Upload folder using huggingface_hub

README.md

@@ -129,9 +129,11 @@ equilibrium-regularized model outputs three stress-tensor channels and derives v
 ### Results
 
 Geo-FNO Elasticity test relative L2. We report **median** (robust to the high run-to-run variance of
-this 1000-sample benchmark) alongside mean ± std and the seed count; full per-seed distributions are in
-the repository (`results/study_summary.json`). All "this work" rows use identical training under one
-method (`torch.compile`); the baseline reproduces the published 0.0064 on its good seeds.
+this 1000-sample benchmark) alongside mean ± std and the seed count; full committed per-seed
+distributions are in the repository (`docs/RESULTS.md`, `results/study_summary.json`). The
+accuracy-comparison rows (baseline, LinearNO M=64/M=256) use `torch.compile`; the
+equilibrium-regularized row is eager (`compile:false`). The baseline reproduces the published 0.0064
+on its good seeds.
 
 | Model | median | mean ± std | n | params |
 |---|---|---|---|---|

model.py

@@ -76,5 +76,7 @@ def predict_stress(model, coords: np.ndarray, norm, info, device: str = "cpu") -
         S = info["scale_S"] or 1.0
         stress = von_mises(out) * S          # 3-channel tensor -> von Mises, undo target scale
     else:
-        stress = out[:, 0] * std + mean      # de-normalize scalar prediction
-    return stress.detach().cpu().numpy()
+        # de-normalize scalar prediction; reshape so the broadcast of a (1,1[,1]) normalizer
+        # against out[:,0] (N,) cannot leak a leading axis (audit bug 1). Always returns (N,).
+        stress = (out[:, 0] * std.reshape(()) + mean.reshape(()))
+    return stress.reshape(-1).detach().cpu().numpy()

stress_operator/data/ood_split.py

@@ -59,6 +59,15 @@ def make_ood_split(
 
     in_dist = np.where(stat <= threshold)[0]
     ood = np.where(stat > threshold)[0]
+    # A strict `> quantile` split yields an empty OOD group if the stat ties at its max
+    # (degenerate distribution). Guard so `.min()/.max()` below cannot crash (audit bug 2);
+    # real continuous `rr` data is non-degenerate (n_ood ~ 400 at train_frac=0.8).
+    if ood.size == 0:
+        raise ValueError(
+            f"OOD split is empty: stat ({stat_kind}) has no values above the "
+            f"{train_frac:.0%} quantile (likely tied at the maximum). Lower train_frac "
+            f"or choose a different stat."
+        )
 
     rng = np.random.default_rng(seed)
     perm = rng.permutation(in_dist)
@@ -73,7 +82,7 @@ def make_ood_split(
         "n_ood_test": int(ood.size),
         "threshold": threshold,
         "stat_kind": stat_kind,
-        "stat_train_max": float(stat[train_idx].max()),
+        "stat_train_max": float(stat[train_idx].max()) if train_idx.size else float("nan"),
         "stat_ood_min": float(stat[ood].min()),
         "stat_ood_max": float(stat[ood].max()),
     }