Fix auto-extract chunk logic and add mutual exclusivity to task buttons

app.py

@@ -817,8 +817,12 @@ def run_auto_extract(input_type, spreadsheet_file, spreadsheet_column,
             divisions = min(3, max(1, num_items // 5))
             categories_per_chunk = 12
         else:
-            divisions = min(5, max(1, num_items // 3))
-            categories_per_chunk = 10
+            # Ensure each chunk has enough items for extraction
+            # We need at least categories_per_chunk items per chunk
+            divisions = max(1, num_items // 15)  # Aim for ~15 items per chunk
+            divisions = min(divisions, 5)  # Cap at 5 divisions
+            chunk_size = num_items // max(1, divisions)
+            categories_per_chunk = min(10, chunk_size - 1)  # Leave room for extraction
 
         # Extract categories
         extract_kwargs = {
@@ -1168,6 +1172,8 @@ def reset_all():
         None,  # image_folder
         "",  # image_description
         None,  # task_mode
+        gr.update(variant="secondary"),  # manual_btn - reset to unselected
+        gr.update(variant="secondary"),  # auto_extract_btn - reset to unselected
     ]
     # Reset category inputs
     for i in range(MAX_CATEGORIES):
@@ -1670,22 +1676,32 @@ Soria, C. (2025). CatLLM: A Python package for LLM-based text classification. DO
 
     # Manual entry button handler
     def select_manual():
-        return ("manual",) + update_task_visibility("manual")
+        visibility = update_task_visibility("manual")
+        return (
+            "manual",
+            gr.update(variant="primary"),    # manual_btn - selected
+            gr.update(variant="secondary"),  # auto_extract_btn - not selected
+        ) + visibility
 
     manual_btn.click(
         fn=select_manual,
         inputs=[],
-        outputs=[task_mode, auto_extract_group, categories_group, model_group, run_btn, classify_output_group, status]
+        outputs=[task_mode, manual_btn, auto_extract_btn, auto_extract_group, categories_group, model_group, run_btn, classify_output_group, status]
     )
 
     # Auto-extract button handler
     def select_auto_extract():
-        return ("auto_extract",) + update_task_visibility("auto_extract")
+        visibility = update_task_visibility("auto_extract")
+        return (
+            "auto_extract",
+            gr.update(variant="secondary"),  # manual_btn - not selected
+            gr.update(variant="primary"),    # auto_extract_btn - selected
+        ) + visibility
 
     auto_extract_btn.click(
         fn=select_auto_extract,
         inputs=[],
-        outputs=[task_mode, auto_extract_group, categories_group, model_group, run_btn, classify_output_group, status]
+        outputs=[task_mode, manual_btn, auto_extract_btn, auto_extract_group, categories_group, model_group, run_btn, classify_output_group, status]
     )
 
     # Run auto-extract button - extracts categories then shows next steps
@@ -1762,7 +1778,7 @@ Soria, C. (2025). CatLLM: A Python package for LLM-based text classification. DO
             spreadsheet_file, spreadsheet_column,
             pdf_upload_type, pdf_file, pdf_folder, pdf_description, pdf_mode,
             image_upload_type, image_file, image_folder, image_description,
-            task_mode
+            task_mode, manual_btn, auto_extract_btn
         ] + category_inputs + [
             add_category_btn, category_count,
             auto_extract_group, max_categories, auto_extract_status,