Update src/streamlit_app.py

src/streamlit_app.py

@@ -79,7 +79,7 @@ TRANSLATIONS = {
         "confidence_orange": "Orange = medium confidence. The answer is partly supported but should be revised.",
         "confidence_red": "Red = low confidence. The system found weak or insufficient support.",
         "similarity_help": "Similarity is a value from 0 to 1. A higher value means the retrieved course material is closer to the question.",
-        "badge_help": "Badges are earned by repeated good performance. A topic badge is awarded after at least 3 strong answers or quiz successes in that topic.",
+        "badge_help": "Badges are earned from quiz marks only: Bronze = 70% or above in 2 quizzes of the same topic; Silver = 80% or above in 3 quizzes; Gold = 90% or above in 5 quizzes. General badges are also awarded for overall learning consistency.",
         "ask_question": "Write your free question here",
         "send": "Ask BrainChat",
         "student_tip": "Student Mode stores your quiz attempts, scores, weak topics, and badge progress.",
@@ -113,7 +113,7 @@ TRANSLATIONS = {
         "confidence_orange": "Naranja = confianza media. La respuesta tiene apoyo parcial y debe revisarse.",
         "confidence_red": "Rojo = baja confianza. El sistema encontró apoyo débil o insuficiente.",
         "similarity_help": "La similitud es un valor de 0 a 1. Un valor más alto significa que el material recuperado se parece más a la pregunta.",
-        "badge_help": "Las insignias se obtienen con buen rendimiento repetido. Una insignia de tema se concede después de al menos 3 respuestas o intentos fuertes en ese tema.",
+        "badge_help": "Las insignias se obtienen solo con las notas del cuestionario: Bronce = 70% o más en 2 cuestionarios del mismo tema; Plata = 80% o más en 3 cuestionarios; Oro = 90% o más en 5 cuestionarios. También hay insignias generales por constancia.",
         "ask_question": "Escribe aquí tu pregunta libre",
         "send": "Preguntar a BrainChat",
         "student_tip": "El modo estudiante guarda intentos, puntuaciones, temas débiles y progreso de insignias.",
@@ -412,17 +412,49 @@ def confidence_from_similarity(similarity: float) -> str:
 
 
 def badges_for_student(student_id: str) -> List[str]:
+    """
+    Quiz-score based badge system.
+
+    Topic badges:
+    - Bronze: 70% or above in at least 2 quizzes of the same topic
+    - Silver: 80% or above in at least 3 quizzes of the same topic
+    - Gold: 90% or above in at least 5 quizzes of the same topic
+
+    General badges:
+    - Consistent Learner: completed at least 10 quizzes
+    - Neurology Master: overall average score 85% or above after at least 5 quizzes
+    """
     df = load_attempts_df()
     if df.empty:
         return []
-    sdf = df[df["student_id"] == student_id]
+
+    sdf = df[df["student_id"] == student_id].copy()
+    if sdf.empty:
+        return []
+
     badges = set()
+
     for topic in TOPICS:
-        good = sdf[(sdf["topic"] == topic) & (sdf["percent"] >= 70)]
-        if len(good) >= 3:
-            badges.add(topic.split(" /")[0] + " Badge")
-    if len(sdf[sdf["percent"] >= 70]) >= 10:
-        badges.add("Clinical Reasoning Builder")
+        topic_df = sdf[sdf["topic"] == topic]
+        topic_short = topic.split(" /")[0]
+
+        bronze_count = len(topic_df[topic_df["percent"] >= 70])
+        silver_count = len(topic_df[topic_df["percent"] >= 80])
+        gold_count = len(topic_df[topic_df["percent"] >= 90])
+
+        if bronze_count >= 2:
+            badges.add(f"🥉 {topic_short} Bronze")
+        if silver_count >= 3:
+            badges.add(f"🥈 {topic_short} Silver")
+        if gold_count >= 5:
+            badges.add(f"🥇 {topic_short} Gold")
+
+    if len(sdf) >= 10:
+        badges.add("📘 Consistent Learner")
+
+    if len(sdf) >= 5 and float(sdf["percent"].mean()) >= 85:
+        badges.add("🏆 Neurology Master")
+
     return sorted(badges)
 
 # =====================================================
@@ -440,6 +472,24 @@ def safe_json_from_text(text: str):
     return json.loads(text)
 
 
+def normalize_mcq_option(opt: Any, index: int) -> str:
+    """Convert option formats into clean display strings like 'A. Text'."""
+    letter = chr(65 + index)
+    if isinstance(opt, dict):
+        opt_letter = str(opt.get("letter", letter)).strip().upper()[:1] or letter
+        text = str(opt.get("text", opt.get("option", opt.get("value", "")))).strip()
+        if not text:
+            text = str(opt)
+        return f"{opt_letter}. {text}"
+
+    text = str(opt).strip()
+    # If the model already gives 'A. text' or 'A) text', keep it clean.
+    m = re.match(r"^([A-Ea-e])\s*[\.|\)]\s*(.+)$", text)
+    if m:
+        return f"{m.group(1).upper()}. {m.group(2).strip()}"
+    return f"{letter}. {text}"
+
+
 def generate_mcqs(topic: str, difficulty: str, n_questions: int, language: str) -> Tuple[List[Dict[str, Any]], str]:
     records, err = search_hybrid(topic + " neurology PMQSN exam questions", final_k=8)
     context = build_context(records)
@@ -450,17 +500,21 @@ def generate_mcqs(topic: str, difficulty: str, n_questions: int, language: str)
         return fallback_mcqs(topic, n_questions, language), "OPENAI_API_KEY missing. Showing demo questions."
 
     lang_instruction = "Write everything in English." if language == "English" else "Escribe todo en español."
+    requested_from_model = n_questions + 2  # ask for a few extra, then keep exactly the selected number
     prompt = f"""
 You are BrainChat, an exam-focused neurology tutor.
-Generate {n_questions} MCQ questions for the topic: {topic}.
+Generate at least {requested_from_model} MCQ questions for the topic: {topic}.
+The final app will keep exactly {n_questions} questions, so do not return fewer than {n_questions}.
 Difficulty: {difficulty}.
 {lang_instruction}
 
 Rules:
-- Output ONLY valid JSON array.
+- Output ONLY a valid JSON array.
 - Each item must have: question, options, correct_option, explanation, subtopic.
 - options must be exactly 5 options labelled A, B, C, D, E.
+- The options may be strings only, for example: ["A. ...", "B. ...", "C. ...", "D. ...", "E. ..."].
 - Only one correct answer.
+- Avoid generic demo questions such as "Option A".
 - The style should follow PMQSN neurology exam questions.
 - Use the course context as the main knowledge source.
 - Do not mention JSON, files, or internal retrieval.
@@ -475,38 +529,91 @@ Course context:
         resp = client.chat.completions.create(
             model=OPENAI_MODEL,
             messages=[{"role": "user", "content": prompt}],
-            temperature=0.35,
+            temperature=0.30,
         )
         mcqs = safe_json_from_text(resp.choices[0].message.content or "[]")
-        clean = []
-        for item in mcqs[:n_questions]:
+        clean: List[Dict[str, Any]] = []
+
+        for item in mcqs:
+            if len(clean) >= n_questions:
+                break
             opts = item.get("options", [])
             if isinstance(opts, dict):
-                opts = [f"{k}. {v}" for k, v in opts.items()]
-            opts = [str(o) for o in opts][:5]
-            while len(opts) < 5:
-                opts.append(f"Option {chr(65+len(opts))}")
-            correct = str(item.get("correct_option", "A")).strip().upper()[:1]
+                # Handles {"A":"...", "B":"..."} format
+                opts = [{"letter": k, "text": v} for k, v in opts.items()]
+            if not isinstance(opts, list):
+                continue
+
+            formatted_opts = [normalize_mcq_option(opt, idx) for idx, opt in enumerate(opts[:5])]
+            if len(formatted_opts) != 5:
+                continue
+
+            question_text = str(item.get("question", "")).strip()
+            if not question_text:
+                continue
+
+            correct = str(item.get("correct_option", item.get("answer", "A"))).strip().upper()[:1]
             if correct not in list("ABCDE"):
                 correct = "A"
+
+            explanation = str(item.get("explanation", "")).strip()
+            if not explanation:
+                explanation = "Review the selected topic and compare the clinical features carefully."
+
             clean.append({
-                "question": str(item.get("question", "")),
-                "options": opts,
+                "question": question_text,
+                "options": formatted_opts,
                 "correct_option": correct,
-                "explanation": str(item.get("explanation", "")),
-                "subtopic": str(item.get("subtopic", topic)),
+                "explanation": explanation,
+                "subtopic": str(item.get("subtopic", topic)).strip() or topic,
             })
-        # Sometimes the model returns one fewer question than requested.
-        # Complete the quiz with fallback questions so the selected number is always respected.
+
         if len(clean) < n_questions:
-            clean.extend(fallback_mcqs(topic, n_questions - len(clean), language))
+            # Try one more small request instead of showing generic fallback questions.
+            missing = n_questions - len(clean)
+            retry_prompt = f"""
+Generate exactly {missing} additional PMQSN-style neurology MCQs for topic: {topic}.
+{lang_instruction}
+Return ONLY valid JSON array. Each question must have exactly 5 string options labelled A-E, one correct_option, explanation, and subtopic.
+Do not use generic placeholders.
+"""
+            retry = client.chat.completions.create(
+                model=OPENAI_MODEL,
+                messages=[{"role": "user", "content": retry_prompt}],
+                temperature=0.30,
+            )
+            more = safe_json_from_text(retry.choices[0].message.content or "[]")
+            for item in more:
+                if len(clean) >= n_questions:
+                    break
+                opts = item.get("options", [])
+                if isinstance(opts, dict):
+                    opts = [{"letter": k, "text": v} for k, v in opts.items()]
+                if not isinstance(opts, list):
+                    continue
+                formatted_opts = [normalize_mcq_option(opt, idx) for idx, opt in enumerate(opts[:5])]
+                if len(formatted_opts) != 5:
+                    continue
+                question_text = str(item.get("question", "")).strip()
+                if not question_text:
+                    continue
+                correct = str(item.get("correct_option", item.get("answer", "A"))).strip().upper()[:1]
+                if correct not in list("ABCDE"):
+                    correct = "A"
+                clean.append({
+                    "question": question_text,
+                    "options": formatted_opts,
+                    "correct_option": correct,
+                    "explanation": str(item.get("explanation", "Review the selected topic carefully.")),
+                    "subtopic": str(item.get("subtopic", topic)).strip() or topic,
+                })
+
         if clean:
-            return clean[:n_questions], ""
+            return clean[:n_questions], "" if len(clean) >= n_questions else "Generated fewer questions than requested."
         return fallback_mcqs(topic, n_questions, language), "Could not generate AI quiz. Showing demo questions."
     except Exception as e:
         return fallback_mcqs(topic, n_questions, language), f"AI generation failed: {e}"
 
-
 def fallback_mcqs(topic: str, n: int, language: str) -> List[Dict[str, Any]]:
     if language == "Spanish":
         q = f"Pregunta de práctica sobre {topic}: ¿cuál opción es más correcta?"
@@ -578,7 +685,7 @@ def html_report_student(student_id: str, name: str, language: str) -> str:
 <p><span class='orange'><b>Orange</b></span>: medium confidence / needs revision.</p>
 <p><span class='red'><b>Red</b></span>: low confidence / weak support or low performance.</p>
 <p><b>Similarity</b> is from 0 to 1 and shows how closely course material matched the question.</p>
-<p><b>Badges</b> are earned after repeated good performance in a topic.</p></div>
+<p><b>Badges</b> are earned from quiz marks only: Bronze = 70% or above in 2 quizzes of the same topic; Silver = 80% or above in 3 quizzes; Gold = 90% or above in 5 quizzes. General badges are awarded for consistency and strong overall performance.</p></div>
 <div class='card'><h2>Quiz attempts</h2><table><tr><th>Date</th><th>Topic</th><th>Difficulty</th><th>Score</th><th>Percent</th><th>Confidence</th></tr>{rows}</table></div>
 </body></html>
 """