Auto-sync: 2026-06-30 04:59:14 (part 2)

results/paper_table_status.json

@@ -1467,6 +1467,44 @@
       "best_config": null,
       "gain_vs_h16_policy": 0.06260869565217392
     },
+    {
+      "key": "retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scale035",
+      "label": "K4 compatible residual retrieval, near-miss challenger gate 0.01, scale-gated 0.35",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_scale035_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "trace-motivated near-miss challenger with tangent-length reliability gating at the shortest scale",
+      "fallback_success": null,
+      "pending_job": "14955663/14955664",
+      "path_exists": false,
+      "status": "pending",
+      "success": null,
+      "std_success": null,
+      "completed_seeds": null,
+      "num_completed": null,
+      "best_config": null,
+      "gain_vs_h16_policy": null
+    },
+    {
+      "key": "retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scales035040",
+      "label": "K4 compatible residual retrieval, near-miss challenger gate 0.01, scale-gated 0.35/0.40",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_scales035040_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "trace-motivated near-miss challenger with tangent-length reliability gating over the two shortest scales",
+      "fallback_success": null,
+      "pending_job": "14955665/14955666",
+      "path_exists": false,
+      "status": "pending",
+      "success": null,
+      "std_success": null,
+      "completed_seeds": null,
+      "num_completed": null,
+      "best_config": null,
+      "gain_vs_h16_policy": null
+    },
     {
       "key": "retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
       "label": "K4 compatible residual retrieval, near-miss challenger gate 0.03",

results/paper_table_status.md

@@ -80,6 +80,8 @@ Baseline h=16 policy: 29.74%
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001 | K4 compatible residual retrieval, near-miss challenger gate 0.01 | complete | 36.06% | +6.32 pp | yes | no | no | lower-margin sensitivity for trace-motivated singleton near-miss challenger calibration |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger0005 | K4 compatible residual retrieval, near-miss challenger gate 0.005 | complete | 36.00% | +6.26 pp | yes | no | no | fine lower-margin calibration for the trace-motivated singleton near-miss challenger gate |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger0015 | K4 compatible residual retrieval, near-miss challenger gate 0.015 | complete | 36.00% | +6.26 pp | yes | no | no | fine upper-margin calibration for the trace-motivated singleton near-miss challenger gate |
+| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scale035 | K4 compatible residual retrieval, near-miss challenger gate 0.01, scale-gated 0.35 | pending 14955663/14955664 | pending | pending | yes | no | no | trace-motivated near-miss challenger with tangent-length reliability gating at the shortest scale |
+| retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scales035040 | K4 compatible residual retrieval, near-miss challenger gate 0.01, scale-gated 0.35/0.40 | pending 14955665/14955666 | pending | pending | yes | no | no | trace-motivated near-miss challenger with tangent-length reliability gating over the two shortest scales |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003 | K4 compatible residual retrieval, near-miss challenger gate 0.03 | complete | 35.94% | +6.20 pp | yes | no | no | upper-margin sensitivity for trace-motivated singleton near-miss challenger calibration |
 | retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgchallenger001 | K4 compatible residual retrieval, near-miss/wrong-gripper challenger gate 0.01 | complete | 35.94% | +6.20 pp | yes | no | no | trace-motivated challenger calibration test for whether wrong-gripper residuals carry conditional selector headroom |
 | 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 |

scripts/build_paper_analysis.py

@@ -425,6 +425,22 @@ METHODS = [
             "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p015_summary.json"
         ),
     ),
+    MethodSpec(
+        key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scale035",
+        label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.01, scale-gated 0.35",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_scale035_summary.json"
+        ),
+    ),
+    MethodSpec(
+        key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scales035040",
+        label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.01, scale-gated 0.35/0.40",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_scales035040_summary.json"
+        ),
+    ),
     MethodSpec(
         key="residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
         label="K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.03",

scripts/build_paper_table_status.py

@@ -785,6 +785,26 @@ SPECS = [
         story_role="fine upper-margin calibration for the trace-motivated singleton near-miss challenger gate",
         pending_job="14955093/14955094",
     ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scale035",
+        label="K4 compatible residual retrieval, near-miss challenger gate 0.01, scale-gated 0.35",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_scale035_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="trace-motivated near-miss challenger with tangent-length reliability gating at the shortest scale",
+        pending_job="14955663/14955664",
+    ),
+    ResultSpec(
+        key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scales035040",
+        label="K4 compatible residual retrieval, near-miss challenger gate 0.01, scale-gated 0.35/0.40",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_scales035040_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="trace-motivated near-miss challenger with tangent-length reliability gating over the two shortest scales",
+        pending_job="14955665/14955666",
+    ),
     ResultSpec(
         key="retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003",
         label="K4 compatible residual retrieval, near-miss challenger gate 0.03",

