diff --git a/tests/test_ebook_search_core.py b/tests/test_ebook_search_core.py
new file mode 100644
index 0000000..75ba762
--- /dev/null
+++ b/tests/test_ebook_search_core.py
@@ -0,0 +1,380 @@
+"""Tests for EPUB search core helpers."""
+
+from __future__ import annotations
+
+from dataclasses import replace
+from datetime import UTC, datetime
+from os import environ
+from pathlib import Path
+from threading import Event
+from types import ModuleType
+
+import pytest
+from sqlalchemy import create_engine, select
+from sqlalchemy.orm import sessionmaker
+
+from python.ebook_search.answer import answer_query
+from python.ebook_search.bm25_corpus import (
+    BM25Corpus,
+    BM25CorpusUnavailableError,
+    BM25Manifest,
+    ensure_bm25_corpus,
+    load_bm25_corpus,
+)
+from python.ebook_search.config import EbookSearchConfig, RerankConfig, load_config, normalize_embedding_model
+from python.ebook_search.embeddings import MODEL_DIMENSIONS, ensure_embedding_models
+from python.ebook_search.ingest import chunk_text, find_existing_source
+from python.ebook_search.search import (
+    SearchResponse,
+    SearchResult,
+    bm25_candidates,
+    reciprocal_rank_fusion,
+    retrieval_query_from_text,
+    search_ebooks,
+)
+from python.ebook_search.timing import RuntimeStep
+from python.orm.richie import EbookEmbeddingModel, EbookSource, RichieBase
+
+
+def test_chunk_text_uses_overlap() -> None:
+    chunks = chunk_text(" ".join(str(index) for index in range(100)), chunk_tokens=20, overlap_tokens=5)
+
+    assert len(chunks) > 1
+    assert chunks[0].token_start == 0
+    assert chunks[1].token_start == 15
+    assert all(chunk.token_count <= 20 for chunk in chunks)
+
+
+def test_reciprocal_rank_fusion_combines_vector_and_bm25_rankings() -> None:
+    vector_results = [
+        SearchResult(chunk_id=1, text="a", source_title="A", score=0.9, vector_score=0.9),
+        SearchResult(chunk_id=2, text="b", source_title="B", score=0.8, vector_score=0.8),
+    ]
+    lexical_results = [
+        SearchResult(chunk_id=2, text="b", source_title="B", score=4.2, bm25_score=4.2),
+        SearchResult(chunk_id=3, text="c", source_title="C", score=2.1, bm25_score=2.1),
+    ]
+
+    fused = reciprocal_rank_fusion(vector_results, lexical_results)
+
+    assert [result.chunk_id for result in fused] == [2, 1, 3]
+    assert fused[0].rank_source == "Hybrid"
+    assert fused[0].vector_score == 0.8
+    assert fused[0].bm25_score == 4.2
+    assert fused[0].fused_score == fused[0].score
+
+
+def test_find_existing_source_matches_path_or_hash() -> 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=None,
+            language=None,
+            publisher=None,
+            identifier=None,
+            file_path="/old/book.epub",
+            file_sha256="a" * 64,
+            file_mtime=datetime.now(tz=UTC),
+            file_size=10,
+        )
+        session.add(source)
+        session.commit()
+
+        assert find_existing_source(session, Path("/old/book.epub"), "b" * 64) == source
+        assert find_existing_source(session, Path("/new/book.epub"), "a" * 64) == source
+
+
+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")]
+
+    fused = reciprocal_rank_fusion(vector_results, lexical_results)
+
+    assert {result.rank_source for result in fused} == {"Hybrid"}
+
+
+def test_search_response_sums_runtime_steps() -> None:
+    response = SearchResponse(
+        query="query",
+        results=[],
+        rank_label="Hybrid",
+        timings=(
+            RuntimeStep(name="A", duration_ms=1.25),
+            RuntimeStep(name="B", duration_ms=2.75),
+            RuntimeStep(name="Parallel detail", duration_ms=10.0, counts_toward_total=False),
+        ),
+    )
+
+    assert response.total_runtime_ms == 4.0
+
+
+def test_search_ebooks_runs_vector_and_bm25_in_parallel(monkeypatch) -> None:
+    engine = create_engine("sqlite+pysqlite:///:memory:", future=True)
+    vector_started = Event()
+    bm25_started = Event()
+    received_engines: list[object] = []
+
+    def fake_vector_candidates(received_engine, query, _config):
+        """Return vector candidates after confirming BM25 has started."""
