auto-sync 2026-07-04T06:15:18Z workspace (part 3)

workspace/scripts/build_selector_diagnostic_sweep.py

@@ -20,6 +20,7 @@ DEFAULT_PATTERNS = (
     "runs/ctt_base_context_obs_learned_dominance_chartcompat_obs_utility_task_envclip_k16_train_to_test/metrics.json",
     "runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test/metrics.json",
     "runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test_tau0/metrics.json",
+    "runs/ctt_dominance_utility_energy_val_to_test_seed*/metrics.json",
     "runs/ctt_base_context_obs_learned_dominance_*_tanh_train_to_test/metrics.json",
     "runs/ctt_base_context_obs_dominance_tanh_train_to_test/metrics.json",
     "runs/ctt_base_context_obs_learned_dominance_*_perdim_trainmax_train_to_test/metrics.json",
@@ -79,7 +80,7 @@ def main(argv: list[str] | None = None) -> int:
         json.dumps(_group_rows(rows, "family"), indent=2, sort_keys=True) + "\n"
     )
     (out_dir / "metrics_by_seed.json").write_text(
-        json.dumps(_group_rows(rows, "k"), indent=2, sort_keys=True) + "\n"
+        json.dumps(_group_rows(rows, "seed"), indent=2, sort_keys=True) + "\n"
     )
     (out_dir / "table.tex").write_text(_table(best_rows) + "\n")
     (out_dir / "config.yaml").write_text(_config(args, metric_paths) + "\n")
@@ -137,6 +138,7 @@ def _row(path: Path) -> dict[str, Any]:
         "run_path": str(path.parent),
         "family": family,
         "selector": selector,
+        "seed": _infer_seed(run_name, data),
         "report_type": data.get("report_type", "unknown"),
         "k": int(data.get("k") or _infer_k(run_name)),
         "base_success": _num(summary.get("base_success")),
@@ -165,6 +167,8 @@ def _micro_summary(summary: dict[str, Any]) -> dict[str, Any]:
 
 def _family(run_name: str, data: dict[str, Any]) -> str:
     k = int(data.get("k") or _infer_k(run_name))
+    if "ctt_dominance_utility_energy" in run_name or "utility_energy" in run_name:
+        return f"K{k} env_clip utility-energy"
     if "envclip_k16" in run_name:
         return "K16 env_clip"
     if "envclip" in run_name:
@@ -178,6 +182,9 @@ def _family(run_name: str, data: dict[str, Any]) -> str:
 
 def _selector_name(run_name: str, data: dict[str, Any]) -> str:
     report_type = str(data.get("report_type", ""))
+    if data.get("score_source") == "checkpoint" or "utility_energy" in run_name:
+        tau_mode = str(data.get("tau_mode", "auto"))
+        return f"checkpoint utility energy/LCB {tau_mode}, seed={_infer_seed(run_name, data)}"
     if report_type == "dominance_calibrated_selector_eval":
         tau_mode = str(data.get("tau_mode", "auto"))
         return f"LCB {tau_mode}"
@@ -196,6 +203,24 @@ def _infer_k(run_name: str) -> int:
     return 16 if "k16" in run_name else 8
 
 
+def _infer_seed(run_name: str, data: dict[str, Any]) -> str:
+    seed = data.get("seed")
+    if seed is not None:
+        return str(seed)
+    marker = "seed"
+    if marker in run_name:
+        suffix = run_name.rsplit(marker, 1)[-1]
+        digits = []
+        for char in suffix:
+            if char.isdigit():
+                digits.append(char)
+            else:
+                break
+        if digits:
+            return "".join(digits)
+    return "pooled"
+
+
 def _best_by_family(rows: list[dict[str, Any]]) -> list[dict[str, Any]]:
     grouped: dict[str, list[dict[str, Any]]] = defaultdict(list)
     for row in rows: