"""Unit tests for Hybrid Retrieval, Query Rewriting, and Decomposition.""" import unittest from ingest import DocumentChunk from retrieve import HybridRetriever, QueryProcessor class TestRetrieval(unittest.TestCase): @classmethod def setUpClass(cls): cls.retriever = HybridRetriever() cls.c1 = DocumentChunk( chunk_id="chunk_01", doc_id="doc_1", file_name="cloud.docx", text="Amazon Aurora PostgreSQL powers the relational database tier with multi-master replication.", page_number=1, section_title="Database Tier", chunk_index=0, token_count=15 ) cls.c2 = DocumentChunk( chunk_id="chunk_02", doc_id="doc_1", file_name="cloud.docx", text="Redis Cluster version 7.2 handles user session caching across 6 shards.", page_number=1, section_title="Caching Layer", chunk_index=1, token_count=13 ) cls.c3 = DocumentChunk( chunk_id="chunk_03", doc_id="doc_2", file_name="security.txt", text="FIDO2 WebAuthn security keys are mandatory for all production console access.", page_number=1, section_title="IAM Policy", chunk_index=2, token_count=14 ) cls.retriever.index_documents([cls.c1, cls.c2, cls.c3]) def test_dense_search(self): res = self.retriever.dense_index.search("relational database", top_k=2) self.assertGreater(len(res), 0) self.assertEqual(res[0][0].chunk_id, "chunk_01") def test_bm25_search(self): res = self.retriever.sparse_index.search("Redis cluster shards", top_k=2) self.assertGreater(len(res), 0) self.assertEqual(res[0][0].chunk_id, "chunk_02") def test_hybrid_search_with_rerank(self): res = self.retriever.retrieve( query="What hardware keys are required for production access?", initial_top_k=5, rerank_top_k=2, mode="hybrid_rerank" ) self.assertGreater(len(res["chunks"]), 0) # Top chunk should be chunk_03 (security keys) self.assertEqual(res["chunks"][0].chunk_id, "chunk_03") def test_query_rewriting(self): qp = QueryProcessor() history = [ {"role": "user", "content": "What is the replication lag for Aurora PostgreSQL?"}, {"role": "assistant", "content": "Average replication lag is less than 12 milliseconds."} ] # Coreference question should incorporate context from previous turn rewritten = qp.rewrite_query("How many shards does it use?", history) self.assertTrue(len(rewritten) > len("How many shards does it use?")) self.assertIn("Aurora", rewritten) # Standalone short query should NOT be rewritten or contaminated standalone = qp.rewrite_query("explain hyperledger", history) self.assertEqual(standalone, "explain hyperledger") # List/dict content formats from Gradio should be handled safely without TypeError list_history = [ {"role": "user", "content": [{"type": "text", "text": "What was total revenue?"}]}, {"role": "assistant", "content": ["$520M"]} ] res = qp.rewrite_query("hyperledger", list_history) self.assertEqual(res, "hyperledger") def test_query_decomposition(self): qp = QueryProcessor() # Multi-question test sub_queries = qp.decompose_query("What database is used? How does caching work?") self.assertEqual(len(sub_queries), 2) # Comparative question test comp_queries = qp.decompose_query("Compare Aurora PostgreSQL and Redis Cluster performance") self.assertGreaterEqual(len(comp_queries), 2) if __name__ == "__main__": unittest.main()