+        received_engines.append(received_engine)
+        assert query == "parallel"
+        vector_started.set()
+        assert bm25_started.wait(timeout=2)
+        return [SearchResult(chunk_id=1, text="vector", source_title="Vector", vector_score=0.9)]
+
+    def fake_bm25_candidates(query, _config):
+        """Return BM25 candidates after confirming vector search has started."""
+        assert query == "parallel"
+        bm25_started.set()
+        assert vector_started.wait(timeout=2)
+        return [SearchResult(chunk_id=2, text="bm25", source_title="BM25", bm25_score=2.0)]
+
+    monkeypatch.setattr("python.ebook_search.search.vector_candidates", fake_vector_candidates)
+    monkeypatch.setattr("python.ebook_search.search.bm25_candidates", fake_bm25_candidates)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+
+    response = search_ebooks(engine, "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 received_engines == [engine]
+
+
+def test_retrieval_query_keeps_entity_and_series_terms() -> None:
+    assert retrieval_query_from_text("what does Damien Montgomery stand for in starship mage") == (
+        "damien montgomery stand starship mage"
+    )
+
+
+def test_bm25_candidates_scores_whole_corpus(monkeypatch) -> None:
+    record = {
+        "chunk_id": 2,
+        "text": "high",
+        "source_title": "B",
+        "source_author": None,
+        "chapter_title": None,
+        "page_label": None,
+        "bm25_text": "high",
+    }
+    manifest = BM25Manifest(created_at=datetime.now(tz=UTC), db_updated_at=None, chunk_count=1)
+    corpus = BM25Corpus(retriever=object(), records=(record,), manifest=manifest)
+    captured: dict[str, object] = {}
+
+    def fake_score_bm25_corpus(query, saved_corpus, *, limit):
+        captured["query"] = query
+        captured["corpus"] = saved_corpus
+        captured["limit"] = limit
+        return [(record, 1.5)]
+
+    monkeypatch.setattr("python.ebook_search.search.load_bm25_corpus", lambda _config: corpus)
+    monkeypatch.setattr("python.ebook_search.search.score_bm25_corpus", fake_score_bm25_corpus)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+
+    results = bm25_candidates("high", config)
+
+    assert captured["query"] == "high"
+    assert captured["corpus"] == corpus
+    assert captured["limit"] == 120
+    assert [result.chunk_id for result in results] == [2]
+    assert [result.bm25_score for result in results] == [1.5]
+
+
+def test_bm25_candidates_raises_when_corpus_is_unavailable(monkeypatch) -> None:
+    def fake_load_bm25_corpus(_config):
+        raise BM25CorpusUnavailableError
+
+    monkeypatch.setattr("python.ebook_search.search.load_bm25_corpus", fake_load_bm25_corpus)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+
+    with pytest.raises(BM25CorpusUnavailableError):
+        bm25_candidates("high", config)
+
+
+def test_load_bm25_corpus_caches_disk_load(monkeypatch, tmp_path) -> None:
+    load_bm25_corpus.cache_clear()
+    manifest = BM25Manifest(created_at=datetime.now(tz=UTC), db_updated_at=None, chunk_count=1)
+    record = {
+        "chunk_id": 2,
+        "text": "cached",
+        "source_title": "B",
+        "source_author": None,
+        "chapter_title": None,
+        "page_label": None,
+        "bm25_text": "cached",
+    }
+    load_count = 0
+
+    class FakeRetriever:
+        """Fake persisted BM25 retriever."""
+
+        corpus = (record,)
+
+    class FakeBM25:
+        """Fake BM25 class with observable load count."""