scripts/eval_maniskill_policy_rollout.py

@@ -240,6 +240,12 @@ def main(argv: list[str] | None = None) -> int:
         help="Optional comma-separated residual candidate types that may override the "
         "primary retrieval-residual selection in a second-stage challenger gate.",
     )
+    parser.add_argument(
+        "--retrieval-residual-challenger-scales",
+        default="",
+        help="Optional comma-separated residual scales where challenger candidates may "
+        "override the primary retrieval-residual selection. Empty allows all scales.",
+    )
     parser.add_argument(
         "--retrieval-residual-challenger-margin",
         type=float,
@@ -290,6 +296,16 @@ def main(argv: list[str] | None = None) -> int:
         for item in args.retrieval_residual_challenger_types.split(",")
         if item.strip()
     )
+    try:
+        retrieval_residual_challenger_scales = tuple(
+            float(item.strip())
+            for item in args.retrieval_residual_challenger_scales.split(",")
+            if item.strip()
+        )
+    except ValueError as exc:
+        raise SystemExit(
+            "--retrieval-residual-challenger-scales must contain comma-separated floats"
+        ) from exc
     candidate_type_bonuses: dict[str, float] = {}
     try:
         for item in args.candidate_type_bonuses.split(","):
@@ -358,6 +374,7 @@ def main(argv: list[str] | None = None) -> int:
         retrieval_residual_direction=args.retrieval_residual_direction,
         retrieval_residual_reduce=args.retrieval_residual_reduce,
         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,
         lattice_exclude_types=lattice_exclude_types,
         candidate_type_bonuses=candidate_type_bonuses,

scripts/slurm/eval_maniskill_policy_rollout.sbatch

@@ -74,6 +74,10 @@ RETRIEVAL_RESIDUAL_CHALLENGER_TYPES="${RETRIEVAL_RESIDUAL_CHALLENGER_TYPES:-}"
 if [[ -n "${RETRIEVAL_RESIDUAL_CHALLENGER_TYPES_COLON:-}" ]]; then
   RETRIEVAL_RESIDUAL_CHALLENGER_TYPES="${RETRIEVAL_RESIDUAL_CHALLENGER_TYPES_COLON//:/,}"
 fi
+RETRIEVAL_RESIDUAL_CHALLENGER_SCALES="${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES:-}"
+if [[ -n "${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES_COLON:-}" ]]; then
+  RETRIEVAL_RESIDUAL_CHALLENGER_SCALES="${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES_COLON//:/,}"
+fi
 RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN="${RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN:-0.0}"
 LATTICE_EXCLUDE_TYPES="${LATTICE_EXCLUDE_TYPES:-}"
 if [[ -n "${LATTICE_EXCLUDE_TYPES_COLON:-}" ]]; then
@@ -154,6 +158,7 @@ apptainer exec --nv \
     --retrieval-residual-direction "$RETRIEVAL_RESIDUAL_DIRECTION" \
     --retrieval-residual-reduce "$RETRIEVAL_RESIDUAL_REDUCE" \
     --retrieval-residual-challenger-types "$RETRIEVAL_RESIDUAL_CHALLENGER_TYPES" \
+    --retrieval-residual-challenger-scales "$RETRIEVAL_RESIDUAL_CHALLENGER_SCALES" \
     --retrieval-residual-challenger-margin "$RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN" \
     --lattice-exclude-types "$LATTICE_EXCLUDE_TYPES" \
     --candidate-type-bonuses "$CANDIDATE_TYPE_BONUSES" \

scripts/slurm/summarize_h16_policy_ckpt.sbatch

@@ -104,6 +104,9 @@ for result_path in sorted(base_dir.glob(f"seed_*/{out_name}")):
             "retrieval_residual_challenger_types": data.get(
                 "retrieval_residual_challenger_types", []
             ),
+            "retrieval_residual_challenger_scales": data.get(
+                "retrieval_residual_challenger_scales", []
+            ),
             "retrieval_residual_challenger_margin": data.get(
                 "retrieval_residual_challenger_margin", 0.0
             ),
@@ -403,6 +406,16 @@ for row in rows:
             challenger=(
                 " + challenger "
                 + ",".join(row.get("retrieval_residual_challenger_types") or [])
+                + (
+                    "["
+                    + ",".join(
+                        str(scale)
+                        for scale in (row.get("retrieval_residual_challenger_scales") or [])
+                    )
+                    + "]"
+                    if row.get("retrieval_residual_challenger_scales")
+                    else ""
+                )
                 + f"@{row.get('retrieval_residual_challenger_margin') or 0.0:.2f}"
                 if row.get("retrieval_residual_challenger_types")
                 else ""