Update LLM.py

LLM.py

@@ -1,4 +1,4 @@
-# LLM.py (Updated to V19.4 - Monte Carlo Formatting Fix)
+# LLM.py (V19.5 - Remove Bias Scores from Prompt)
 import os, traceback, json, time, re
 import httpx
 from datetime import datetime
@@ -21,13 +21,11 @@ try:
 except ImportError:
     NewsFetcher = None
 
-# 🔴 --- START OF CHANGE (V19.3) --- 🔴
 # (استيراد VADER هنا أيضاً للـ type hinting)
 try:
     from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer
 except ImportError:
     SentimentIntensityAnalyzer = None
-# 🔴 --- END OF CHANGE --- 🔴
 
 
 # (تحديث الإعدادات الافتراضية لتطابق NVIDIA)
@@ -56,13 +54,13 @@ class LLMService:
                 api_key=LLM_API_KEY,
                 timeout=CLIENT_TIMEOUT
             )
-            # 🔴 --- START OF CHANGE (V19.4) --- 🔴
-            print(f"✅ [LLMService V19.4] مهيأ. النموذج: {LLM_MODEL}")
+            # 🔴 --- START OF CHANGE (V19.5) --- 🔴
+            print(f"✅ [LLMService V19.5] مهيأ. النموذج: {LLM_MODEL}")
             # 🔴 --- END OF CHANGE --- 🔴
             print(f"   -> Endpoint: {LLM_API_URL}")
         except Exception as e:
-            # 🔴 --- START OF CHANGE (V19.4) --- 🔴
-            print(f"❌ [LLMService V19.4] فشل تهيئة AsyncOpenAI: {e}")
+            # 🔴 --- START OF CHANGE (V19.5) --- 🔴
+            print(f"❌ [LLMService V19.5] فشل تهيئة AsyncOpenAI: {e}")
             # 🔴 --- END OF CHANGE --- 🔴
             traceback.print_exc()
             raise
@@ -71,11 +69,7 @@ class LLMService:
         self.r2_service: Optional[R2Service] = None
         self.learning_hub: Optional[LearningHubManager] = None
         self.news_fetcher: Optional[NewsFetcher] = None
-        
-        # 🔴 --- START OF CHANGE (V19.3) --- 🔴
-        # (إضافة متغير VADER)
         self.vader_analyzer: Optional[SentimentIntensityAnalyzer] = None
-        # 🔴 --- END OF CHANGE --- 🔴
 
     async def _call_llm(self, prompt: str) -> Optional[str]:
         """
@@ -247,14 +241,12 @@ class LLMService:
             latest_news_text = "News data unavailable for re-analysis."
             latest_news_score = 0.0
             
-            # 🔴 --- START OF CHANGE (V19.3) --- 🔴
             # (استخدام self.vader_analyzer الذي تم حقنه)
             if self.news_fetcher:
                 latest_news_text = await self.news_fetcher.get_news_for_symbol(symbol)
                 if self.vader_analyzer and latest_news_text: # (التحقق من المحلل المُمرر)
                      vader_scores = self.vader_analyzer.polarity_scores(latest_news_text)
                      latest_news_score = vader_scores.get('compound', 0.0)
-            # 🔴 --- END OF CHANGE --- 🔴
             
             current_data['latest_news_text'] = latest_news_text
             current_data['latest_news_score'] = latest_news_score
@@ -289,16 +281,15 @@ class LLMService:
                                candidate_data: Dict[str, Any], 
                                learning_context: str) -> str:
         """
-        (معدل V19.4)
+        (معدل V19.5)
         إنشاء الـ Prompt (باللغة الإنجليزية) لاتخاذ قرار التداول الأولي (Explorer).
-        (تم إصلاح تنسيق مونت كارلو)
+        (تمت إزالة الدرجات المسبقة لتقليل الانحياز)
         """
         
         symbol = candidate_data.get('symbol', 'N/A')
         