+
+        @staticmethod
+        def load(index_path, *, load_corpus, mmap):
+            nonlocal load_count
+            load_count += 1
+            assert index_path == tmp_path
+            assert load_corpus is True
+            assert mmap is True
+            return FakeRetriever()
+
+    fake_bm25s = ModuleType("bm25s")
+    fake_bm25s.BM25 = FakeBM25
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.read_bm25_manifest", lambda _path: manifest)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.bm25_index_exists", lambda _path, _manifest: True)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.bm25s", fake_bm25s)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False), bm25_index_dir=str(tmp_path))
+
+    try:
+        first = load_bm25_corpus(config)
+        second = load_bm25_corpus(config)
+    finally:
+        load_bm25_corpus.cache_clear()
+
+    assert first is second
+    assert first is not None
+    assert first.records == (record,)
+    assert load_count == 1
+
+
+def test_load_bm25_corpus_raises_when_index_is_missing(monkeypatch, tmp_path) -> None:
+    load_bm25_corpus.cache_clear()
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.read_bm25_manifest", lambda _path: None)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.bm25_index_exists", lambda _path, _manifest: False)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False), bm25_index_dir=str(tmp_path))
+
+    try:
+        with pytest.raises(BM25CorpusUnavailableError, match="BM25 corpus is not available"):
+            load_bm25_corpus(config)
+    finally:
+        load_bm25_corpus.cache_clear()
+
+
+def test_ensure_bm25_corpus_refreshes_missing_index(monkeypatch) -> None:
+    refreshed: list[object] = []
+    db_updated_at = datetime.now(tz=UTC)
+
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.read_bm25_manifest", lambda _path: None)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.bm25_index_exists", lambda _path, _manifest: False)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.corpus_last_updated_at", lambda _session: db_updated_at)
+    monkeypatch.setattr(
+        "python.ebook_search.bm25_corpus.refresh_bm25_corpus",
+        lambda session, config, *, db_updated_at: refreshed.append((session, config, db_updated_at)),
+    )
+
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+    session = object()
+
+    ensure_bm25_corpus(session, config)
+
+    assert refreshed == [(session, config, db_updated_at)]
+
+
+def test_ensure_bm25_corpus_refreshes_stale_index(monkeypatch) -> None:
+    refreshed: list[object] = []
+    created_at = datetime(2026, 1, 1, tzinfo=UTC)
+    db_updated_at = datetime(2026, 1, 2, tzinfo=UTC)
+    manifest = BM25Manifest(created_at=created_at, db_updated_at=created_at, chunk_count=10)
+
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.read_bm25_manifest", lambda _path: manifest)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.bm25_index_exists", lambda _path, _manifest: True)
+    monkeypatch.setattr("python.ebook_search.bm25_corpus.corpus_last_updated_at", lambda _session: db_updated_at)
+    monkeypatch.setattr(
+        "python.ebook_search.bm25_corpus.refresh_bm25_corpus",
