update

src/similarity_model/hybrid_ranker.py

@@ -135,12 +135,25 @@ def compute_originality(
     """
     Originality Score (0-100)
 
-    - Driven primarily by inverse of hybrid similarity.
+    - Driven primarily by inverse of hybrid similarity (non-linear).
+    - Boosted by the uniqueness of the features.
     """
 
     hybrid_score = clamp(hybrid_score)
 
-    originality = 1.0 - hybrid_score
+    uniqueness_ratio = (
+        unique_query_features / total_query_features
+        if total_query_features > 0
+        else 1.0
+    )
+
+    # Non-linear penalty: 50% similarity drops originality slightly, 
+    # but 90% similarity drops it massively.
+    similarity_penalty = hybrid_score ** 2
+    base_originality = 1.0 - similarity_penalty
+
+    # Blend uniqueness and base originality
+    originality = (0.40 * uniqueness_ratio) + (0.60 * base_originality)
 
     return round(
         max(0.0, min(100.0, originality * 100)),