binary check: keep float dtype during validation, reject non-{0,1}

Old check cast to int first, which silently truncated probabilities
(0.0007 → 0) so prob CSVs slipped past the binary gate. Stay in float
land and require exact equality with 0.0 or 1.0.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

graphtestbed/submit.py

@@ -61,15 +61,17 @@ def validate_submission(task: str, csv_path: Path) -> dict:
         if (preds < 0).any() or (preds > 1).any():
             raise SystemExit("predictions must lie in [0, 1]")
     elif dtype == "binary":
+        raw = df[s["pred_col"]]
         try:
-            preds = df[s["pred_col"]].astype(int)
+            preds = raw.astype(float)
         except (TypeError, ValueError) as e:
-            raise SystemExit(f"pred_col not int-castable: {e}")
-        bad = ~preds.isin([0, 1])
+            raise SystemExit(f"pred_col not numeric: {e}")
+        bad = ~preds.isin([0.0, 1.0])
         if bad.any():
             raise SystemExit(
-                f"binary submission must contain only 0 or 1; "
-                f"got {int(bad.sum())} other values"
+                f"binary submission must contain only 0 or 1 "
+                f"(no probabilities); got {int(bad.sum())} other values "
+                f"(e.g. {preds[bad].iloc[0]!r})"
             )
 
     return {

server/api.py

@@ -221,14 +221,14 @@ def _validate_schema(sub_df: pd.DataFrame, cfg: dict) -> None:
             raise ValueError("predictions must lie in [0, 1]")
     elif dtype == "binary":
         try:
-            preds = sub_df[s["pred_col"]].astype(int)
+            preds = sub_df[s["pred_col"]].astype(float)
         except (TypeError, ValueError) as e:
-            raise ValueError(f"pred_col not int-castable: {e}")
-        bad = ~preds.isin([0, 1])
+            raise ValueError(f"pred_col not numeric: {e}")
+        bad = ~preds.isin([0.0, 1.0])
         if bad.any():
             raise ValueError(
-                f"binary submission must contain only 0 or 1; "
-                f"got {int(bad.sum())} other values"
+                f"binary submission must contain only 0 or 1 "
+                f"(no probabilities); got {int(bad.sum())} other values"
             )