+        lambda session, config, *, db_updated_at: refreshed.append((session, config, db_updated_at)),
+    )
+
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+    session = object()
+
+    ensure_bm25_corpus(session, config)
+
+    assert refreshed == [(session, config, db_updated_at)]
+
+
+def test_supported_embedding_models_match_service_names() -> None:
+    assert MODEL_DIMENSIONS == {
+        "qwen3-embedding-0.6b": 1024,
+        "qwen3-embedding-4b": 2560,
+        "qwen3-embedding-8b": 4096,
+    }
+
+
+def test_ensure_embedding_models_registers_service_names() -> 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:
+        ensure_embedding_models(session)
+        session.commit()
+
+        models = list(session.scalars(select(EbookEmbeddingModel).order_by(EbookEmbeddingModel.name)))
+
+    assert [(model.name, model.dimension) for model in models] == [
+        ("qwen3-embedding-0.6b", 1024),
+        ("qwen3-embedding-4b", 2560),
+        ("qwen3-embedding-8b", 4096),
+    ]
+
+
+def test_embedding_model_aliases_normalize_to_provider_names() -> None:
+    assert normalize_embedding_model() == "qwen3-embedding-0.6b"
+
+    environ["EBOOK_SEARCH_EMBEDDING_MODEL"] = "qwen3-embedding-0.6b"
+    assert normalize_embedding_model() == "qwen3-embedding-0.6b"
+
+    environ["EBOOK_SEARCH_EMBEDDING_MODEL"] = "Qwen3-Embedding-0.6B"
+    assert normalize_embedding_model() == "qwen3-embedding-0.6b"
+
+    environ["EBOOK_SEARCH_EMBEDDING_MODEL"] = "Qwen/Qwen3-Embedding-4B"
+
+    assert normalize_embedding_model() == "qwen3-embedding-4b"
+
+    environ["EBOOK_SEARCH_EMBEDDING_MODEL"] = "qwen3-embedding:8b"
+    assert normalize_embedding_model() == "qwen3-embedding-8b"
+
+    environ["EBOOK_SEARCH_EMBEDDING_MODEL"] = "qwen3-embedding-8b"
+    assert normalize_embedding_model() == "qwen3-embedding-8b"
+
+
+def test_answer_generation_is_enabled_by_default(monkeypatch) -> None:
+    monkeypatch.delenv("EBOOK_SEARCH_ANSWER_ENABLED", raising=False)
+
+    config = load_config()
+
+    assert config.answer_enabled is True
+
+
+def test_chat_defaults_use_ollama_cloud(monkeypatch) -> None:
+    monkeypatch.delenv("EBOOK_SEARCH_VLLM_BASE_URL", raising=False)
+    monkeypatch.delenv("EBOOK_SEARCH_CHAT_MODEL", raising=False)
+
+    config = load_config()
+
+    assert config.vllm_base_url == "https://ollama.com/v1"
+    assert config.chat_model == "deepseek-v4-flash"
+
+
+def test_chat_api_key_falls_back_to_ollama_api_key(monkeypatch) -> None:
+    monkeypatch.delenv("EBOOK_SEARCH_VLLM_API_KEY", raising=False)
+    monkeypatch.setenv("OLLAMA_API_KEY", "ollama-key")
+
+    config = load_config()
+
+    assert config.vllm_api_key == "ollama-key"
+
+
+def test_answer_query_does_not_call_model_when_disabled() -> None:
+    config = replace(load_config(), answer_enabled=False)
+    result = SearchResult(chunk_id=1, text="source text", source_title="Book")
+
+    answer = answer_query("question", [result], config)
+
+    assert "Answer generation is disabled" in answer
diff --git a/tests/test_ebook_search_http.py b/tests/test_ebook_search_http.py
new file mode 100644
index 0000000..993f7bd
--- /dev/null
+++ b/tests/test_ebook_search_http.py
@@ -0,0 +1,84 @@
+"""Tests for EPUB search HTTP model adapters."""
+
+from __future__ import annotations
+
+import httpx
+import pytest
+
+from python.ebook_search.answer import answer_query
+from python.ebook_search.config import EbookSearchConfig, RerankConfig
+from python.ebook_search.embeddings import embed_texts
+from python.ebook_search.search import SearchResult
+
+
+def test_answer_query_uses_httpx_chat_completions(monkeypatch) -> None:
+    captured: dict[str, object] = {}
+
+    def fake_post(url: str, **kwargs: object) -> httpx.Response:
+        captured["url"] = url
+        captured["kwargs"] = kwargs
+        return httpx.Response(
+            200,
+            json={"choices": [{"message": {"content": "grounded answer"}}]},
+            request=httpx.Request("POST", url),
+        )
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+    config = EbookSearchConfig(
+        rerank=RerankConfig(enabled=False),
+        vllm_base_url="https://ollama.com/v1",
+        vllm_api_key="secret",
+        chat_model="deepseek-v4-flash",
+    )
+
+    answer = answer_query("question", [SearchResult(chunk_id=1, text="source", source_title="Book")], config)
+
+    assert answer == "grounded answer"
+    assert captured["url"] == "https://ollama.com/v1/chat/completions"
+    kwargs = captured["kwargs"]
+    assert isinstance(kwargs, dict)
+    assert kwargs["headers"] == {"Authorization": "Bearer secret"}
+    payload = kwargs["json"]
+    assert isinstance(payload, dict)
+    assert payload["model"] == "deepseek-v4-flash"
+
+
+def test_embed_texts_uses_httpx_embeddings(monkeypatch) -> None:
+    captured: dict[str, object] = {}
+    vector = [0.0] * 1024
+
+    def fake_post(url: str, **kwargs: object) -> httpx.Response:
+        captured["url"] = url
+        captured["kwargs"] = kwargs
+        return httpx.Response(
+            200,
+            json={"data": [{"embedding": vector}]},
+            request=httpx.Request("POST", url),
+        )
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+    config = EbookSearchConfig(
+        rerank=RerankConfig(enabled=False),
+        embedding_base_url="http://bob:8000/v1",
+        embedding_model="qwen3-embedding-0.6b",
+    )
+
+    embeddings = embed_texts(["hello"], config)
+
+    assert embeddings == [vector]
+    assert captured["url"] == "http://bob:8000/v1/embeddings"
+    kwargs = captured["kwargs"]
+    assert isinstance(kwargs, dict)
+    assert kwargs["headers"] == {}
+    assert kwargs["json"] == {"model": "qwen3-embedding-0.6b", "input": ["hello"]}
+
+
+def test_embed_texts_rejects_bad_response_shape(monkeypatch) -> None:
+    def fake_post(url: str, **_kwargs: object) -> httpx.Response:
+        return httpx.Response(200, json={"data": [{}]}, request=httpx.Request("POST", url))
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False))
+
+    with pytest.raises(RuntimeError, match="Embedding request failed"):
+        embed_texts(["hello"], config)
diff --git a/tests/test_ebook_search_rerank.py b/tests/test_ebook_search_rerank.py
new file mode 100644
index 0000000..7ccae46
--- /dev/null
+++ b/tests/test_ebook_search_rerank.py
@@ -0,0 +1,150 @@
+"""Tests for EPUB search reranking."""
+
+from __future__ import annotations
+
+import httpx
+import pytest
+
+from python.ebook_search.config import EbookSearchConfig, RerankConfig, load_rerank_config
+from python.ebook_search.rerank import rerank_chunks
+from python.ebook_search.search import SearchResult, apply_rerank, skip_rerank
+
+
+def candidates() -> list[SearchResult]:
+    return [
+        SearchResult(chunk_id=1, text="alpha", source_title="A", score=0.9),
+        SearchResult(chunk_id=2, text="beta", source_title="B", score=0.8),
+        SearchResult(chunk_id=3, text="gamma", source_title="C", score=0.7),
+    ]
+
+
+def rerank_response(payload: dict[str, object] | None = None, *, content: bytes | None = None) -> httpx.Response:
+    return httpx.Response(
+        200,
+        content=content,
+        json=payload,
+        request=httpx.Request("POST", "http://rerank.test/rerank"),
+    )
+
+
+def test_config_defaults_keep_reranking_optional(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.delenv("EBOOK_SEARCH_RERANK_ENABLED", raising=False)
+    monkeypatch.delenv("EBOOK_SEARCH_RERANK_BASE_URL", raising=False)
+    monkeypatch.delenv("EBOOK_SEARCH_RERANK_MODEL", raising=False)
+    monkeypatch.delenv("EBOOK_SEARCH_RERANK_CANDIDATES", raising=False)
+    monkeypatch.delenv("EBOOK_SEARCH_RERANK_TIMEOUT_SECONDS", raising=False)
+
+    config = load_rerank_config()
+
+    assert config.enabled is False
+    assert config.base_url == "http://192.168.90.25:8001"
+    assert config.model == "qwen3-reranker-06b"
+    assert config.candidates == 24
+    assert config.timeout_seconds == 30
+
+
+def test_reranking_disabled_returns_original_fused_order() -> None:
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=False), top_k=2)
+
+    response = skip_rerank("query", candidates(), config)
+
+    assert response.rank_label == "Hybrid"
+    assert [result.chunk_id for result in response.results] == [1, 2]
+
+
+def test_reranking_enabled_reorders_candidates(monkeypatch: pytest.MonkeyPatch) -> None:
+    def fake_post(_url: str, *, json: dict[str, object], timeout: float) -> httpx.Response:
+        assert timeout == 30
+        assert json == {
+            "model": "qwen3-reranker-06b",
+            "query": "query",
+            "documents": ["alpha", "beta", "gamma"],
+        }
+        return rerank_response(
+            {
+                "results": [
+                    {"index": 0, "relevance_score": 0.1},
+                    {"index": 1, "relevance_score": 0.9},
+                    {"index": 2, "relevance_score": 0.4},
+                ]
+            }
+        )
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+
+    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 [result.rerank_score for result in results] == [0.9, 0.1, 0.4]
+
+
+def test_reranking_cannot_ignore_hybrid_score(monkeypatch: pytest.MonkeyPatch) -> None:
+    candidates = [
+        SearchResult(chunk_id=1, text="strong hybrid", source_title="A", score=1.0),
+        SearchResult(chunk_id=2, text="weak hybrid", source_title="B", score=0.1),
+    ]
+
+    def fake_post(_url: str, **_kwargs: object) -> httpx.Response:
+        return rerank_response(
+            {
+                "results": [
+                    {"index": 0, "relevance_score": 0.7},
+                    {"index": 1, "relevance_score": 1.0},
+                ]
+            }
+        )
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+
+    results = rerank_chunks("query", candidates, RerankConfig())
+
+    assert [result.chunk_id for result in results] == [1, 2]
+    assert results[0].score == 0.7
+    assert results[1].score == 0.0
+    assert results[1].rerank_score == 1.0
+
+
+def test_vllm_rerank_timeout_raises(monkeypatch: pytest.MonkeyPatch) -> None:
+    def fake_rerank_chunks(
+        _query: str,
+        _candidates: list[SearchResult],
+        _config: RerankConfig,
+    ) -> list[SearchResult]:
+        message = "timeout"
+        raise httpx.TimeoutException(message)
+
+    monkeypatch.setattr("python.ebook_search.search.rerank_chunks", fake_rerank_chunks)
+    config = EbookSearchConfig(rerank=RerankConfig(enabled=True), top_k=2)
+
+    with pytest.raises(httpx.TimeoutException, match="timeout"):
+        apply_rerank("query", candidates(), config)
+
+
+def test_malformed_vllm_rerank_json_does_not_crash_search(monkeypatch: pytest.MonkeyPatch) -> None:
+    def fake_post(_url: str, **_kwargs: object) -> httpx.Response:
+        return rerank_response(content=b"not-json")
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+
+    results = rerank_chunks("query", candidates()[:1], RerankConfig())
+
+    assert results[0].score == 0.0
+
+
+def test_vllm_rerank_scores_are_clamped(monkeypatch: pytest.MonkeyPatch) -> None:
+    def fake_post(_url: str, **_kwargs: object) -> httpx.Response:
+        return rerank_response(
+            {
+                "results": [
+                    {"index": 0, "relevance_score": -1},
+                    {"index": 1, "relevance_score": 2},
+                ]
+            }
+        )
+
+    monkeypatch.setattr(httpx, "post", fake_post)
+
+    results = rerank_chunks("query", candidates()[:2], RerankConfig())
+
+    assert [result.rerank_score for result in results] == [0.0, 1.0]
diff --git a/tests/test_ebook_search_ui.py b/tests/test_ebook_search_ui.py
new file mode 100644
index 0000000..03b31b2
--- /dev/null
+++ b/tests/test_ebook_search_ui.py
@@ -0,0 +1,265 @@
+"""Tests for EPUB search HTMX routes."""
