diff --git a/python/ebook_search/config.py b/python/ebook_search/config.py
index dd1cd7b..6a1e736 100644
--- a/python/ebook_search/config.py
+++ b/python/ebook_search/config.py
@@ -42,7 +42,7 @@ class RerankConfig(BaseSettings):
 
     model_config = SettingsConfigDict(env_prefix="EBOOK_SEARCH_RERANK_", frozen=True, protected_namespaces=())
 
-    enabled: bool = False
+    enabled: bool = True
     base_url: str = "http://192.168.90.25:8001"
     model: str = "qwen3-reranker-06b"
     candidates: int = 24
diff --git a/python/ebook_search/search.py b/python/ebook_search/search.py
index db7b1eb..6e3e8d1 100644
--- a/python/ebook_search/search.py
+++ b/python/ebook_search/search.py
@@ -4,6 +4,7 @@ from __future__ import annotations
 
 import logging
 import re
+from collections import defaultdict
 from concurrent.futures import ThreadPoolExecutor
 from dataclasses import dataclass, replace
 from typing import TYPE_CHECKING
@@ -93,14 +94,11 @@ def search_ebooks(
 
     logger.info("ebook_search_start query_length=%s rerank=%s", len(query), rerank)
     timings: list[RuntimeStep] = []
-    bm25_query, timing = timed_result("BM25 query preparation", retrieval_query_from_text, query)
-    timings.append(timing)
     retrieval, timing = timed_result(
         "Hybrid retrieval",
         parallel_retrieval,
         engine,
         query,
-        bm25_query,
         config,
     )
     timings.extend(retrieval.timings)
@@ -134,8 +132,7 @@ def search_ebooks(
 
 def parallel_retrieval(
     engine: Engine,
-    vector_query: str,
-    bm25_query: str,
+    query: str,
     config: EbookSearchConfig,
 ) -> RetrievalResponse:
     """Run vector and BM25 candidate retrieval concurrently with separate database sessions."""
@@ -145,14 +142,14 @@ def parallel_retrieval(
             "Embedding + vector search",
             vector_candidates,
             engine,
-            vector_query,
+            query,
             config,
         )
         bm25_future = executor.submit(
             timed_result,
             "BM25 search",
             bm25_candidates,
-            bm25_query,
+            query,
             config,
         )
         vector_results, vector_timing = vector_future.result()
@@ -263,7 +260,8 @@ def bm25_candidates(query: str, config: EbookSearchConfig) -> list[SearchResult]
         logger.info("ebook_bm25_search_complete corpus=0 candidates=0")
         return []
 
-    scored_records = score_bm25_corpus(query, corpus, limit=config.bm25_candidate_limit)
+    bm25_query = retrieval_query_from_text(query)
+    scored_records = score_bm25_corpus(bm25_query, corpus, limit=config.bm25_candidate_limit)
     results = [
         replace(search_result_from_row(record), score=score, vector_score=None, bm25_score=score)
         for record, score in scored_records
@@ -282,24 +280,23 @@ def bm25_candidates(query: str, config: EbookSearchConfig) -> list[SearchResult]
 def reciprocal_rank_fusion(
     vector_results: list[SearchResult],
     lexical_results: list[SearchResult],
-    *,
-    rank_constant: int = 60,
+    rank_constant: int,
 ) -> list[SearchResult]:
     """Fuse vector and lexical rankings with Reciprocal Rank Fusion."""
     by_chunk: dict[int, SearchResult] = {}
-    scores: dict[int, float] = {}
+    scores: defaultdict[int, float] = defaultdict(float)
     vector_scores: dict[int, float] = {}
     bm25_scores: dict[int, float] = {}
 
     for rank, result in enumerate(vector_results, start=1):
         by_chunk.setdefault(result.chunk_id, result)
         vector_scores[result.chunk_id] = result.vector_score if result.vector_score is not None else result.score
-        scores[result.chunk_id] = scores.get(result.chunk_id, 0.0) + (1 / (rank_constant + rank))
+        scores[result.chunk_id] += 1 / (rank_constant + rank)
 
     for rank, result in enumerate(lexical_results, start=1):
         by_chunk.setdefault(result.chunk_id, result)
         bm25_scores[result.chunk_id] = result.bm25_score if result.bm25_score is not None else result.score
-        scores[result.chunk_id] = scores.get(result.chunk_id, 0.0) + (1 / (rank_constant + rank))
+        scores[result.chunk_id] += 1 / (rank_constant + rank)
 
     return sorted(
         (