updated tests

tests/test_ebook_search_core.py

@@ -19,6 +19,7 @@ from python.ebook_search.bm25_corpus import (
     BM25CorpusUnavailableError,
     BM25Manifest,
     ensure_bm25_corpus,
+    fetch_bm25_corpus_records,
     load_bm25_corpus,
 )
 from python.ebook_search.config import EbookSearchConfig, RerankConfig, load_config, normalize_embedding_model
@@ -33,7 +34,7 @@ from python.ebook_search.search import (
     search_ebooks,
 )
 from python.ebook_search.timing import RuntimeStep
-from python.orm.richie import EbookEmbeddingModel, EbookSource, RichieBase
+from python.orm.richie import EbookChapter, EbookChunk, EbookEmbeddingModel, EbookSource, RichieBase
 
 
 def test_chunk_text_uses_overlap() -> None:
@@ -86,6 +87,47 @@ def test_find_existing_source_matches_path_or_hash() -> None:
         assert find_existing_source(session, Path("/new/book.epub"), "a" * 64) == source
 
 
+def test_bm25_corpus_uses_existing_search_text_without_duplicate_metadata() -> None:
+    engine = create_engine("sqlite+pysqlite:///:memory:", future=True)
+    RichieBase.metadata.create_all(engine)
+    with sessionmaker(bind=engine, expire_on_commit=False, future=True)() as session:
+        source = EbookSource(
+            title="Book",
+            author="Author",
+            language=None,
+            publisher=None,
+            identifier=None,
+            file_path="/book.epub",
+            file_sha256="a" * 64,
+            file_mtime=datetime.now(tz=UTC),
+            file_size=10,
+        )
+        session.add(source)
+        session.flush()
+        chapter = EbookChapter(source_id=source.id, spine_index=0, title="Chapter", href=None)
+        session.add(chapter)
+        session.flush()
+        session.add(
+            EbookChunk(
+                id=1,
+                source_id=source.id,
+                chapter_id=chapter.id,
+                chunk_index=0,
+                text="content",
+                token_start=0,
+                token_count=1,
+                page_label=None,
+                content_sha256="b" * 64,
+                search_text="Book Author Chapter content",
+            )
+        )
+        session.commit()
+
+        records = fetch_bm25_corpus_records(session)
+
+    assert records[0]["bm25_text"] == "Book Author Chapter content"
+
+
 def test_reciprocal_rank_fusion_marks_hybrid_source() -> None:
     vector_results = [SearchResult(chunk_id=1, text="a", source_title="A")]
     lexical_results = [SearchResult(chunk_id=2, text="b", source_title="B")]
@@ -119,7 +161,7 @@ def test_search_ebooks_runs_vector_and_bm25_in_parallel(monkeypatch) -> None:
     def fake_vector_candidates(received_engine, query, _config):
         """Return vector candidates after confirming BM25 has started."""
         received_engines.append(received_engine)
-        assert query == "parallel"
+        assert query == "what is parallel"
         vector_started.set()
         assert bm25_started.wait(timeout=2)
         return [SearchResult(chunk_id=1, text="vector", source_title="Vector", vector_score=0.9)]
@@ -135,13 +177,14 @@ def test_search_ebooks_runs_vector_and_bm25_in_parallel(monkeypatch) -> None:
     monkeypatch.setattr("python.ebook_search.search.bm25_candidates", fake_bm25_candidates)
     config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
 
-    response = search_ebooks(engine, "parallel", config)
+    response = search_ebooks(engine, "what is parallel", config)
 
     timings = {step.name: step for step in response.timings}
     assert [result.chunk_id for result in response.results] == [1, 2]
     assert timings["Embedding + vector search"].counts_toward_total is False
     assert timings["BM25 search"].counts_toward_total is False
     assert timings["Hybrid retrieval"].counts_toward_total is True
+    assert timings["BM25 query preparation"].counts_toward_total is True
     assert received_engines == [engine]
 
 

tests/test_ebook_search_rerank.py

@@ -75,7 +75,7 @@ def test_reranking_enabled_reorders_candidates(monkeypatch: pytest.MonkeyPatch)
     results = rerank_chunks("query", candidates(), RerankConfig())
 
     assert [result.chunk_id for result in results] == [2, 1, 3]
-    assert [round(result.score, 3) for result in results] == [0.45, 0.1, 0.0]
+    assert [round(result.score, 3) for result in results] == [0.78, 0.37, 0.28]
     assert [result.rerank_score for result in results] == [0.9, 0.1, 0.4]
 
 
@@ -100,8 +100,8 @@ def test_reranking_cannot_ignore_hybrid_score(monkeypatch: pytest.MonkeyPatch) -
     results = rerank_chunks("query", candidates, RerankConfig())
 
     assert [result.chunk_id for result in results] == [1, 2]
-    assert results[0].score == 0.7
+    assert results[0].score == pytest.approx(0.79)
-    assert results[1].score == 0.0
+    assert results[1].score == 0.7
     assert results[1].rerank_score == 1.0
 
 
@@ -129,7 +129,7 @@ def test_malformed_vllm_rerank_json_does_not_crash_search(monkeypatch: pytest.Mo
 
     results = rerank_chunks("query", candidates()[:1], RerankConfig())
 
-    assert results[0].score == 0.0
+    assert results[0].score == 0.3
 
 
 def test_vllm_rerank_scores_are_clamped(monkeypatch: pytest.MonkeyPatch) -> None:
@@ -147,4 +147,4 @@ def test_vllm_rerank_scores_are_clamped(monkeypatch: pytest.MonkeyPatch) -> None
 
     results = rerank_chunks("query", candidates()[:2], RerankConfig())
 
-    assert [result.rerank_score for result in results] == [0.0, 1.0]
+    assert {result.chunk_id: result.rerank_score for result in results} == {1: 0.0, 2: 1.0}