src/app.py

import streamlit as st
import requests
import os
import re
import io
import time
import contextlib
import zipfile
import tracker
import rag_engine
import doc_loader 
import modules.admin_panel as admin_panel

from openai import OpenAI
from google import genai
from google.genai import types
from datetime import datetime
from test_integration import run_tests
from core.QuizEngine import QuizEngine
from core.PineconeManager import PineconeManager
from huggingface_hub import hf_hub_download

# --- CONFIGURATION ---
st.set_page_config(page_title="Navy AI Toolkit", page_icon="⚓", layout="wide")

API_URL_ROOT = os.getenv("API_URL") 
OPENAI_KEY = os.getenv("OPENAI_API_KEY") 
GOOGLE_KEY = os.getenv("GOOGLE_API_KEY") # NEW: Google Key

# --- INITIALIZATION ---
if "roles" not in st.session_state:
    st.session_state.roles = []

if "quiz_state" not in st.session_state:
    st.session_state.quiz_state = {
        "active": False, "question_data": None, "user_answer": "", 
        "feedback": None, "streak": 0, "generated_question_text": ""
    }

if "quiz_history" not in st.session_state: st.session_state.quiz_history = []
if "active_index" not in st.session_state: st.session_state.active_index = None

# Debug State Variables
if "last_prompt_sent" not in st.session_state: st.session_state.last_prompt_sent = ""
if "last_context_used" not in st.session_state: st.session_state.last_context_used = ""

# --- FLATTENER LOGIC ---
class OutlineProcessor:
    """Parses text outlines for the Flattener tool."""
    def __init__(self, file_content):
        self.raw_lines = file_content.split('\n')
    
    def _is_list_item(self, line):
        pattern = r"^\s*(\d+\.|[a-zA-Z]\.|-|\*)\s+"
        return bool(re.match(pattern, line))

    def _merge_multiline_items(self):
        merged_lines = []
        for line in self.raw_lines:
            stripped = line.strip()
            if not stripped: continue
            if not merged_lines:
                merged_lines.append(line)
                continue
            if not self._is_list_item(line):
                merged_lines[-1] = merged_lines[-1].rstrip() + " " + stripped
            else:
                merged_lines.append(line)
        return merged_lines

    def parse(self):
        clean_lines = self._merge_multiline_items()
        stack = []
        results = []
        for line in clean_lines:
            stripped = line.strip()
            indent = len(line) - len(line.lstrip())
            while stack and stack[-1]['indent'] >= indent:
                stack.pop()
            stack.append({'indent': indent, 'text': stripped})
            if len(stack) > 1:
                context_str = " > ".join([item['text'] for item in stack[:-1]])
            else:
                context_str = "ROOT"
            results.append({"context": context_str, "target": stripped})
        return results

# --- HELPER FUNCTIONS ---
def query_model_universal(messages, max_tokens, model_choice, user_key=None):
    """Unified router for Chat, Tools, and Quiz."""
    
    # 1. DEBUG CAPTURE
    if messages and messages[-1]['role'] == 'user':
        st.session_state.last_prompt_sent = messages[-1]['content']

    # --- ROUTE 1: GOOGLE GEMINI (NEW) ---
    if "Gemini" in model_choice:
        # Use System Key (Env Var) or User Override if you allow it
        # For now, we strictly use the Hugging Face Secret as requested
        if not GOOGLE_KEY: return "[Error: No GOOGLE_API_KEY found in Secrets]", None
        
        try:
            client = genai.Client(api_key=GOOGLE_KEY)
            
            # Convert Chat History to Single String for 'generate_content'
            # (Gemini supports chat history objects, but string is more robust for RAG contexts)
            full_prompt = ""
            for m in messages:
                role = m["role"].upper()
                content = m["content"]
                full_prompt += f"{role}: {content}\n\n"
            full_prompt += "ASSISTANT: "

            # RETRY LOGIC (User Provided)
            max_retries = 3 # Slightly conservative for UI responsiveness
            model_id = "gemini-2.0-flash" # or "gemini-1.5-pro" depending on your access

            for attempt in range(max_retries):
                try:
                    response = client.models.generate_content(
                        model=model_id,
                        contents=full_prompt,
                        config=types.GenerateContentConfig(
                            max_output_tokens=max_tokens,
                            temperature=0.3
                        )
                    )
                    # Usage tracking is different for Gemini, we estimate or grab from response if available
                    # usage_meta = response.usage_metadata (if available)
                    return response.text.strip(), {"input": 0, "output": 0} 
                    
                except Exception as e:
                    error_msg = str(e)
                    if "429" in error_msg or "RESOURCE_EXHAUSTED" in error_msg:
                        wait_time = 10 # Short wait
                        time.sleep(wait_time)
                    else:
                        return f"[Gemini Error: {error_msg}]", None
            
            return "[Error: Gemini Rate Limit Exceeded]", None

        except Exception as e:
             return f"[Gemini Client Error: {e}]", None

    # --- ROUTE 2: OPENAI GPT-4o ---
    elif "GPT-4o" in model_choice:
        key = user_key if user_key else OPENAI_KEY
        if not key: return "[Error: No OpenAI API Key]", None
        
        client = OpenAI(api_key=key)
        try:
            resp = client.chat.completions.create(
                model="gpt-4o", max_tokens=max_tokens, messages=messages, temperature=0.3
            )
            usage = {"input": resp.usage.prompt_tokens, "output": resp.usage.completion_tokens}
            return resp.choices[0].message.content, usage
        except Exception as e:
            return f"[OpenAI Error: {e}]", None

    
    # --- ROUTE 3: LOCAL/OPEN SOURCE ---
    else:
        model_map = {
            "Granite 4 (IBM)": "granite4:latest", 
            "Llama 3.2 (Meta)": "llama3.2:latest", 
            "Gemma 3 (Google)": "gemma3:latest"
        }
        tech_name = model_map.get(model_choice)
        if not tech_name: return "[Error: Model Map Failed]", None
        
        url = f"{API_URL_ROOT}/generate"
        
        hist = ""
        sys_msg = "You are a helpful assistant."
        for m in messages:
            if m['role']=='system': sys_msg = m['content']
            elif m['role']=='user': hist += f"User: {m['content']}\n"
            elif m['role']=='assistant': hist += f"Assistant: {m['content']}\n"
        hist += "Assistant: "
        
        try:
            r = requests.post(url, json={"text": hist, "persona": sys_msg, "max_tokens": max_tokens, "model": tech_name}, timeout=600)
            if r.status_code == 200:
                d = r.json()
                return d.get("response", ""), d.get("usage", {"input":0,"output":0})
            return f"[Local Error {r.status_code}]", None
        except Exception as e:
            return f"[Conn Error: {e}]", None

def update_sidebar_metrics():
    if metric_placeholder:
        stats = tracker.get_daily_stats()
        u_stats = stats["users"].get(st.session_state.username, {"input":0, "output":0})
        metric_placeholder.metric("My Tokens Today", u_stats["input"] + u_stats["output"])

def generate_study_guide_md(history):
    md = "# ⚓ Study Guide\n\nGenerated: " + datetime.now().strftime('%Y-%m-%d %H:%M') + "\n\n"
    for item in history:
        md += f"## Q: {item['question']}\n**Your Answer:** {item['user_answer']}\n\n**Grade:** {item['grade']}\n\n**Context/Correct Info:**\n> {item['context']}\n\n---\n\n"
    return md

# --- LOGIN ---
if "authentication_status" not in st.session_state or st.session_state["authentication_status"] is None:
    login_tab, register_tab = st.tabs(["🔑 Login", "📝 Register"])
    with login_tab:
        if tracker.check_login():
            if "last_user" in st.session_state and st.session_state.last_user != st.session_state.username:
                st.session_state.messages = []
                st.session_state.user_openai_key = None
            st.session_state.last_user = st.session_state.username
            tracker.download_user_db(st.session_state.username)
            st.rerun()
    with register_tab:
        st.header("Create Account")
        with st.form("reg_form"):
            new_user = st.text_input("Username"); new_name = st.text_input("Display Name")
            new_email = st.text_input("Email"); new_pwd = st.text_input("Password", type="password")
            invite = st.text_input("Invitation Passcode") 
            if st.form_submit_button("Register"):
                success, msg = tracker.register_user(new_email, new_user, new_name, new_pwd, invite)
                if success: st.success(msg)
                else: st.error(msg)
    if not st.session_state.get("authentication_status"): st.stop()

# --- SIDEBAR ---
metric_placeholder = None
with st.sidebar:
    st.header("👤 User Profile")
    st.write(f"Welcome, **{st.session_state.name}**")
    st.header("📊 Usage Tracker")
    metric_placeholder = st.empty()
    
    if "admin" in st.session_state.roles:
        admin_panel.render_admin_sidebar()
            
    st.divider()
    st.header("🌲 Pinecone Settings")
    pc_key = os.getenv("PINECONE_API_KEY")
    if pc_key:
        pm = PineconeManager(pc_key)
        indexes = pm.list_indexes()
        selected_index = st.selectbox("Active Index", indexes)
        st.session_state.active_index = selected_index
        if selected_index:
            current_model = st.session_state.get("active_embed_model", "sentence-transformers/all-MiniLM-L6-v2")
            try:
                emb_fn = rag_engine.get_embedding_func(current_model)
                test_vec = emb_fn.embed_query("test")
                active_model_dim = len(test_vec)
                if pm.check_dimension_compatibility(selected_index, active_model_dim): st.caption(f"✅ Compatible ({active_model_dim}d)")
                else: st.error(f"❌ Mismatch! Model: {active_model_dim}d")
            except Exception as e: st.caption(f"⚠️ Check failed: {e}")
        with st.expander("Create New Index"):
            new_idx_name = st.text_input("Index Name")
            new_idx_dim = st.selectbox("Dimension", [384, 768, 1024, 1536, 3072], index=0)
            if st.button("Create"):
                with st.spinner("Provisioning..."):
                    ok, msg = pm.create_index(new_idx_name, dimension=new_idx_dim)
                    if ok: st.success(msg); time.sleep(2); st.rerun()
                    else: st.error(msg)
    else: st.warning("No Pinecone Key")
    
    st.header("🧠 Intelligence")
    st.subheader("1. Embeddings")
    embed_options = {
        "Standard (All-MiniLM, 384d)": "sentence-transformers/all-MiniLM-L6-v2",
        "High-Perf (MPNet, 768d)": "sentence-transformers/all-mpnet-base-v2",
        "OpenAI Small (1536d)": "text-embedding-3-small",
        "Custom Navy (BGE, 768d)": "NavyDevilDoc/navy-custom-models/bge-finetuned" 
    }
    embed_choice_label = st.selectbox("Select Embedding Model", list(embed_options.keys()))
    st.session_state.active_embed_model = embed_options[embed_choice_label]
    
    st.subheader("2. Chat Model")
    # Base local models
    model_map = {"Granite 4 (IBM)": "granite4:latest", 
                 "Llama 3.2 (Meta)": "llama3.2:latest", 
                 "Gemma 3 (Google)": "gemma3:latest"}
    opts = list(model_map.keys())
    
    is_admin = "admin" in st.session_state.roles
    user_key = None
    
    # Logic for Premium Models
    if not is_admin:
        user_key = st.text_input("Unlock GPT-4o", type="password")
        st.session_state.user_openai_key = user_key if user_key else None
    else: st.session_state.user_openai_key = None
    
    # Add Premium Options if Admin or Key provided
    if is_admin or st.session_state.get("user_openai_key"):
        opts.append("GPT-4o (Omni)")
    
    # Add Gemini if Key exists (System wide)
    if GOOGLE_KEY:
        opts.append("Gemini 2.5 (Google)")

    model_choice = st.radio("Select Model:", opts, key="model_selector_radio")
    st.info(f"Connected to: **{model_choice}**")
    st.divider()
    if st.session_state.authenticator: st.session_state.authenticator.logout(location='sidebar')

update_sidebar_metrics()

# --- MAIN APP ---
st.title("⚓ Navy AI Toolkit")
tab1, tab2, tab3 = st.tabs(["💬 Chat Playground", "📂 Knowledge & Tools", "⚡ Quiz Mode"])

