🔧 Fix web interface accuracy by removing redundant answer extraction

Fixed critical issue where solve_question() output was being double-processed,
causing accuracy to drop from 90% to 30%. The solve_question method already
returns clean, processed answers, so removed redundant _extract_answer() call.
Also fixed import paths to ensure GAIASolver initializes properly.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

app/app.py

@@ -17,6 +17,7 @@ from pathlib import Path
 
 # Add current directory to Python path to find main modules
 sys.path.insert(0, '/home/user/app')
+sys.path.insert(0, '/home/user')
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@@ -138,21 +139,19 @@ class AdvancedGAIAAgent:
                     "question": question,
                     "file_name": ""
                 }
-                result = self.solver.solve_question(question_data)
-                answer = self._extract_answer(result)
+                # solve_question already returns a clean, processed answer string
+                answer = self.solver.solve_question(question_data)
             elif self.solver == "refactored":
                 # For refactored architecture
                 try:
                     from main_refactored import main as refactored_main
-                    result = refactored_main(question)
-                    answer = self._extract_answer(result)
+                    answer = refactored_main(question)
                 except Exception as e:
                     print(f"Refactored solver error: {e}")
                     answer = f"Refactored solver error: {e}"
             elif hasattr(self.solver, '__call__'):
                 # Generic callable solver
-                result = self.solver(question)
-                answer = self._extract_answer(result)
+                answer = self.solver(question)
             else:
                 # Last resort
                 answer = "Unable to process question with current solver"
@@ -260,9 +259,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                 "Our Answer": submitted_answer[:50] + "..." if len(str(submitted_answer)) > 50 else submitted_answer,
                 "Expected Answer": correct_answer,
                 "Result": f"{validation_result['icon']} {validation_result['status']}",
-                "Score": f"{validation_result['score']:.1f}",
-                "Level": level,
-                "Time (s)": f"{question_time:.2f}"
+                "Time (s)": f"{question_time:.2f}",
+                "_score": validation_result['score']  # Keep for calculation but don't display
             })
             print(f"✅ Completed in {question_time:.2f}s - {validation_result['icon']} {validation_result['status']}")
             
@@ -274,9 +272,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                 "Our Answer": f"ERROR: {e}",
                 "Expected Answer": correct_answers.get(task_id, {}).get('answer', 'Not available'),
                 "Result": "❌ ERROR",
-                "Score": "0.0",
-                "Level": f"Level {correct_answers.get(task_id, {}).get('level', 'Unknown')}",
-                "Time (s)": "Error"
+                "Time (s)": "Error",
+                "_score": 0.0  # Keep for calculation but don't display
             })
 
     total_time = time.time() - start_time
@@ -289,12 +286,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     
     for result in results_log:
         try:
-            score = float(result.get('Score', '0.0'))
+            score = float(result.get('_score', 0.0))
             total_score += score
             validated_count += 1
             if score >= 1.0:
                 correct_count += 1
-        except ValueError:
+        except (ValueError, TypeError):
             pass
     
     local_accuracy = (total_score / validated_count * 100) if validated_count > 0 else 0
@@ -306,7 +303,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     if not answers_payload:
         print("❌ Agent did not produce any answers to submit.")
-        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
+        display_results = [{k: v for k, v in result.items() if not k.startswith('_')} for result in results_log]
+        return "Agent did not produce any answers to submit.", pd.DataFrame(display_results)
 
     # 4. Prepare Submission
     submission_data = {
@@ -347,7 +345,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             f"- Features: Enhanced reasoning, 42 specialized tools, domain expertise"
         )
         print("✅ Submission successful.")
-        results_df = pd.DataFrame(results_log)
+        # Create DataFrame excluding hidden score field
+        display_results = [{k: v for k, v in result.items() if not k.startswith('_')} for result in results_log]
+        results_df = pd.DataFrame(display_results)
         return final_status, results_df
         
     except requests.exceptions.HTTPError as e: