make model less aggressive

moderator.py

@@ -242,7 +242,7 @@ class SimpleModerator:
             final_confidence = hate_score
         
         # If both pattern and model agree on FLAG, escalate to DELETE
-        if pattern_decision == "FLAG" and (toxic_score > 0.80 or hate_score > 0.70):
+        if pattern_decision == "FLAG" and (toxic_score > 0.95 or hate_score > 0.90):
             action = "delete"
             reason = f"Pattern + Model agreement: {reason}"
             final_confidence = max(pattern_confidence, toxic_score, hate_score)