# === TAB 1: CHAT ===
with tab1:
    # 1. LAYOUT: Header + Placeholder for Download Button
    col_header, col_btn = st.columns([6, 1])
    with col_header:
        st.header("Discussion & Analysis")
    download_placeholder = col_btn.empty()
    
    if "messages" not in st.session_state: st.session_state.messages = []
    
    # RENDER DEBUG OVERLAY (If enabled in Admin)
    admin_panel.render_debug_overlay("Chat Tab")

    c1, c2 = st.columns([3, 1])
    with c1: st.caption(f"Active Model: **{st.session_state.get('model_selector_radio', 'Granite')}**")
    with c2: use_rag = st.toggle("Enable Knowledge Base", value=False)
    
    for msg in st.session_state.messages:
        with st.chat_message(msg["role"]): st.markdown(msg["content"])
        
    if prompt := st.chat_input("Input command..."):
        st.session_state.messages.append({"role": "user", "content": prompt})
        with st.chat_message("user"): st.markdown(prompt)
        context_txt = ""
        sys_p = "You are a helpful AI assistant."
        st.session_state.last_context_used = "" # Reset context debug

        if use_rag:
            if not st.session_state.active_index: st.error("⚠️ Please select an Active Index in the sidebar first.")
            else:
                with st.spinner("Searching Knowledge Base..."):
                    docs = rag_engine.search_knowledge_base(
                        query=prompt, 
                        username=st.session_state.username,
                        index_name=st.session_state.active_index,
                        embed_model_name=st.session_state.active_embed_model
                    )
                    if docs:
                        sys_p = "You are a Navy Document Analyst. Answer based PRIMARILY on the Context."
                        for i, d in enumerate(docs):
                            src = d.metadata.get('source', 'Unknown')
                            context_txt += f"<document index='{i+1}' source='{src}'>\n{d.page_content}\n</document>\n"
                        st.session_state.last_context_used = context_txt
        
        if context_txt:
            final_prompt = f"User Question: {prompt}\n\n<context>\n{context_txt}\n</context>\n\nInstruction: Answer using the context above."
        else: final_prompt = prompt
        
        with st.chat_message("assistant"):
            with st.spinner("Thinking..."):
                hist = [{"role":"system", "content":sys_p}] + st.session_state.messages[-6:-1] + [{"role":"user", "content":final_prompt}]
                resp, usage = query_model_universal(hist, 2000, model_choice, st.session_state.get("user_openai_key"))
                st.markdown(resp)
                if usage:
                    m_name = "GPT-4o" if "GPT-4o" in model_choice else model_choice.split()[0]
                    tracker.log_usage(m_name, usage["input"], usage["output"])
                    update_sidebar_metrics()
        
        st.session_state.messages.append({"role": "assistant", "content": resp})
        if use_rag and context_txt:
            with st.expander("📚 View Context Used"): st.text(context_txt)

    # 3. LATE RENDER: Fill Download Button
    if st.session_state.messages:
        chat_log = f"# ⚓ Navy AI Toolkit - Chat Log\nDate: {datetime.now().strftime('%Y-%m-%d %H:%M')}\nModel: {st.session_state.get('model_selector_radio', 'Unknown')}\n\n---\n\n"
        for msg in st.session_state.messages:
            chat_log += f"**{msg['role'].upper()}**: {msg['content']}\n\n"
        with download_placeholder:
            st.download_button("💾 Save", chat_log, f"chat_{datetime.now().strftime('%Y%m%d_%H%M')}.md", "text/markdown")

# === TAB 2: KNOWLEDGE & TOOLS ===
with tab2:
    st.header("Document Processor")
    c1, c2 = st.columns([1, 1])
    with c1: uploaded_file = st.file_uploader("Upload File", type=["pdf", "docx", "pptx", "txt", "md"])
    with c2:
        use_vision = st.toggle("👁️ Enable Vision Mode")
        if use_vision and "GPT-4o" not in opts: st.warning("Vision requires OpenAI.")
        
    if uploaded_file:
        temp_path = rag_engine.save_uploaded_file(uploaded_file, st.session_state.username)
        col_a, col_b, col_c = st.columns(3)
        
        # COLUMN A: Ingest
        with col_a:
            chunk_strategy = st.selectbox("Chunking Strategy", ["paragraph", "token"])
            if st.button("📥 Add to KB", type="primary"):
                if not st.session_state.active_index: st.error("Select Index first.")
                else:
                    with st.spinner("Ingesting..."):
                        ok, msg = rag_engine.ingest_file(temp_path, st.session_state.username, st.session_state.active_index, st.session_state.active_embed_model, chunk_strategy)
                        if ok: tracker.upload_user_db(st.session_state.username); st.success(msg)
                        else: st.error(msg)
        
        # COLUMN B: Summarize
        with col_b:
            st.write(""); st.write("")
            if st.button("📝 Summarize"):
                with st.spinner("Summarizing..."):
                    key = st.session_state.get("user_openai_key") or OPENAI_KEY
                    class FileObj:
                        def __init__(self, p, n): self.path=p; self.name=n
                        def read(self): 
                            with open(self.path, "rb") as f: return f.read()
                    raw = doc_loader.extract_text_from_file(FileObj(temp_path, uploaded_file.name), use_vision=use_vision, api_key=key)
                    prompt = f"Summarize:\n\n{raw[:20000]}"
                    msgs = [{"role":"user", "content": prompt}]
                    summ, usage = query_model_universal(msgs, 1000, model_choice, st.session_state.get("user_openai_key"))
                    st.subheader("Summary"); st.markdown(summ)
        
        # COLUMN C: Flatten
        with col_c:
            st.write(""); st.write("")
            if "flattened_result" not in st.session_state: st.session_state.flattened_result = None
            
            if st.button("📄 Flatten"):
                with st.spinner("Flattening..."):
                    key = st.session_state.get("user_openai_key") or OPENAI_KEY
                    
                    # 1. Read File
                    with open(temp_path, "rb") as f:
                        class Wrapper:
                            def __init__(self, data, n): self.data=data; self.name=n
                            def read(self): return self.data
                        raw = doc_loader.extract_text_from_file(Wrapper(f.read(), uploaded_file.name), use_vision=use_vision, api_key=key)
                    
                    # 2. Parse Outline (This was missing logic previously)
                    proc = OutlineProcessor(raw)
                    items = proc.parse()
                    
                    # 3. Process Items
                    out_txt = []
                    bar = st.progress(0)
                    for i, item in enumerate(items):
                        p = f"Context: {item['context']}\nTarget: {item['target']}\nRewrite as one sentence."
                        m = [{"role":"user", "content": p}]
                        res, _ = query_model_universal(m, 300, model_choice, st.session_state.get("user_openai_key"))
                        out_txt.append(res)
                        bar.progress((i+1)/len(items))
                    
                    final_flattened_text = "\n".join(out_txt)
                    st.session_state.flattened_result = {"text": final_flattened_text, "source": f"{uploaded_file.name}_flat"}
                    st.rerun()

            if st.session_state.flattened_result:
                res = st.session_state.flattened_result
                st.success("Complete!")
                st.text_area("Result", res["text"], height=200)
                if st.button("📥 Index Flat"):
                    if not st.session_state.active_index: 
                        st.error("Please select an Active Index.")
                    else:
                        with st.spinner("Indexing..."):
                            # FIX: Pass the active_embed_model here!
                            ok, msg = rag_engine.process_and_add_text(
                                text=res["text"], 
                                source_name=res["source"], 
                                username=st.session_state.username,
                                index_name=st.session_state.active_index,
                                embed_model_name=st.session_state.active_embed_model
                            )
                            if ok: 
                                tracker.upload_user_db(st.session_state.username)
                                st.success(msg)
                            else: 
                                st.error(msg)
                            
    st.divider()
    st.subheader("Database Management")
    c1, c2 = st.columns([2, 1])
    with c1: st.info("Missing local files? Resync below.")
    with c2:
        if st.button("🔄 Resync from Pinecone"):
            if not st.session_state.active_index: st.error("Select Index.")
            else:
                with st.spinner("Resyncing..."):
                    ok, msg = rag_engine.rebuild_cache_from_pinecone(st.session_state.username, st.session_state.active_index)
                    if ok: st.success(msg); time.sleep(1); st.rerun()
                    else: st.error(msg)
                        
    docs = rag_engine.list_documents(st.session_state.username) 
    if docs:
        for d in docs:
            c1, c2 = st.columns([4,1])
            c1.text(f"📄 {d['filename']}")
            if c2.button("🗑️", key=d['source']):
                if not st.session_state.active_index: st.error("Select Index.")
                else:
                    rag_engine.delete_document(st.session_state.username, d['source'], st.session_state.active_index)
                    tracker.upload_user_db(st.session_state.username); st.rerun()
    else: st.warning("Cache Empty.")
        
# === TAB 3: QUIZ MODE ===
with tab3:
    st.header("⚓ Qualification Board Simulator")
    admin_panel.render_debug_overlay("Quiz Tab")
    
    col_mode, col_streak = st.columns([3, 1])
    with col_mode: 
        quiz_mode = st.radio("Mode:", ["⚡ Acronym Lightning Round", "📖 Document Deep Dive"], horizontal=True)
    
    if "Document" in quiz_mode: 
        focus_topic = st.text_input("🎯 Focus Topic", placeholder="e.g., PPBE...", help="Leave empty for random.")
    else: 
        focus_topic = None

    if "last_quiz_mode" not in st.session_state: st.session_state.last_quiz_mode = quiz_mode
    if "quiz_trigger" not in st.session_state: st.session_state.quiz_trigger = False
    
    if st.session_state.last_quiz_mode != quiz_mode:
        st.session_state.quiz_state["active"] = False
        st.session_state.quiz_state["question_data"] = None
        st.session_state.quiz_state["feedback"] = None
        st.session_state.quiz_state["generated_question_text"] = ""
        st.session_state.last_quiz_mode = quiz_mode
        st.rerun()
    
    quiz = QuizEngine()
    qs = st.session_state.quiz_state
    
    with col_streak: 
        st.metric("Streak", qs["streak"])
        if st.button("Reset"): qs["streak"] = 0
        
    if st.session_state.quiz_history:
        with st.expander(f"📚 Review Study Guide ({len(st.session_state.quiz_history)})"):
            st.download_button(
                "📥 Download Markdown", 
                generate_study_guide_md(st.session_state.quiz_history), 
                f"StudyGuide_{datetime.now().strftime('%Y%m%d')}.md"
            )
    st.divider()

    def generate_question():
        with st.spinner("Consulting Board..."):
            st.session_state.last_context_used = "" 
            
            if "Acronym" in quiz_mode:
                q_data = quiz.get_random_acronym()
                if q_data: 
                    qs["active"]=True
                    qs["question_data"]=q_data
                    qs["feedback"]=None
                    qs["generated_question_text"]=q_data["question"]
                else: 
                    st.error("No acronyms.")
            else:
                valid_question_found = False
                attempts = 0
                last_error = None
                
                while not valid_question_found and attempts < 5:
                    attempts += 1
                    q_ctx = quiz.get_document_context(st.session_state.username, topic_filter=focus_topic)
                    
                    if q_ctx and "error" in q_ctx: 
                        last_error = q_ctx["error"]
                        break
                        
                    if q_ctx:
                        # NEW: Use the Scenario Prompt
                        prompt = quiz.construct_scenario_prompt(q_ctx["context_text"])
                        st.session_state.last_context_used = q_ctx["context_text"]
                        
                        # Generate
                        response_text, usage = query_model_universal([{"role": "user", "content": prompt}], 600, model_choice, st.session_state.get("user_openai_key"))
                        
                        # PARSE OUTPUT (Scenario vs Solution)
                        if "SCENARIO:" in response_text and "SOLUTION:" in response_text:
                            parts = response_text.split("SOLUTION:")
                            scenario_text = parts[0].replace("SCENARIO:", "").strip()
                            solution_text = parts[1].strip()
                            
