update

src/similarity_model/feature_similarity.py

@@ -11,7 +11,7 @@ logger = logging.getLogger(__name__)
 MODEL_NAME = "all-MiniLM-L6-v2"
 SIMILARITY_WEIGHT = 0.70
 COVERAGE_WEIGHT = 0.30
-DEFAULT_THRESHOLD = 0.80
+DEFAULT_THRESHOLD = 0.65
 
 @lru_cache(maxsize=1)
 def load_feature_model():

src/similarity_model/hybrid_ranker.py

@@ -109,7 +109,7 @@ def compute_hybrid_score(
     # No feature overlap → rely on semantics only
     # ==========================================
     if feature_score == 0 or coverage == 0:
-        return round(clamp(0.20 * semantic_score), 4)
+        return round(clamp(0.60 * semantic_score), 4)
 
     # ==========================================
     # Normal scoring with dynamic weights
@@ -135,22 +135,12 @@ def compute_originality(
     """
     Originality Score (0-100)
 
-    - More unique features -> higher originality
-    - More similarity -> lower originality
+    - Driven primarily by inverse of hybrid similarity.
     """
 
     hybrid_score = clamp(hybrid_score)
 
-    uniqueness_ratio = (
-        unique_query_features / total_query_features
-        if total_query_features > 0
-        else 0.0
-    )
-
-    originality = (
-        0.70 * uniqueness_ratio +
-        0.30 * (1.0 - hybrid_score)
-    )
+    originality = 1.0 - hybrid_score
 
     return round(
         max(0.0, min(100.0, originality * 100)),