v3 prompt

app.py

@@ -137,16 +137,18 @@ def classify_text(claim):
 
             User's statement: "{claim}"
 
-            Relevant evidences: {chr(10).join(f"- {e}" for e in evidences)}
+            Information you have received (use this for reasoning, but do not mention or list it directly):
+            {chr(10).join(f"- {e}" for e in evidences)}
 
             The system’s current assessment is that the claim is: "{label_str}".
 
-            Evaluate the claim, evidences, and system's assessment. You may disagree with the system if the evidences clearly contradict the claim.
+            Now, carefully evaluate the statement and the assessment. You may disagree with the system if the evidences clearly contradict the claim.
             Write your reasoning and return it STRICTLY as a JSON object with the following fields:
             {{
                 "verdict": "Real / Fake / Uncertain",
                 "explanation": "3–5 natural sentences explaining what makes the claim true or fake or uncertain. 
-                                User does not have the context of evidences. In case you are referring to evidences, briefly summarize them.",
+                                Do NOT mention words like 'evidence', 'sources', or 'provided information'. 
+                                Instead, explain the reasoning naturally as if you are telling it from general knowledge.",",
                 "confidence": "Low / Medium / High 
                                Decide this depending on how strong the evidences are, how clear the reasoning is,
                                and how certain you are about your verdict."