Align leaderboard with SpectrumWorld paper Table 3

app.py

@@ -28,10 +28,10 @@ class SpectralLeaderboard:
         """格式化准确率显示"""
         if accuracy is None:
             return "-"
-        return f"{accuracy:.1f}"
+        return f"{accuracy:.2f}"
 
     def _calculate_average(self, results: Dict) -> Optional[float]:
-        """计算平均准确率，使用overall_accuracy字段"""
+        """Return Avg. Perf from Table 3: arithmetic mean across 14 subtasks."""
         return results.get("overall_accuracy")
 
     def _get_model_type_icon(self, model_type: str) -> str:
@@ -363,7 +363,7 @@ def create_leaderboard():
                 - **Rank**: 🥇 1st place, 🥈 2nd place, 🥉 3rd place, then numbers
                 - **Type**: 🔓 Open Source, 🔒 Proprietary, 📊 Baseline
                 - **MM**: 👁️ Multimodal, 📝 Text-only  
-                - **Overall**: Average accuracy across all evaluated levels
+                - **Overall**: Average accuracy across all 14 evaluated subtasks, matching Table 3 in the paper
                 - **Signal**: Low-level signal processing tasks
                 - **Perception**: Mid-level feature extraction tasks
                 - **Semantic**: High-level understanding tasks