Update app.py

app.py

@@ -362,16 +362,17 @@ def split_content_in_half(text: str) -> tuple:
     return first_half, second_half
 
 
-def analyze_content_halves(model_manager, text: str) -> Dict:
+def analyze_content_halves(model_manager, text: str, overall_result: Dict = None) -> Dict:
     """
     Analyze text by splitting it into two halves after cleaning
     
     Args:
         model_manager: The ModelManager instance
         text: Original text to analyze
+        overall_result: Overall classification result for variance calculation
     
     Returns:
-        Dictionary with analysis of both halves
+        Dictionary with analysis of both halves and final decision
     """
     try:
         # Clean the content first
@@ -394,6 +395,104 @@ def analyze_content_halves(model_manager, text: str) -> Dict:
         second_half_result = model_manager.classify_text(second_half)
         second_half_words = len(second_half.split())
         
+        # Extract key metrics
+        first_ai = first_half_result["ai_percentage"]
+        second_ai = second_half_result["ai_percentage"]
+        first_model = first_half_result["predicted_model"]
+        second_model = second_half_result["predicted_model"]
+        
+        # Calculate average AI score from both halves
+        avg_halves_ai_score = (first_ai + second_ai) / 2
+        
+        # Calculate variance between halves
+        variance_between_halves = abs(first_ai - second_ai)
+        
+        # Overall AI probability (use overall_result if provided, otherwise calculate)
+        overall_ai_prob = overall_result["ai_percentage"] / 100 if overall_result else avg_halves_ai_score / 100
+        
+        # ===== FINAL DECISION LOGIC =====
+        verdict = None
+        confidence = None
+        reasoning = None
+        
+        # Condition 1: Both halves < 50% AI AND second_half predicted_model is "human"
+        if first_ai < 50 and second_ai < 50 and second_model.lower() == "human":
+            verdict = "HUMAN"
+            confidence = "High" if variance_between_halves < 15 else "Medium"
+            reasoning = (
+                f"Both halves scored below 50% AI probability (First: {first_ai}%, Second: {second_ai}%). "
+                f"The second half was classified as human-written. "
+                f"Variance between halves is {variance_between_halves:.2f}%, indicating "
+                f"{'consistent human patterns' if variance_between_halves < 15 else 'some variation but still human-like'}."
+            )
+        
+        # Condition 2: Both halves > 50% AI AND second_half predicted_model is NOT "human"
+        elif first_ai > 50 and second_ai > 50 and second_model.lower() != "human":
+            verdict = "AI"
+            
+            # Determine confidence based on scores
+            if first_ai > 80 and second_ai > 80:
+                confidence = "Very High"
+            elif first_ai > 70 and second_ai > 70:
+                confidence = "High"
+            else:
+                confidence = "Medium"
+            
+            reasoning = (
+                f"Both halves scored above 50% AI probability (First: {first_ai}%, Second: {second_ai}%). "
+                f"The pattern matches {second_model} outputs. "
+                f"Variance between halves is {variance_between_halves:.2f}%, "
+                f"{'showing consistent AI patterns throughout' if variance_between_halves < 20 else 'with some variation in AI generation style'}."
+            )
+        
+        # Condition 3: Mixed results - one half AI, one half human
+        elif (first_ai > 50 and second_ai < 50) or (first_ai < 50 and second_ai > 50):
+            verdict = "MIXED"
+            confidence = "Low"
+            reasoning = (
+                f"Mixed signals detected. First half: {first_ai}% AI, Second half: {second_ai}% AI. "
+                f"One portion appears AI-generated while the other seems human-written. "
+                f"This could indicate: partial AI assistance, human editing of AI content, "
+                f"or AI completion of human-started text. High variance of {variance_between_halves:.2f}% supports mixed authorship."
+            )
+        
+        # Condition 4: Both around 50% - uncertain
+        else:
+            # Check if second_model is human but scores are borderline
+            if second_model.lower() == "human":
+                verdict = "LIKELY_HUMAN"
+                confidence = "Low"
+                reasoning = (
+                    f"Borderline case with scores near 50% threshold (First: {first_ai}%, Second: {second_ai}%). "
+                    f"Second half classified as human-written. The text shows characteristics of both "
+                    f"human and AI writing. Variance: {variance_between_halves:.2f}%."
+                )
+            else:
+                verdict = "LIKELY_AI"
+                confidence = "Low"
+                reasoning = (
+                    f"Borderline case with scores near 50% threshold (First: {first_ai}%, Second: {second_ai}%). "
+                    f"Pattern suggests {second_model} but confidence is low. "
+                    f"Variance: {variance_between_halves:.2f}%."
+                )
+        
+        # Prepare final decision structure
+        final_decision = {
+            "verdict": verdict,
+            "confidence": confidence,
+            "reasoning": reasoning,
+            "supporting_data": {
+                "overall_ai_prob": round(overall_ai_prob, 3),
+                "first_half_ai_score": round(first_ai / 100, 3),
+                "second_half_ai_score": round(second_ai / 100, 3),
+                "avg_halves_ai_score": round(avg_halves_ai_score / 100, 3),
+                "variance_between_halves": round(variance_between_halves, 2),
+                "first_half_model": first_model,
+                "second_half_model": second_model,
+                "models_agree": first_model == second_model
+            }
+        }
+        
         return {
             "halves_analysis_available": True,
             "cleaned_content": {
@@ -402,19 +501,20 @@ def analyze_content_halves(model_manager, text: str) -> Dict:
                 "second_half_words": second_half_words
             },
             "first_half": {
-                "ai_percentage": first_half_result["ai_percentage"],
+                "ai_percentage": first_ai,
                 "human_percentage": first_half_result["human_percentage"],
-                "predicted_model": first_half_result["predicted_model"],
+                "predicted_model": first_model,
                 "word_count": first_half_words,
                 "preview": first_half[:200] + "..." if len(first_half) > 200 else first_half
             },
             "second_half": {
-                "ai_percentage": second_half_result["ai_percentage"],
+                "ai_percentage": second_ai,
                 "human_percentage": second_half_result["human_percentage"],
-                "predicted_model": second_half_result["predicted_model"],
+                "predicted_model": second_model,
                 "word_count": second_half_words,
                 "preview": second_half[:200] + "..." if len(second_half) > 200 else second_half
-            }
+            },
+            "final_decision": final_decision
         }
         
     except Exception as e:
@@ -589,8 +689,8 @@ async def analyze_text(data: TextInput):
                 human_percentage = round(100 - ai_percentage, 2)
                 ai_words = int(recalc_ai_words)
         
-        # 🆕 NEW FEATURE: Analyze content by halves
-        halves_analysis = analyze_content_halves(model_manager, text)
+        # 🆕 NEW FEATURE: Analyze content by halves (pass overall result for variance calculation)
+        halves_analysis = analyze_content_halves(model_manager, text, result)
         
         # إنشاء رسالة التغذية الراجعة
         if ai_percentage > 50: