Auto-sync: 2026-06-30 14:47:01 (part 2)

results/paper_analysis.json

@@ -1,6 +1,6 @@
 {
   "best_clean_key": "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001",
-  "generated_utc": "2026-06-30T15:06:34+00:00",
+  "generated_utc": "2026-06-30T18:49:56+00:00",
   "mechanism_gap": {
     "best_clean_vs_direct_same_ckpt": 0.0776811594202898,
     "best_clean_vs_h16": 0.06318840579710144,
@@ -2208,6 +2208,11 @@
       "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_summary.json",
       "std_success": 0.011315490606343858
     },
+    "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",
+      "missing": true,
+      "source": "results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_summary.json"
+    },
     "residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_oraclek8": {
       "candidate_oracle_rollouts": 8,
       "candidate_oracle_type_counts": {

results/paper_analysis.md

@@ -1,6 +1,6 @@
 # Paper Analysis
 
-Generated: `2026-06-30T15:06:34+00:00`
+Generated: `2026-06-30T18:49:56+00:00`
 
 ## Main Seed Statistics
 
@@ -60,6 +60,7 @@ Generated: `2026-06-30T15:06:34+00:00`
 | residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger001_scales035040 | K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.01, scale-gated 0.35/0.40 | 3 | 36.00% +/- 1.31 | +/- 3.26 | 57.37% | 0.407 | +6.26 pp |
 | residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmchallenger003 | K4 compatible tangents, no-op bonus 0.03, near-miss challenger gate 0.03 | 3 | 35.94% +/- 1.28 | +/- 3.18 | 57.36% | 0.407 | +6.20 pp |
 | residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgchallenger001 | K4 compatible tangents, no-op bonus 0.03, near-miss/wrong-gripper challenger gate 0.01 | 3 | 35.94% +/- 1.13 | +/- 2.81 | 57.40% | 0.424 | +6.20 pp |
+| residual_k4_composemasked_dropnmnoop_grid035040045_noopbonus003_nmwgchallenger001_pickpull | K4 compatible tangents, no-op bonus 0.03, near-miss/wrong-gripper challenger gate 0.01 on Pick/Pull only | 0 | missing | missing | missing | missing | missing |
 | typed_proposal_lattice_types6_prepend_margin000 | Typed proposal lattice head, six families, policy-prepended margin 0.00 | 3 | 31.30% +/- 3.81 | +/- 9.48 | 54.03% | 0.480 | +1.57 pp |
 | typed_proposal_lattice_types6_prepend_margin005 | Typed proposal lattice head, six families, policy-prepended margin 0.05 | 3 | 32.17% +/- 3.05 | +/- 7.57 | 54.74% | 0.467 | +2.43 pp |
 | typed_proposal_lattice_types4safe_prepend_margin005 | Typed proposal lattice head, safe four families, policy-prepended margin 0.05 | 3 | 33.45% +/- 3.39 | +/- 8.43 | 55.47% | 0.456 | +3.71 pp |

results/paper_table_status.json

@@ -1543,6 +1543,25 @@
       "best_config": null,
       "gain_vs_h16_policy": 0.06202898550724639
     },
+    {
+      "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",
+      "path": "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_summary.json",
+      "clean_deployment": "yes",
+      "same_state_proposals": "no",
+      "expert_proposal": "no",
+      "story_role": "task-conditional challenger gate using branch trace evidence that wrong-gripper tangents help Pick/Pull but harm other tasks",
+      "fallback_success": null,
+      "pending_job": "14974294/14974296",
+      "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": "typed_proposal_lattice_types6_prepend_margin000",
       "label": "Typed proposal lattice head, six families, policy-prepended margin 0.00",

results/paper_table_status.md

@@ -84,6 +84,7 @@ Baseline h=16 policy: 29.74%
 | 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 | complete | 36.00% | +6.26 pp | 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_nmwgchallenger001_pickpull | K4 compatible residual retrieval, near-miss/wrong-gripper challenger gate 0.01 on Pick/Pull only | pending 14974294/14974296 | pending | pending | yes | no | no | task-conditional challenger gate using branch trace evidence that wrong-gripper tangents help Pick/Pull but harm other tasks |
 | typed_proposal_lattice_types6_prepend_margin000 | Typed proposal lattice head, six families, policy-prepended margin 0.00 | complete | 31.30% | +1.57 pp | yes | no | no | model-generated counterfactual proposal support test; field scores typed proposals rather than retrieved train-state residuals |
 | typed_proposal_lattice_types6_prepend_margin005 | Typed proposal lattice head, six families, policy-prepended margin 0.05 | complete | 32.17% | +2.43 pp | yes | no | no | model-generated counterfactual proposal support test with a conservative policy-abstention margin |
 | typed_proposal_lattice_types4safe_prepend_margin005 | Typed proposal lattice head, safe four families, policy-prepended margin 0.05 | complete | 33.45% | +3.71 pp | yes | no | no | anti-goal proposal-family mask test for the model-generated typed lattice |

scripts/build_paper_analysis.py

@@ -457,6 +457,14 @@ METHODS = [
             "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_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",
+        summary_path=(
+            "h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_"
+            "k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_summary.json"
+        ),
+    ),
     MethodSpec(
         key="typed_proposal_lattice_types6_prepend_margin000",
         label="Typed proposal lattice head, six families, policy-prepended margin 0.00",

scripts/build_paper_table_status.py

@@ -825,6 +825,16 @@ 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_nmwgchallenger001_pickpull",
+        label="K4 compatible residual retrieval, near-miss/wrong-gripper challenger gate 0.01 on Pick/Pull only",
+        path="h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_summary.json",
+        clean_deployment="yes",
+        same_state_proposals="no",
+        expert_proposal="no",
+        story_role="task-conditional challenger gate using branch trace evidence that wrong-gripper tangents help Pick/Pull but harm other tasks",
+        pending_job="14974294/14974296",
+    ),
     ResultSpec(
         key="typed_proposal_lattice_types6_prepend_margin000",
         label="Typed proposal lattice head, six families, policy-prepended margin 0.00",

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-tasks",
+        default="",
+        help="Optional comma-separated task IDs where challenger candidates may override "
+        "the primary retrieval-residual selection. Empty enables challengers for all tasks.",
+    )
     parser.add_argument(
         "--lattice-exclude-types",
         default="",
@@ -304,6 +310,11 @@ def main(argv: list[str] | None = None) -> int:
         for item in args.retrieval_residual_challenger_types.split(",")
         if item.strip()
     )
+    retrieval_residual_challenger_tasks = tuple(
+        item.strip()
+        for item in args.retrieval_residual_challenger_tasks.split(",")
+        if item.strip()
+    )
     proposal_lattice_types = tuple(
         item.strip()
         for item in args.proposal_lattice_types.split(",")
@@ -390,6 +401,7 @@ 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_tasks=retrieval_residual_challenger_tasks,
         lattice_exclude_types=lattice_exclude_types,
         candidate_type_bonuses=candidate_type_bonuses,
         candidate_type_bonus_components=args.candidate_type_bonus_components,

scripts/slurm/eval_maniskill_policy_rollout.sbatch

@@ -83,6 +83,10 @@ if [[ -n "${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES_COLON:-}" ]]; then
   RETRIEVAL_RESIDUAL_CHALLENGER_SCALES="${RETRIEVAL_RESIDUAL_CHALLENGER_SCALES_COLON//:/,}"
 fi
 RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN="${RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN:-0.0}"
+RETRIEVAL_RESIDUAL_CHALLENGER_TASKS="${RETRIEVAL_RESIDUAL_CHALLENGER_TASKS:-}"
+if [[ -n "${RETRIEVAL_RESIDUAL_CHALLENGER_TASKS_COLON:-}" ]]; then
+  RETRIEVAL_RESIDUAL_CHALLENGER_TASKS="${RETRIEVAL_RESIDUAL_CHALLENGER_TASKS_COLON//:/,}"
+fi
 LATTICE_EXCLUDE_TYPES="${LATTICE_EXCLUDE_TYPES:-}"
 if [[ -n "${LATTICE_EXCLUDE_TYPES_COLON:-}" ]]; then
   LATTICE_EXCLUDE_TYPES="${LATTICE_EXCLUDE_TYPES_COLON//:/,}"
@@ -165,6 +169,7 @@ apptainer exec --nv \
     --retrieval-residual-challenger-types "$RETRIEVAL_RESIDUAL_CHALLENGER_TYPES" \
     --retrieval-residual-challenger-scales "$RETRIEVAL_RESIDUAL_CHALLENGER_SCALES" \
     --retrieval-residual-challenger-margin "$RETRIEVAL_RESIDUAL_CHALLENGER_MARGIN" \
+    --retrieval-residual-challenger-tasks "$RETRIEVAL_RESIDUAL_CHALLENGER_TASKS" \
     --lattice-exclude-types "$LATTICE_EXCLUDE_TYPES" \
     --candidate-type-bonuses "$CANDIDATE_TYPE_BONUSES" \
     --candidate-oracle-rollouts "$CANDIDATE_ORACLE_ROLLOUTS" \

scripts/slurm/summarize_h16_policy_ckpt.sbatch

@@ -112,6 +112,9 @@ for result_path in sorted(base_dir.glob(f"seed_*/{out_name}")):
             "retrieval_residual_challenger_margin": data.get(
                 "retrieval_residual_challenger_margin", 0.0
             ),
+            "retrieval_residual_challenger_tasks": data.get(
+                "retrieval_residual_challenger_tasks", []
+            ),
             "lattice_exclude_types": data.get("lattice_exclude_types", []),
             "candidate_type_bonuses": data.get("candidate_type_bonuses", {}),
             "candidate_type_bonus_components": data.get(
@@ -392,6 +395,12 @@ lines.extend(
 for row in rows:
     scales = row.get("retrieval_residual_scales") or []
     scale_grid = ",".join(f"{float(scale):.2f}" for scale in scales) if scales else "none"
+    challenger_tasks = row.get("retrieval_residual_challenger_tasks") or []
+    challenger_task_suffix = (
+        " on " + ",".join(challenger_tasks)
+        if challenger_tasks
+        else ""
+    )
     lines.append(
         "| {seed} | {mode} | {k} | {policy_cand} | {retrieval} | {metric} | {anchor} | {direction} | {reduce}{challenger} | {min_success:.2f} | {type_success_bonus:.3f} | {consensus_penalty:.3f} | {min_source_progress:.2f} | {source_progress_bonus:.3f} | {source_score_bonus:.3f} | {scale:.2f} | {scale_grid} | {margin:.3f} | {sigma:.2f} | {steps} | {trust:.2f} | "
         "{success:.2%} | {progress:.2%} | {oracle:.2%} | {candidate_oracle} | {oracle_gain} | {mse:.3f} |".format(
@@ -418,6 +427,7 @@ for row in rows:
                     if row.get("retrieval_residual_challenger_scales")
                     else ""
                 )
+                + challenger_task_suffix
                 + f"@{row.get('retrieval_residual_challenger_margin') or 0.0:.2f}"
                 if row.get("retrieval_residual_challenger_types")
                 else ""