                            valid_question_found = True
                            qs["active"] = True
                            qs["question_data"] = q_ctx
                            qs["generated_question_text"] = scenario_text
                            qs["hidden_solution"] = solution_text 
                            qs["feedback"] = None
                        else:
                            # Fallback if model ignores format
                            valid_question_found = True
                            qs["active"] = True
                            qs["question_data"] = q_ctx
                            qs["generated_question_text"] = response_text
                            qs["hidden_solution"] = "Refer to Source Text."
                            qs["feedback"] = None

                if not valid_question_found:
                    if last_error == "topic_not_found": 
                        st.warning(f"Topic '{focus_topic}' not found.")
                    elif focus_topic: 
                        st.warning(f"Found '{focus_topic}' but could not generate question.")
                    else: 
                        st.warning("Could not generate question. Try Resync.")

    if st.session_state.quiz_trigger: 
        st.session_state.quiz_trigger = False
        generate_question()
        st.rerun()
        
    if not qs["active"]:
        if st.button("🚀 New Question", type="primary"): 
            generate_question()
            st.rerun()

    if qs["active"]:
        st.markdown(f"### {qs['generated_question_text']}")
        
        if "document" in qs.get("question_data", {}).get("type", ""): 
            st.caption(f"Source: *{qs['question_data']['source_file']}*")
            
        with st.form(key="quiz_response"):
            user_ans = st.text_area("Answer:")
            sub = st.form_submit_button("Submit")
            
        if sub and user_ans:
            with st.spinner("Board is deliberating..."):
                data = qs["question_data"]
                
                if data["type"] == "acronym":
                    prompt = quiz.construct_acronym_grading_prompt(data["term"], data["correct_definition"], user_ans)
                    final_context_for_history = data["correct_definition"]
                    msgs = [{"role": "user", "content": prompt}]
                    grade, _ = query_model_universal(msgs, 1000, model_choice, st.session_state.get("user_openai_key"))
                    qs["feedback"] = grade
                else:
                    # NEW: Scenario Grading Logic
                    scenario = qs["generated_question_text"]
                    solution = qs.get("hidden_solution", "")
                    context_ref = data["context_text"]
                    
                    prompt = quiz.construct_scenario_grading_prompt(scenario, user_ans, solution, context_ref)
                    st.session_state.last_context_used = f"SCENARIO: {scenario}\n\nSOLUTION: {solution}\n\nREF: {context_ref}"

                    msgs = [{"role": "user", "content": prompt}]
                    grade, _ = query_model_universal(msgs, 1000, model_choice, st.session_state.get("user_openai_key"))
                    qs["feedback"] = grade
                
                # Logic to determine PASS/FAIL
                is_pass = False
                if "10/10" in grade or "9/10" in grade or "8/10" in grade or "7/10" in grade or "PASS" in grade:
                    is_pass = True
                    qs["streak"] += 1
                elif "FAIL" in grade or " 6/" in grade or " 5/" in grade: 
                    qs["streak"] = 0
                else:
                    is_pass = True
                    qs["streak"] += 1 

                # Save history
                st.session_state.quiz_history.append({
                    "question": qs["generated_question_text"], 
                    "user_answer": user_ans, 
                    "grade": "PASS" if is_pass else "FAIL", # Simplified for history list
                    "context": f"**Official Solution:** {qs.get('hidden_solution', '')}\n\n**Source Text:** {data.get('context_text', '')[:500]}..."
                })
                
                st.rerun()

    if qs["feedback"]:
        st.divider()
        if "PASS" in qs["feedback"] or "7/10" in qs["feedback"] or "8/10" in qs["feedback"] or "9/10" in qs["feedback"] or "10/10" in qs["feedback"]: 
            st.success("✅ CORRECT / PASSING")
        else:
            if "FAIL" in qs["feedback"]: st.error("❌ INCORRECT")
            else: st.warning("⚠️ PARTIAL / CRITIQUE")
            
        st.markdown(qs["feedback"])
        
        data = qs["question_data"]
        if data["type"] == "acronym": 
            st.info(f"**Definition:** {data['correct_definition']}")
        elif data["type"] == "document":
            with st.expander("Show Official Solution"): 
                st.info(qs.get("hidden_solution", "No solution generated."))
            
        if st.button("Next Question ➡️"):
            st.session_state.quiz_trigger = True
            qs["active"] = False
            qs["question_data"] = None
            qs["feedback"] = None
            st.rerun()