Auto-sync: 2026-06-29 11:36:27 (part 2)

scripts/build_paper_analysis.py

@@ -331,6 +331,22 @@ METHODS = [
             "k4_compose_grid035040045_safe_margin0p20_noopbonus0p03_summary.json"
         ),
     ),
+    MethodSpec(
+        key="residual_k4_composemasked_grid035040045",
+        label="K4 composed type-consensus tangents, masked, scales 0.35/0.40/0.45",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_grid035040045_safe_margin0p20_summary.json"
+        ),
+    ),
+    MethodSpec(
+        key="residual_k4_composemasked_grid035040045_noopbonus003",
+        label="K4 composed type-consensus tangents, masked, scales 0.35/0.40/0.45, no-op bonus 0.03",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_grid035040045_safe_margin0p20_noopbonus0p03_summary.json"
+        ),
+    ),
     MethodSpec(
         key="repair_nearmiss_k4_grid025035050_margin020",
         label="K4 near-miss-to-expert repair tangent, scales 0.25/0.35/0.50, margin 0.20",

scripts/build_paper_table_status.py

@@ -675,6 +675,26 @@ SPECS = [
         story_role="local tangent composition on the current best typed prior",
         pending_job="14905957/14905964",
     ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_grid035040045",
+        label="K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_grid035040045_safe_margin0p20_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="local tangent composition with anti-goal composite masks",
+        pending_job="14911979/14911982",
+    ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_grid035040045_noopbonus003",
+        label="K4 composed type-consensus residual retrieval, masked, scales 0.35/0.40/0.45, margin 0.20, no-op bonus 0.03",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_grid035040045_safe_margin0p20_noopbonus0p03_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="local tangent composition with anti-goal composite masks on the current best typed prior",
+        pending_job="14911980/14911983",
+    ),
     ResultSpec(
         key="retrieval_repair_nearmiss_k4_grid025035050_margin020",
         label="K4 near-miss-to-expert repair tangent, scales 0.25/0.35/0.50, margin 0.20",

tests/test_maniskill_policy_rollout.py

@@ -19,6 +19,7 @@ from dovla_cil.eval.maniskill_policy_rollout import (
     _adapt_action_dim,
     _attach_retrieved_residual_candidates,
     _effective_lattice_candidate_count,
+    _lattice_candidate_mask,
     _load_state_archive,
     _numeric_action_values,
     _reduce_residual_candidates_by_type,
@@ -353,6 +354,39 @@ def test_lattice_mode_respects_candidate_mask() -> None:
     assert index.tolist() == [1]
 
 
+def test_lattice_candidate_mask_excludes_composite_candidate_parts() -> None:
+    import torch
+
+    case = _RolloutCase(
+        group_id="g",
+        task_id="PickCube-v1",
+        source_dataset=Path("."),
+        state={},
+        observation={"features": [0.0]},
+        instruction="pick",
+        oracle_score=1.0,
+        oracle_success=True,
+        expert_score=1.0,
+        expert_success=True,
+        best_action_values=[[0.0]],
+        candidate_action_values=[[[0.0]], [[0.2]], [[0.4]]],
+        candidate_types=[
+            "policy_residual",
+            "residual_no_op+residual_random_negative",
+            "residual_no_op+residual_wrong_gripper",
+        ],
+    )
+
+    mask = _lattice_candidate_mask(
+        [case],
+        torch=torch,
+        device="cpu",
+        exclude_types=("residual_random_negative",),
+    )
+
+    assert mask.tolist() == [[True, False, True]]
+
+
 def test_lattice_mode_can_prepend_policy_baseline_for_margin_abstention() -> None:
     import torch