Update src/core/QuizEngine.py

src/core/QuizEngine.py

@@ -25,58 +25,70 @@ class QuizEngine:
         }
 
     # --- MODE 2: DOCUMENTS (NEW) ---
+# ... inside QuizEngine class ...
+
     def get_document_context(self, username):
-        """
-        Picks a random file, reads it, and selects a random segment.
-        Returns the segment to be used as the 'Answer Key'.
-        """
         user_dir = os.path.join(self.source_dir, username)
-        if not os.path.exists(user_dir):
-            return None
+        if not os.path.exists(user_dir): return None
 
-        # 1. Get list of text/md files
+        # 1. Get files
         files = [f for f in os.listdir(user_dir) if f.lower().endswith(('.txt', '.md'))]
-        if not files:
-            return None
+        if not files: return None
 
-        # 2. Pick a random file
-        selected_file = random.choice(files)
-        file_path = os.path.join(user_dir, selected_file)
+        # Retry Loop: Try up to 5 times to find a "worthy" chunk
+        for attempt in range(5):
+            selected_file = random.choice(files)
+            try:
+                with open(os.path.join(user_dir, selected_file), 'r', encoding='utf-8', errors='ignore') as f:
+                    text = f.read()
 
-        try:
-            with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
-                text = f.read()
+                # DEFENSE 1: Aggressive Heuristic Filtering
+                # Split by double newline (paragraphs)
+                paragraphs = text.split('\n\n')
+                candidates = []
+                for p in paragraphs:
+                    p = p.strip()
+                    # A. Too short?
+                    if len(p) < 250: continue 
+                    # B. Looks like a list item or table row? (starts with number/bullet)
+                    if p[0].isdigit() or p.startswith(('-', '*', '•')): continue 
+                    # C. Looks like administrative noise?
+                    if "intentionally left blank" in p.lower(): continue
+                    
+                    candidates.append(p)
 
-            # 3. Split into rough paragraphs (double line break)
-            # Filter out short/empty paragraphs to avoid bad questions
-            paragraphs = [p.strip() for p in text.split('\n\n') if len(p.strip()) > 200]
-            
-            if not paragraphs:
-                return None
+                if not candidates: continue
 
-            # 4. Pick a random paragraph
-            selected_context = random.choice(paragraphs)
-            
-            return {
-                "type": "document",
-                "source_file": selected_file,
-                "context_text": selected_context
-            }
+                # Pick a random survivor
+                selected_context = random.choice(candidates)
+                
+                return {
+                    "type": "document",
+                    "source_file": selected_file,
+                    "context_text": selected_context
+                }
 
-        except Exception as e:
-            self.logger.error(f"Error fetching quiz context: {e}")
-            return None
+            except Exception as e:
+                self.logger.error(f"Error fetching context: {e}")
+                continue
+        
+        return None # Failed to find good text after 5 tries
 
-    # --- PROMPT CONSTRUCTORS ---
     def construct_question_generation_prompt(self, context_text):
-        """Asks the LLM to write the question for us."""
+        """
+        DEFENSE 2: Explicit Instructions to Ignore Trivia
+        """
         return (
             f"Act as a US Navy Engineering Duty Officer Board Examiner.\n"
-            f"Read the following technical excerpt from a Navy reference:\n\n"
-            f"'{context_text}'\n\n"
-            f"TASK: Generate a single, tough, short-answer board question based ONLY on this text.\n"
-            f"The question should test recall of specific details (e.g., 'What are the 3 documents required for...').\n"
-            f"OUTPUT: Just the question text. No preamble."
+            f"Review the following source text for suitability:\n"
+            f"'''{context_text}'''\n\n"
+            f"DECISION LOGIC:\n"
+            f"1. Does this text contain a specific Engineering Concept, Responsibility, or Procedural Rule?\n"
+            f"2. Is it free of pure administrative trivia (dates, page numbers, formatting rules)?\n\n"
+            f"IF NO: Output the word 'SKIP'.\n"
+            f"IF YES: Generate a difficult, scenario-based question that tests the candidate's understanding of the concept. "
+            f"Do not ask 'What does the text say?'. Ask 'How would you apply...?' or 'What are the requirements for...?'\n\n"
+            f"OUTPUT: Just the question text."
         )
 
     def construct_grading_prompt(self, question, answer, context_text):