step 5: leaderboard reads from submissions dataset on hub

Replace the local-only JSONL read with hf_hub_download(force_download=True)
against michaelr27/cadgenbench-submissions. force_download avoids the
stale-snapshot trap that datasets.load_dataset would create per process.

Falls back to AI4Engineering/results.jsonl on any Hub error (broad except)
so the table stays up during a transient outage or a mid-migration rename.

HF_TOKEN secret already set on the Space via HfApi.add_space_secret to
read the private dataset. Secret becomes write-only at launch when the
dataset goes public; rotation TODO documented in space-setup/migration.md.

Co-authored-by: Cursor <cursoragent@cursor.com>

app.py

@@ -15,6 +15,7 @@ from pathlib import Path
 
 import gradio as gr
 import pandas as pd
+from huggingface_hub import hf_hub_download
 
 HF_ORG = os.getenv("HF_ORG", "michaelr27")
 HF_SUBMISSIONS_REPO = os.getenv(
@@ -34,14 +35,43 @@ LEADERBOARD_COLS = [
 ]
 
 
-def load_leaderboard() -> pd.DataFrame:
+def _load_rows_from_hub() -> list[dict] | None:
+    """Pull results.jsonl from the submissions dataset.
+
+    Returns None on any failure so callers can fall back to the local file.
+    """
+    try:
+        path = hf_hub_download(
+            repo_id=HF_SUBMISSIONS_REPO,
+            filename="results.jsonl",
+            repo_type="dataset",
+            force_download=True,
+        )
+        return [
+            json.loads(line)
+            for line in Path(path).read_text().splitlines()
+            if line.strip()
+        ]
+    except Exception as e:  # noqa: BLE001 — any failure should fall back
+        print(f"[load_leaderboard] Hub fetch failed ({type(e).__name__}: {e})")
+        return None
+
+
+def _load_rows_from_local() -> list[dict]:
     if not LOCAL_RESULTS_PATH.exists():
-        return pd.DataFrame(columns=LEADERBOARD_COLS)
-    rows = [
+        return []
+    return [
         json.loads(line)
         for line in LOCAL_RESULTS_PATH.read_text().splitlines()
         if line.strip()
     ]
+
+
+def load_leaderboard() -> pd.DataFrame:
+    rows = _load_rows_from_hub()
+    if rows is None:
+        print("[load_leaderboard] falling back to local results.jsonl")
+        rows = _load_rows_from_local()
     if not rows:
         return pd.DataFrame(columns=LEADERBOARD_COLS)
     df = pd.DataFrame(rows)