fix: normalize vector store scores to similarity (higher = better)

The threshold parameter in Memory.search() was silently broken for 11
vector stores because they returned raw distance scores (lower = better)
while the threshold check assumed similarity scores (higher = better).
This caused threshold filtering to be inverted — good matches got
dropped and bad matches passed through.

Convert all affected vector stores to return similarity scores:
- Cosine distance stores: score = max(0.0, 1.0 - distance)
- L2 distance stores: score = 1.0 / (1.0 + distance)
- Stores computing similarity then discarding it: return similarity directly

Fixes #4453
Related: #3283

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

tests/vector_stores/test_chroma.py

@@ -45,7 +45,7 @@ def test_search_vectors(chromadb_instance, mock_chromadb_client):
 
     assert len(results) == 2
     assert results[0].id == "id1"
-    assert results[0].score == 0.1
+    assert abs(results[0].score - (1.0 / (1.0 + 0.1))) < 1e-9  # L2 distance 0.1 -> similarity
     assert results[0].payload == {"name": "vector1"}
 
 
@@ -150,7 +150,7 @@ def test_get_vector(chromadb_instance):
     chromadb_instance.collection.get.assert_called_once_with(ids=["id1"])
 
     assert result.id == "id1"
-    assert result.score == 0.1
+    assert abs(result.score - (1.0 / (1.0 + 0.1))) < 1e-9  # L2 distance 0.1 -> similarity
     assert result.payload == {"name": "vector1"}
 
 

tests/vector_stores/test_milvus.py

@@ -129,7 +129,7 @@ class TestMilvusDB:
         # Verify results are parsed correctly
         assert len(results) == 1
         assert results[0].id == "mem1"
-        assert results[0].score == 0.8
+        assert results[0].score == 0.8  # COSINE: Milvus returns similarity directly
 
     def test_search_different_user_ids(self, milvus_db, mock_milvus_client):
         """Test that search works with different user_ids (reproduces reported bug)."""
@@ -228,10 +228,10 @@ class TestMilvusDB:
         
         assert len(parsed) == 2
         assert parsed[0].id == "mem1"
-        assert parsed[0].score == 0.9
+        assert parsed[0].score == 0.9  # COSINE: Milvus returns similarity directly
         assert parsed[0].payload == {"user_id": "alice"}
         assert parsed[1].id == "mem2"
-        assert parsed[1].score == 0.85
+        assert parsed[1].score == 0.85  # COSINE: Milvus returns similarity directly
 
     def test_update_with_none_vector_fetches_existing(self, milvus_db, mock_milvus_client):
         """Test that update with vector=None fetches the existing vector (fixes #3708)."""

tests/vector_stores/test_pgvector.py

@@ -445,9 +445,9 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 2)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[1].id, self.test_ids[1])
-        self.assertEqual(results[1].score, 0.2)
+        self.assertEqual(results[1].score, 0.8)
 
     @patch('mem0.vector_stores.pgvector.PSYCOPG_VERSION', 2)
     @patch('mem0.vector_stores.pgvector.ConnectionPool')
@@ -494,9 +494,9 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 2)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[1].id, self.test_ids[1])
-        self.assertEqual(results[1].score, 0.2)
+        self.assertEqual(results[1].score, 0.8)
 
     @patch('mem0.vector_stores.pgvector.PSYCOPG_VERSION', 3)
     @patch('mem0.vector_stores.pgvector.ConnectionPool')
@@ -1140,7 +1140,7 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 1)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[0].payload["user_id"], "alice")
         self.assertEqual(results[0].payload["agent_id"], "agent1")
         self.assertEqual(results[0].payload["run_id"], "run1")
@@ -1190,7 +1190,7 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 1)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[0].payload["user_id"], "alice")
         self.assertEqual(results[0].payload["agent_id"], "agent1")
         self.assertEqual(results[0].payload["run_id"], "run1")
@@ -1240,7 +1240,7 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 1)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[0].payload["user_id"], "alice")
 
     @patch('mem0.vector_stores.pgvector.PSYCOPG_VERSION', 2)
@@ -1288,7 +1288,7 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 1)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[0].payload["user_id"], "alice")
 
     @patch('mem0.vector_stores.pgvector.PSYCOPG_VERSION', 3)
@@ -1336,9 +1336,9 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 2)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[1].id, self.test_ids[1])
-        self.assertEqual(results[1].score, 0.2)
+        self.assertEqual(results[1].score, 0.8)
 
     @patch('mem0.vector_stores.pgvector.PSYCOPG_VERSION', 2)
     @patch('mem0.vector_stores.pgvector.ConnectionPool')
@@ -1385,9 +1385,9 @@ class TestPGVector(unittest.TestCase):
         # Verify results
         self.assertEqual(len(results), 2)
         self.assertEqual(results[0].id, self.test_ids[0])
-        self.assertEqual(results[0].score, 0.1)
+        self.assertEqual(results[0].score, 0.9)
         self.assertEqual(results[1].id, self.test_ids[1])
-        self.assertEqual(results[1].score, 0.2)
+        self.assertEqual(results[1].score, 0.8)
 
     @patch('mem0.vector_stores.pgvector.PSYCOPG_VERSION', 3)
     @patch('mem0.vector_stores.pgvector.ConnectionPool')

tests/vector_stores/test_s3_vectors.py

@@ -157,7 +157,7 @@ def test_search(mock_boto_client):
     mock_boto_client.query_vectors.assert_called_once()
     assert len(results) == 1
     assert results[0].id == "id1"
-    assert results[0].score == 0.9
+    assert abs(results[0].score - 0.1) < 1e-9  # distance 0.9 -> similarity 0.1
 
 
 def test_get(mock_boto_client):

tests/vector_stores/test_supabase.py

@@ -75,7 +75,7 @@ def test_search_vectors(supabase_instance, mock_collection):
 
     assert len(results) == 2
     assert results[0].id == "id1"
-    assert results[0].score == 0.9
+    assert abs(results[0].score - 0.1) < 1e-9  # cosine distance 0.9 -> similarity 0.1
     assert results[0].payload == {"name": "vector1"}
 
 

tests/vector_stores/test_vertex_ai_vector_search.py

@@ -112,7 +112,7 @@ def test_search_vectors(vector_store, mock_vertex_ai):
 
     assert len(results) == 1
     assert results[0].id == "test-id"
-    assert results[0].score == 0.1
+    assert results[0].score == 0.9  # distance 0.1 -> similarity 0.9
     assert results[0].payload == {"user_id": "test_user"}