Update app.py

app.py

@@ -347,7 +347,7 @@ def run_hypothesis_testing(systems_data, selected_test_metrics, n_samples, signi
             except Exception as bert_error:
                 return f"Error testing BERTScore evaluator: {str(bert_error)}", ""
                 
-        if 'Word Count' in selected_test_metrics:
+        if 'custom: Word Count' in selected_test_metrics:
             metrics['word_count'] = word_count_metric  # ← example of a simple custom-defined metric
         
         if not metrics:
@@ -499,7 +499,7 @@ with gr.Blocks(title="Null Hypothesis Testing", theme=gr.themes.Soft()) as hypot
         with gr.Column(scale=1):
             test_metrics_selection = gr.CheckboxGroup(
                 label="🎯 Select Metrics for Testing",
-                choices=["BLEU", "ROUGE", "BERTScore", "Word Count"],
+                choices=["BLEU", "ROUGE", "BERTScore", "custom: Word Count"],
                 value=["BLEU", "ROUGE", "BERTScore"],
                 interactive=True,
                 info="Only fast metrics are shown to ensure quick evaluation (slow ones are excluded)"