Fix numeric runtime sorting in leaderboard tables

Co-authored-by: openhands <openhands@all-hands.dev>

leaderboard_transformer.py

@@ -1472,38 +1472,47 @@ def format_score_column(df: pd.DataFrame, score_col_name: str) -> pd.DataFrame:
     return df.assign(**{score_col_name: df[score_col_name].apply(apply_formatting)})
 
 
+def _hidden_runtime_sort_key(runtime_value: float | int | None, score_value: float | int | None) -> str:
+    """Build a hidden prefix so Gradio's string-based runtime sorting behaves numerically."""
+    if pd.notna(runtime_value) and isinstance(runtime_value, (int, float)):
+        return f"{float(runtime_value):020.6f}"
+    if pd.notna(score_value):
+        return "99999999999999999998"
+    return "99999999999999999999"
+
+
 def format_runtime_column(df: pd.DataFrame, runtime_col_name: str) -> pd.DataFrame:
     """
     Applies custom formatting to a runtime column based on its corresponding score column.
     - If runtime is not null, formats as time with 's' suffix.
     - If runtime is null but score is not, it becomes "Missing".
     - If both runtime and score are null, it becomes "Not Submitted".
+    - Adds a hidden, zero-padded numeric prefix so Gradio sorts the column numerically.
     Args:
         df: The DataFrame to modify.
         runtime_col_name: The name of the runtime column to format (e.g., "Average Runtime").
     Returns:
         The DataFrame with the formatted runtime column.
     """
-    # Find the corresponding score column by replacing "Runtime" with "Score"
     score_col_name = runtime_col_name.replace("Runtime", "Score")
 
-    # Ensure the score column actually exists to avoid errors
     if score_col_name not in df.columns:
-        return df  # Return the DataFrame unmodified if there's no matching score
+        return df
 
     def apply_formatting_logic(row):
         runtime_value = row[runtime_col_name]
         score_value = row[score_col_name]
         status_color = "#ec4899"
+        sort_key = _hidden_runtime_sort_key(runtime_value, score_value)
+        hidden_sort_prefix = f'<span style="display:none">{sort_key}</span>'
 
         if pd.notna(runtime_value) and isinstance(runtime_value, (int, float)):
-            return f"{runtime_value:.0f}s"
+            return f"{hidden_sort_prefix}{runtime_value:.0f}s"
         elif pd.notna(score_value):
-            return f'<span style="color: {status_color};">Missing</span>'  # Score exists, but runtime is missing
+            return f'{hidden_sort_prefix}<span style="color: {status_color};">Missing</span>'
         else:
-            return f'<span style="color: {status_color};">Not Submitted</span>'  # Neither score nor runtime exists
+            return f'{hidden_sort_prefix}<span style="color: {status_color};">Not Submitted</span>'
 
-    # Apply the logic to the specified runtime column and update the DataFrame
     df[runtime_col_name] = df.apply(apply_formatting_logic, axis=1)
 
     return df

tests/test_runtime_sorting.py

@@ -0,0 +1,40 @@
+import pandas as pd
+
+from leaderboard_transformer import format_runtime_column
+
+
+def test_runtime_strings_sort_numerically_in_ascending_order():
+    df = pd.DataFrame(
+        {
+            "Average Score": [0.8, 0.8, 0.8, 0.8, None],
+            "Average Runtime": [1323.0, 372.0, 410.0, None, None],
+        }
+    )
+
+    formatted = format_runtime_column(df.copy(), "Average Runtime")
+    runtimes = formatted["Average Runtime"].tolist()
+
+    assert sorted(runtimes) == [
+        runtimes[1],
+        runtimes[2],
+        runtimes[0],
+        runtimes[3],
+        runtimes[4],
+    ]
+
+
+def test_runtime_formatting_preserves_visible_labels():
+    df = pd.DataFrame(
+        {
+            "Average Score": [0.8, 0.8, None],
+            "Average Runtime": [45.2, None, None],
+        }
+    )
+
+    formatted = format_runtime_column(df.copy(), "Average Runtime")
+    values = formatted["Average Runtime"].tolist()
+
+    assert values[0].endswith("45s")
+    assert values[1].endswith("Missing</span>")
+    assert values[2].endswith("Not Submitted</span>")
+    assert 'display:none' in values[0]