29%/33.16%; no-op-only sparse-head retrain is pending | Negative/diagnostic: direct proposal generation helps only when sparse and anti-goal masked, but learned proposals still lack the local same-state counterfactual geometry |
+| Typed proposal generation confirms support geometry, not just support count, is the remaining clean bottleneck | six-family generated typed proposal lattices reach 31.30%/32.17%; masking anti-goal families raises safe4 to 33.45%, sparse no-op/wrong-gripper to 34.26%, and no-op-only to 34.43%; best-policy checkpoint ablations drop to 32.29%/33.16%; a no-op-only sparse-head retrain reaches 34.72% | Negative/diagnostic: direct proposal generation helps only when sparse and anti-goal masked, but learned proposals still lack the local same-state counterfactual geometry |
 | Component-wise composite priors do not add the gain | propagating the no-op bonus into composite types reaches 35.36%, below the exact-prior masked composition row at 35.54% and the exact compatibility row at 35.59% | Negative/near-tie diagnostic: sparse exact priors are better than broadly rewarding every no-op-containing composite |
 | Composite trust-radius penalty explains but does not improve the success top line | composite-only L2 penalty 0.02 ties 35.54% while lowering action MSE from 0.4106 to 0.4079; with the exact compatibility mask it lowers MSE further to 0.4048 but drops back to 35.54% | Tie/negative diagnostic: composed tangents need a local trust radius, but compatibility masking gives the success optimum |
 | Minimum-energy residual regularization does not add the gain | action L2 penalty 0.05 ties the previous 35.42% scale-grid row, while 0.10/0.20 reach 35.36% | Negative/tie diagnostic: the clean bridge is not explained by shortest-action bias |
@@ -85,7 +85,7 @@ clean proposal result, the intended main rows are:
 17. K4 masked composed type-consensus tangent transport: 35.30%; with exact near-miss/no-op compatibility mask only: 35.48%; with typed no-op prior: 35.54%; with exact compatibility mask + typed no-op prior: 35.59%; adding singleton near-miss prior 0.01/0.02 ties 35.59%; with source-score prior on that exact chart: 35.48%; with typed no-op + source-score: 35.54%
 18. K4 compatible tangent unique-action candidate-oracle prefix K=8 diagnostic: 43.07%; branch trace shows mean best branch rank 2.85, unique count 8.00, and real but conditional selector headroom
 19. K4 compatible tangent near-miss challenger gate 0.01: 36.06%; current best clean deployment row from trace-motivated two-stage selector calibration. Fine margins 0.005/0.015/0.02 tie at 36.00%; scale-gating the challenger to 0.35 or 0.35/0.40 also ties 36.00% with slightly lower MSE; 0.03 and near-miss+wrong-gripper variants are lower or higher-MSE.
-20. Typed proposal lattice head: six-family generated proposals reach 31.30%/32.17%; safe4 reaches 33.45%; sparse no-op/wrong-gripper reaches 34.26%; no-op-only reaches 34.43%. Checkpointing ablations with `best_policy.pt` are lower at 32.29%/33.16%. A no-op-only sparse-head retrain is pending as `14963500` -> `14963505` -> `14963507`.
+20. Typed proposal lattice head: six-family generated proposals reach 31.30%/32.17%; safe4 reaches 33.45%; sparse no-op/wrong-gripper reaches 34.26%; no-op-only reaches 34.43%. Checkpointing ablations with `best_policy.pt` are lower at 32.29%/33.16%. A no-op-only sparse-head retrain reaches 34.72%, confirming mild multi-family interference but not closing the clean proposal gap.
 21. K4 compatible tangent abstention margin 0.10 diagnostic: 34.67%; naive lower abstention hurts despite candidate-prefix oracle headroom
 22. K4 masked composed type-consensus tangent transport + component-wise no-op prior: 35.36%
 21. K4 masked composed type-consensus tangent transport + composite L2 trust penalty: 35.54% at 0.02; 35.48% at 0.05; 35.54% with exact compatibility mask + L2 0.02

results/paper_table_status.json

@@ -1686,14 +1686,14 @@
       "story_role": "sparse-head retrain test for whether typed proposal failure comes from multi-family interference",
       "fallback_success": null,
       "pending_job": "14963500/14963505/14963507",
-      "path_exists": false,
-      "status": "pending",
-      "success": null,
-      "std_success": null,
+      "path_exists": true,
+      "status": "complete",
+      "success": 0.3472463768115942,
+      "std_success": 0.03607735161159794,
       "completed_seeds": null,
-      "num_completed": null,
+      "num_completed": 3,
       "best_config": null,
-      "gain_vs_h16_policy": null
+      "gain_vs_h16_policy": 0.049855072463768135
     },
     {
       "key": "retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",

results/paper_table_status.md

@@ -91,7 +91,7 @@ Baseline h=16 policy: 29.74%
 | typed_proposal_lattice_nooponly_prepend_margin005 | Typed proposal lattice head, no-op family only, policy-prepended margin 0.05 | complete | 34.43% | +4.70 pp | yes | no | no | minimal generated no-op proposal-family test after sparse typed lattice improved over the full six-family row |
 | typed_proposal_lattice_types2sparse_bestpolicy_prepend_margin005 | Typed proposal lattice head, sparse no-op/wrong-gripper families, best-policy checkpoint | complete | 32.29% | +2.55 pp | yes | no | no | checkpointing ablation for typed generated sparse proposals: lower BC/proposal loss versus weaker field scorer |
 | typed_proposal_lattice_nooponly_bestpolicy_prepend_margin005 | Typed proposal lattice head, no-op family only, best-policy checkpoint | complete | 33.16% | +3.42 pp | yes | no | no | minimal generated no-op checkpointing ablation for typed proposal support |
-| typed_proposal_lattice_nooponly_sparsehead_prepend_margin005 | Typed proposal lattice head, no-op-only head retrain | pending 14963500/14963505/14963507 | pending | pending | yes | no | no | sparse-head retrain test for whether typed proposal failure comes from multi-family interference |
+| typed_proposal_lattice_nooponly_sparsehead_prepend_margin005 | Typed proposal lattice head, no-op-only head retrain | complete | 34.72% | +4.99 pp | yes | no | no | sparse-head retrain test for whether typed proposal failure comes from multi-family interference |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001 | K4 composed compatible residual retrieval, no-op bonus 0.03, singleton near-miss bonus 0.01 | complete | 35.59% | +5.86 pp | yes | no | no | revive high-precision singleton near-miss tangents without boosting toxic near-miss+no-op composites |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus002 | K4 composed compatible residual retrieval, no-op bonus 0.03, singleton near-miss bonus 0.02 | complete | 35.59% | +5.86 pp | yes | no | no | stronger singleton near-miss revival prior on the compatible local tangent chart |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_srcscorebonus002 | K4 composed type-consensus residual retrieval, masked, drop near-miss+no-op composite, source-score bonus 0.02 | complete | 35.48% | +5.74 pp | yes | no | no | train-measured source-score prior on the compatible local tangent chart |

scripts/build_paper_analysis.py

@@ -512,6 +512,16 @@ METHODS = [
         label="Typed proposal lattice head, no-op-only head retrain, policy-prepended margin 0.05",
         summary_path="h16_typedprop_nooponly_sparsehead_prepend_margin0p05_summary.json",
     ),
+    MethodSpec(
+        key="typed_proposal_lattice_nooponly_sparsehead_prepend_margin000",
+        label="Typed proposal lattice head, no-op-only head retrain, policy-prepended margin 0.00",
+        summary_path="h16_typedprop_nooponly_sparsehead_prepend_margin0p00_summary.json",
+    ),
+    MethodSpec(
+        key="typed_proposal_lattice_nooponly_sparsehead_prepend_margin010",
+        label="Typed proposal lattice head, no-op-only head retrain, policy-prepended margin 0.10",
+        summary_path="h16_typedprop_nooponly_sparsehead_prepend_margin0p10_summary.json",
+    ),
     MethodSpec(
         key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",
         label="K4 composed compatible tangents, no-op bonus 0.03, singleton near-miss bonus 0.01",

scripts/build_paper_table_status.py

@@ -905,6 +905,26 @@ SPECS = [
         story_role="sparse-head retrain test for whether typed proposal failure comes from multi-family interference",
         pending_job="14963500/14963505/14963507",
     ),
+    ResultSpec(
+        key="typed_proposal_lattice_nooponly_sparsehead_prepend_margin000",
+        label="Typed proposal lattice head, no-op-only head retrain, margin 0.00",
+        path="h16_typedprop_nooponly_sparsehead_prepend_margin0p00_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="abstention calibration for the no-op-only typed proposal head: select generated no-op more often",
+        pending_job="14964011/14964019",
+    ),
+    ResultSpec(
+        key="typed_proposal_lattice_nooponly_sparsehead_prepend_margin010",
+        label="Typed proposal lattice head, no-op-only head retrain, margin 0.10",
+        path="h16_typedprop_nooponly_sparsehead_prepend_margin0p10_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="abstention calibration for the no-op-only typed proposal head: require stronger field advantage",
+        pending_job="14964016/14964020",
+    ),
     ResultSpec(
         key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmbonus001",
         label="K4 composed compatible residual retrieval, no-op bonus 0.03, singleton near-miss bonus 0.01",