Auto-sync: 2026-06-29 07:04:48 (part 2)

scripts/build_paper_analysis.py

@@ -251,6 +251,38 @@ METHODS = [
             "k4_grid035040045_safe_margin0p20_mean_by_type_noopbonus0p03_srcadvgate0p0_summary.json"
         ),
     ),
+    MethodSpec(
+        key="residual_k4_consensus_grid035040045_typesuccessbonus002",
+        label="K4 mean-by-type tangent consensus, scales 0.35/0.40/0.45, train family-success bonus 0.02",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_grid035040045_safe_margin0p20_mean_by_type_typesuccessbonus0p02_summary.json"
+        ),
+    ),
+    MethodSpec(
+        key="residual_k4_consensus_grid035040045_typesuccessbonus003",
+        label="K4 mean-by-type tangent consensus, scales 0.35/0.40/0.45, train family-success bonus 0.03",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_grid035040045_safe_margin0p20_mean_by_type_typesuccessbonus0p03_summary.json"
+        ),
+    ),
+    MethodSpec(
+        key="residual_k4_consensus_grid035040045_typesuccessbonus005",
+        label="K4 mean-by-type tangent consensus, scales 0.35/0.40/0.45, train family-success bonus 0.05",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_grid035040045_safe_margin0p20_mean_by_type_typesuccessbonus0p05_summary.json"
+        ),
+    ),
+    MethodSpec(
+        key="residual_k4_consensus_grid035040045_noopbonus003_typesuccessbonus002",
+        label="K4 mean-by-type tangent consensus, scales 0.35/0.40/0.45, no-op bonus 0.03, train family-success bonus 0.02",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_grid035040045_safe_margin0p20_mean_by_type_noopbonus0p03_typesuccessbonus0p02_summary.json"
+        ),
+    ),
     MethodSpec(
         key="residual_k4_consensus_grid035040045_noopbonus003_l2penalty005",
         label="K4 mean-by-type tangent consensus, scales 0.35/0.40/0.45, no-op bonus 0.03, action L2 penalty 0.05",

scripts/build_paper_table_status.py

@@ -575,6 +575,46 @@ SPECS = [
         story_role="source-local utility-lift gate on the current best typed prior",
         pending_job="14902721/14902723",
     ),
+    ResultSpec(
+        key="retrieval_residual_k4_mean_grid035040045_typesuccessbonus002",
+        label="K4 mean-by-type residual retrieval, scales 0.35/0.40/0.45, margin 0.20, train family-success bonus 0.02",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_grid035040045_safe_margin0p20_mean_by_type_typesuccessbonus0p02_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="continuous train-family reliability prior for mean-consensus residual transport",
+        pending_job="14903128/14903129",
+    ),
+    ResultSpec(
+        key="retrieval_residual_k4_mean_grid035040045_typesuccessbonus003",
+        label="K4 mean-by-type residual retrieval, scales 0.35/0.40/0.45, margin 0.20, train family-success bonus 0.03",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_grid035040045_safe_margin0p20_mean_by_type_typesuccessbonus0p03_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="continuous train-family reliability prior for mean-consensus residual transport",
+        pending_job="14903130/14903131",
+    ),
+    ResultSpec(
+        key="retrieval_residual_k4_mean_grid035040045_typesuccessbonus005",
+        label="K4 mean-by-type residual retrieval, scales 0.35/0.40/0.45, margin 0.20, train family-success bonus 0.05",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_grid035040045_safe_margin0p20_mean_by_type_typesuccessbonus0p05_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="continuous train-family reliability prior for mean-consensus residual transport",
+        pending_job="14903132/14903133",
+    ),
+    ResultSpec(
+        key="retrieval_residual_k4_mean_grid035040045_noopbonus003_typesuccessbonus002",
+        label="K4 mean-by-type residual retrieval, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03, train family-success bonus 0.02",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_grid035040045_safe_margin0p20_mean_by_type_noopbonus0p03_typesuccessbonus0p02_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="continuous train-family reliability calibration on the current best typed prior",
+        pending_job="14903134/14903135",
+    ),
     ResultSpec(
         key="retrieval_residual_k4_mean_grid035040045_noopbonus003_l2penalty005",
         label="K4 mean-by-type residual retrieval, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03, action L2 penalty 0.05",

scripts/eval_maniskill_policy_rollout.py

@@ -129,6 +129,13 @@ def main(argv: list[str] | None = None) -> int:
         help="Minimum train-split terminal success rate for a residual candidate family to be "
         "eligible in retrieval_residual mode. The policy_residual fallback is always kept.",
     )
+    parser.add_argument(
+        "--retrieval-type-success-bonus-scale",
+        type=float,
+        default=0.0,
+        help="Scale for adding a train-split task/family terminal-success prior to each "
+        "retrieved residual candidate before field selection.",
+    )
     parser.add_argument(
         "--retrieval-residual-min-source-progress",
         type=float,
@@ -262,6 +269,7 @@ def main(argv: list[str] | None = None) -> int:
         retrieval_neighbors=args.retrieval_neighbors,
         retrieval_metric=args.retrieval_metric,
         retrieval_type_min_success=args.retrieval_type_min_success,
+        retrieval_type_success_bonus_scale=args.retrieval_type_success_bonus_scale,
         retrieval_residual_min_source_progress=args.retrieval_residual_min_source_progress,
         retrieval_residual_min_source_advantage=args.retrieval_residual_min_source_advantage,
         retrieval_residual_source_progress_bonus_scale=(

scripts/slurm/eval_maniskill_policy_rollout.sbatch

@@ -53,6 +53,7 @@ FIELD_OPTIM_L2_PENALTY="${FIELD_OPTIM_L2_PENALTY:-0.0}"
 RETRIEVAL_NEIGHBORS="${RETRIEVAL_NEIGHBORS:-1}"
 RETRIEVAL_METRIC="${RETRIEVAL_METRIC:-raw}"
 RETRIEVAL_TYPE_MIN_SUCCESS="${RETRIEVAL_TYPE_MIN_SUCCESS:-0.0}"
+RETRIEVAL_TYPE_SUCCESS_BONUS_SCALE="${RETRIEVAL_TYPE_SUCCESS_BONUS_SCALE:-0.0}"
 RETRIEVAL_RESIDUAL_MIN_SOURCE_PROGRESS="${RETRIEVAL_RESIDUAL_MIN_SOURCE_PROGRESS:-0.0}"
 RETRIEVAL_RESIDUAL_MIN_SOURCE_ADVANTAGE="${RETRIEVAL_RESIDUAL_MIN_SOURCE_ADVANTAGE:--1000000000.0}"
 RETRIEVAL_RESIDUAL_SOURCE_PROGRESS_BONUS_SCALE="${RETRIEVAL_RESIDUAL_SOURCE_PROGRESS_BONUS_SCALE:-0.0}"
@@ -124,6 +125,7 @@ apptainer exec --nv \
     --retrieval-neighbors "$RETRIEVAL_NEIGHBORS" \
     --retrieval-metric "$RETRIEVAL_METRIC" \
     --retrieval-type-min-success "$RETRIEVAL_TYPE_MIN_SUCCESS" \
+    --retrieval-type-success-bonus-scale "$RETRIEVAL_TYPE_SUCCESS_BONUS_SCALE" \
     --retrieval-residual-min-source-progress "$RETRIEVAL_RESIDUAL_MIN_SOURCE_PROGRESS" \
     --retrieval-residual-min-source-advantage "$RETRIEVAL_RESIDUAL_MIN_SOURCE_ADVANTAGE" \
     --retrieval-residual-source-progress-bonus-scale "$RETRIEVAL_RESIDUAL_SOURCE_PROGRESS_BONUS_SCALE" \

scripts/slurm/eval_maniskill_policy_rollout_cpu_smoke.sbatch

@@ -52,6 +52,7 @@ FIELD_OPTIM_L2_PENALTY="${FIELD_OPTIM_L2_PENALTY:-0.02}"
 RETRIEVAL_NEIGHBORS="${RETRIEVAL_NEIGHBORS:-1}"
 RETRIEVAL_METRIC="${RETRIEVAL_METRIC:-raw}"
 RETRIEVAL_TYPE_MIN_SUCCESS="${RETRIEVAL_TYPE_MIN_SUCCESS:-0.0}"
+RETRIEVAL_TYPE_SUCCESS_BONUS_SCALE="${RETRIEVAL_TYPE_SUCCESS_BONUS_SCALE:-0.0}"
 RETRIEVAL_RESIDUAL_MIN_SOURCE_PROGRESS="${RETRIEVAL_RESIDUAL_MIN_SOURCE_PROGRESS:-0.0}"
 RETRIEVAL_RESIDUAL_MIN_SOURCE_ADVANTAGE="${RETRIEVAL_RESIDUAL_MIN_SOURCE_ADVANTAGE:--1000000000.0}"
 RETRIEVAL_RESIDUAL_SOURCE_PROGRESS_BONUS_SCALE="${RETRIEVAL_RESIDUAL_SOURCE_PROGRESS_BONUS_SCALE:-0.0}"
@@ -120,6 +121,7 @@ apptainer exec \
     --retrieval-neighbors "$RETRIEVAL_NEIGHBORS" \
     --retrieval-metric "$RETRIEVAL_METRIC" \
     --retrieval-type-min-success "$RETRIEVAL_TYPE_MIN_SUCCESS" \
+    --retrieval-type-success-bonus-scale "$RETRIEVAL_TYPE_SUCCESS_BONUS_SCALE" \
     --retrieval-residual-min-source-progress "$RETRIEVAL_RESIDUAL_MIN_SOURCE_PROGRESS" \
     --retrieval-residual-min-source-advantage "$RETRIEVAL_RESIDUAL_MIN_SOURCE_ADVANTAGE" \
     --retrieval-residual-source-progress-bonus-scale "$RETRIEVAL_RESIDUAL_SOURCE_PROGRESS_BONUS_SCALE" \

tests/test_maniskill_policy_rollout.py

@@ -1104,9 +1104,11 @@ def test_retrieval_residual_type_success_threshold_filters_train_families() -> N
         observation_mode="state",
         retrieval_neighbors=1,
         retrieval_type_min_success=0.5,
+        retrieval_type_success_bonus_scale=0.2,
     )
 
     assert attached.candidate_types == ["policy_residual", "residual_near_miss"]
+    assert np.allclose(attached.candidate_score_bonuses, [0.0, 0.2])
     assert np.allclose(
         np.asarray(attached.candidate_action_values, dtype=np.float32),
         np.asarray([[[0.0, 0.0]], [[0.2, 0.0]]]),