diff --git a/application/prompt_client/prompt_client.py b/application/prompt_client/prompt_client.py
index d9e151118..1f27b276e 100644
--- a/application/prompt_client/prompt_client.py
+++ b/application/prompt_client/prompt_client.py
@@ -1194,13 +1194,14 @@ def get_id_of_most_similar_cre_paginated(
         most_similar_index = 0
         most_similar_id = ""
         for page in range(starting_page, total_pages + 1):
-            existing_cres, existing_cre_ids = self.__load_cre_embeddings(embeddings)
-
-            similarities = cosine_similarity(embedding_array, existing_cres)
-            if np.max(similarities) > max_similarity:
-                max_similarity = np.max(similarities)
-                most_similar_index = np.argmax(similarities)
-                most_similar_id = existing_cre_ids[most_similar_index]
+            if embeddings:
+                existing_cres, existing_cre_ids = self.__load_cre_embeddings(embeddings)
+
+                similarities = cosine_similarity(embedding_array, existing_cres)
+                if np.max(similarities) > max_similarity:
+                    max_similarity = np.max(similarities)
+                    most_similar_index = np.argmax(similarities)
+                    most_similar_id = existing_cre_ids[most_similar_index]
             if page < total_pages:
                 (
                     embeddings,
@@ -1256,14 +1257,15 @@ def get_id_of_most_similar_node_paginated(
         most_similar_index = 0
         most_similar_id = ""
         for page in range(starting_page, total_pages + 1):
-            existing_standards, existing_standard_ids = self.__load_node_embeddings(
-                embeddings
-            )
-            similarities = cosine_similarity(embedding_array, existing_standards)
-            if np.max(similarities) > max_similarity:
-                max_similarity = np.max(similarities)
-                most_similar_index = int(np.argmax(similarities))
-                most_similar_id = existing_standard_ids[most_similar_index]
+            if embeddings:
+                existing_standards, existing_standard_ids = self.__load_node_embeddings(
+                    embeddings
+                )
+                similarities = cosine_similarity(embedding_array, existing_standards)
+                if np.max(similarities) > max_similarity:
+                    max_similarity = np.max(similarities)
+                    most_similar_index = int(np.argmax(similarities))
+                    most_similar_id = existing_standard_ids[most_similar_index]
 
             if page < total_pages:
                 embeddings, _, _ = self.database.get_embeddings_by_doc_type_paginated(
diff --git a/application/tests/prompt_client_pgvector_similarity_test.py b/application/tests/prompt_client_pgvector_similarity_test.py
index d6e6a32fe..af4f3bee0 100644
--- a/application/tests/prompt_client_pgvector_similarity_test.py
+++ b/application/tests/prompt_client_pgvector_similarity_test.py
@@ -149,6 +149,36 @@ def test_cre_paginated_single_page(self) -> None:
         self.assertEqual(result, ("only-cre", 1.0))
         self.assertEqual(database.get_embeddings_by_doc_type_paginated.call_count, 1)
 
+    def test_node_paginated_skips_empty_final_page(self) -> None:
+        # The final page's embeddings all failed to parse and came back
+        # empty (e.g. malformed stored vectors). Must not crash
+        # cosine_similarity with a feature-count mismatch; the real match
+        # found on an earlier page must still be returned.
+        pages = {
+            1: {"target-node": self.MATCHING_VECTOR},
+            2: {},
+        }
+        handler, database = self._make_handler(can_use_pgvector=False, pages=pages)
+
+        result = handler.get_id_of_most_similar_node_paginated(
+            self.QUERY_EMBEDDING, similarity_threshold=0.5
+        )
+
+        self.assertEqual(result, ("target-node", 1.0))
+
+    def test_cre_paginated_skips_empty_final_page(self) -> None:
+        pages = {
+            1: {"target-cre": self.MATCHING_VECTOR},
+            2: {},
+        }
+        handler, database = self._make_handler(can_use_pgvector=False, pages=pages)
+
+        result = handler.get_id_of_most_similar_cre_paginated(
+            self.QUERY_EMBEDDING, similarity_threshold=0.5
+        )
+
+        self.assertEqual(result, ("target-cre", 1.0))
+
 
 class FindMostSimilarEmbeddingIdResilienceTest(unittest.TestCase):
     def test_query_error_returns_no_match(self) -> None: