Auto-sync: 2026-06-30 20:06:30 (part 2)

scripts/build_paper_analysis.py

@@ -465,6 +465,22 @@ METHODS = [
             "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_summary.json"
         ),
     ),
+    MethodSpec(
+        key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgmargin003_challenger001",
+        label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger 0.01, wrong-gripper margin 0.03",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgmargin0p03_challenger0p01_summary.json"
+        ),
+    ),
+    MethodSpec(
+        key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgmargin005_challenger001",
+        label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger 0.01, wrong-gripper margin 0.05",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgmargin0p05_challenger0p01_summary.json"
+        ),
+    ),
     MethodSpec(
         key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgchallenger001_pickpull",
         label="K4 compatible tangents, no-op bonus 0.03, near-miss/wrong-gripper challenger gate 0.01 on Pick/Pull only",

scripts/build_paper_table_status.py

@@ -835,6 +835,26 @@ SPECS = [
         story_role="trace-motivated challenger calibration test for whether wrong-gripper residuals carry conditional selector headroom",
         pending_job="14954530/14954531",
     ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgmargin003_challenger001",
+        label="K4 compatible residual retrieval, near-miss challenger 0.01, wrong-gripper margin 0.03",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgmargin0p03_challenger0p01_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="family-specific reliability margin: keep near-miss sensitive while requiring stronger wrong-gripper evidence",
+        pending_job="pending",
+    ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgmargin005_challenger001",
+        label="K4 compatible residual retrieval, near-miss challenger 0.01, wrong-gripper margin 0.05",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgmargin0p05_challenger0p01_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="family-specific reliability margin upper setting for wrong-gripper challenger evidence",
+        pending_job="pending",
+    ),
     ResultSpec(
         key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgchallenger001_pickpull",
         label="K4 compatible residual retrieval, near-miss/wrong-gripper challenger gate 0.01 on Pick/Pull only",

scripts/eval_maniskill_policy_rollout.py

@@ -261,6 +261,12 @@ def main(argv: list[str] | None = None) -> int:
         help="Required field-potential margin for a challenger candidate to override the "
         "primary retrieval-residual selection.",
     )
+    parser.add_argument(
+        "--retrieval-residual-challenger-type-margins",
+        default="",
+        help="Optional comma-separated candidate_type=margin overrides for challenger "
+        "families. Types without an entry use --retrieval-residual-challenger-margin.",
+    )
     parser.add_argument(
         "--retrieval-residual-challenger-tasks",
         default="",
@@ -342,6 +348,21 @@ def main(argv: list[str] | None = None) -> int:
         for item in args.retrieval_residual_challenger_tasks.split(",")
         if item.strip()
     )
+    retrieval_residual_challenger_type_margins: dict[str, float] = {}
+    try:
+        for item in args.retrieval_residual_challenger_type_margins.split(","):
+            item = item.strip()
+            if not item:
+                continue
+            key, value = item.split("=", 1)
+            retrieval_residual_challenger_type_margins[key.strip()] = float(
+                value.strip()
+            )
+    except ValueError as exc:
+        raise SystemExit(
+            "--retrieval-residual-challenger-type-margins must contain "
+            "comma-separated candidate_type=margin entries"
+        ) from exc
     retrieval_residual_challenger_type_tasks: dict[str, tuple[str, ...]] = {}
     try:
         for item in args.retrieval_residual_challenger_type_tasks.split(";"):
@@ -443,6 +464,9 @@ def main(argv: list[str] | None = None) -> int:
         retrieval_residual_challenger_types=retrieval_residual_challenger_types,
         retrieval_residual_challenger_scales=retrieval_residual_challenger_scales,
         retrieval_residual_challenger_margin=args.retrieval_residual_challenger_margin,
+        retrieval_residual_challenger_type_margins=(
+            retrieval_residual_challenger_type_margins
+        ),
         retrieval_residual_challenger_tasks=retrieval_residual_challenger_tasks,
         retrieval_residual_challenger_type_tasks=(
             retrieval_residual_challenger_type_tasks

scripts/slurm/eval_maniskill_policy_rollout.sbatch

@@ -83,6 +83,7 @@ if [[ -n "${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES_COLON:-}" ]]; then
   RETRIEVAL_RESIDUAL_CHALLENGER_SCALES="${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES_COLON//:/,}"
 fi
 RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN="${RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN:-0.0}"
+RETRIEVAL_RESIDUAL_CHALLENGER_TYPE_MARGINS="${RETRIEVAL_RESIDUAL_CHALLENGER_TYPE_MARGINS:-}"
 RETRIEVAL_RESIDUAL_CHALLENGER_TASKS="${RETRIEVAL_RESIDUAL_CHALLENGER_TASKS:-}"
 if [[ -n "${RETRIEVAL_RESIDUAL_CHALLENGER_TASKS_COLON:-}" ]]; then
   RETRIEVAL_RESIDUAL_CHALLENGER_TASKS="${RETRIEVAL_RESIDUAL_CHALLENGER_TASKS_COLON//:/,}"
@@ -171,6 +172,7 @@ apptainer exec --nv \
     --retrieval-residual-challenger-types "$RETRIEVAL_RESIDUAL_CHALLENGER_TYPES" \
     --retrieval-residual-challenger-scales "$RETRIEVAL_RESIDUAL_CHALLENGER_SCALES" \
     --retrieval-residual-challenger-margin "$RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN" \
+    --retrieval-residual-challenger-type-margins "$RETRIEVAL_RESIDUAL_CHALLENGER_TYPE_MARGINS" \
     --retrieval-residual-challenger-tasks "$RETRIEVAL_RESIDUAL_CHALLENGER_TASKS" \
     --retrieval-residual-challenger-type-tasks "$RETRIEVAL_RESIDUAL_CHALLENGER_TYPE_TASKS" \
     --lattice-exclude-types "$LATTICE_EXCLUDE_TYPES" \