         # --- 1. استخراج بيانات ML (الطبقة 1) ---
-        l1_score = candidate_data.get('enhanced_final_score', 0)
-        l1_reasons = candidate_data.get('reasons_for_candidacy', [])
+        # (تمت إزالة l1_score و l1_reasons عمداً)
         pattern_data = candidate_data.get('pattern_analysis', {})
         mc_data = candidate_data.get('monte_carlo_distribution', {})
         
@@ -336,12 +327,16 @@ class LLMService:
 {learning_context}
 --- END OF PLAYBOOK ---
 """
-        ml_summary_prompt = f"""
-1.  **ML Analysis (Score: {l1_score:.3f}):**
-    * Reasons: {', '.join(l1_reasons)}
+        
+        # 🔴 --- START OF CHANGE (V19.5) --- 🔴
+        # (تمت إزالة درجة l1_score و l1_reasons من هنا)
+        technical_summary_prompt = f"""
+1.  **Technical Analysis:**
     * Chart Pattern: {pattern_data.get('pattern_detected', 'None')} (Conf: {pattern_data.get('pattern_confidence', 0):.2f})
     * Monte Carlo (1h): {mc_data.get('probability_of_gain', 0) * 100:.1f}% chance of profit (Expected: {mc_data.get('expected_return_pct', 0):.2f}%)
 """
+        # 🔴 --- END OF CHANGE --- 🔴
+
         news_prompt = f"""
 2.  **News & Sentiment Analysis:**
     * Market Trend: {market_trend} (BTC: {btc_sentiment})
@@ -362,24 +357,26 @@ class LLMService:
     * Relative 24h Net Flow (vs Daily Volume): {relative_net_flow_24h_percent:+.2f}%
 """
         
-        # (تم دمج جميع التعليمات هنا في رسالة الـ user)
+        # 🔴 --- START OF CHANGE (V19.5) --- 🔴
+        # (تم تحديث التعليمات ليعكس تحليل البيانات "الخام" بدلاً من الدرجات)
         task_prompt = f"""
 CONTEXT:
 You are an expert AI trading analyst (Explorer Brain). 
-Analyze the provided data for {symbol} and decide if it's a high-potential candidate to 'WATCH'.
+Analyze the following raw technical, news, and whale data for {symbol}. You must make a decision based *only* on the data provided, without any pre-calculated scores.
+Decide if this combination of signals presents a high-potential opportunity to 'WATCH'.
 {playbook_prompt}
 
 --- START OF CANDIDATE DATA ---
-{ml_summary_prompt}
+{technical_summary_prompt}
 {news_prompt}
 {whale_activity_prompt}
 --- END OF CANDIDATE DATA ---
 
 TASK:
 1.  **Internal Thinking (Private):** Perform a step-by-step analysis (as triggered by the system prompt).
-    * Synthesize all data points (ML, News, Whale Flow, 24h Accumulation).
-    * Are the signals aligned? (e.g., ML breakout + Whale Accumulation = Strong).
-    * Are there conflicts? (e.g., Good ML Score but high 24h Deposits = Risky).
+    * Synthesize all data points (Chart Pattern, Monte Carlo, News, Whale Flow, 24h Accumulation).
+    * Are the signals aligned? (e.g., Bullish Pattern + High MC Probability + Whale Accumulation = Strong).
+    * Are there conflicts? (e.g., Bullish Pattern but high 24h Deposits = Risky).
     * Consult the "Playbook" for learned rules.
 2.  **Final Decision:** Based on your internal thinking, decide the final action.
 3.  **Output Constraint:** Provide your final answer ONLY in the requested JSON object format, with no introductory text, markdown formatting, or explanations.
@@ -393,6 +390,7 @@ OUTPUT (JSON Object ONLY):
   "exit_profile": "Aggressive" or "Standard" or "Patient"
 }}
 """
+        # 🔴 --- END OF CHANGE --- 🔴
         
         # (نرسل فقط task_prompt لأنه يحتوي الآن على كل شيء)
         return task_prompt