+
+from __future__ import annotations
+
+from fastapi.testclient import TestClient
+from sqlalchemy import create_engine
+
+from python.ebook_search.api.main import create_app
+from python.ebook_search.config import EbookSearchConfig, RerankConfig
+from python.ebook_search.embeddings import EmbeddingModelStats
+from python.ebook_search.search import SearchResponse, SearchResult
+from python.ebook_search.timing import RuntimeStep
+
+
+def patch_app_runtime(monkeypatch):
+    """Patch app startup dependencies used by UI route tests."""
+    monkeypatch.setattr("python.ebook_search.api.main.get_postgres_engine", fake_get_postgres_engine)
+    monkeypatch.setattr("python.ebook_search.api.main.ensure_bm25_corpus", lambda _session, _config: None)
+
+
+def fake_get_postgres_engine(**_kwargs):
+    """Return an in-memory engine for route tests."""
+    return create_engine("sqlite+pysqlite:///:memory:", future=True)
+
+
+def test_ui_form_passes_rerank_flag_to_search_handler(monkeypatch) -> None:
+    captured: dict[str, object] = {}
+
+    def fake_search_ebooks(_engine, query, config, *, rerank=False):
+        captured["query"] = query
+        captured["rerank"] = rerank
+        captured["config"] = config
+        return SearchResponse(query=query, results=[], rank_label="Hybrid + rerank")
+
+    monkeypatch.setattr("python.ebook_search.api.routes.search.search_ebooks", fake_search_ebooks)
+    monkeypatch.setattr(
+        "python.ebook_search.api.routes.search.answer_query",
+        lambda _query, _results, _config: "answer",
+    )
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+    app.state.config = EbookSearchConfig(rerank=RerankConfig(enabled=False), top_k=12, answer_enabled=True)
+
+    with TestClient(app) as client:
+        response = client.post("/search", data={"query": "where is the quote?", "rerank": "true"})
+
+    assert response.status_code == 200
+    assert "Hybrid + rerank" in response.text
+    assert captured["query"] == "where is the quote?"
+    assert captured["rerank"] is True
+
+
+def test_ui_search_failure_returns_visible_error(monkeypatch) -> None:
+    def fake_search_ebooks(_engine, _query, _config, *, rerank=False):
+        del rerank
+        msg = "search exploded"
+        raise RuntimeError(msg)
+
+    monkeypatch.setattr("python.ebook_search.api.routes.search.search_ebooks", fake_search_ebooks)
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+    app.state.config = EbookSearchConfig(rerank=RerankConfig(enabled=False), top_k=12)
+
+    with TestClient(app) as client:
+        response = client.post("/search", data={"query": "where is the quote?"})
+
+    assert response.status_code == 500
+    assert "search exploded" in response.text
+
+
+def test_ui_answer_failure_still_returns_sources(monkeypatch) -> None:
+    def fake_search_ebooks(_engine, query, _config, *, rerank=False):
+        del rerank
+        return SearchResponse(query=query, results=[], rank_label="Hybrid")
+
+    def fake_answer_query(_query, _results, _config):
+        msg = "answer exploded"
+        raise RuntimeError(msg)
+
+    monkeypatch.setattr("python.ebook_search.api.routes.search.search_ebooks", fake_search_ebooks)
+    monkeypatch.setattr("python.ebook_search.api.routes.search.answer_query", fake_answer_query)
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+    app.state.config = EbookSearchConfig(rerank=RerankConfig(enabled=False), top_k=12, answer_enabled=True)
+
+    with TestClient(app) as client:
+        response = client.post("/search", data={"query": "where is the quote?"})
+
+    assert response.status_code == 200
+    assert "Answer generation failed" in response.text
+
+
+def test_ui_skips_answer_when_disabled(monkeypatch) -> None:
+    called = False
+
+    def fake_search_ebooks(_engine, query, _config, *, rerank=False):
+        del rerank
+        return SearchResponse(query=query, results=[], rank_label="Hybrid")
+
+    def fake_answer_query(_query, _results, _config):
+        nonlocal called
+        called = True
+        return "answer"
+
+    monkeypatch.setattr("python.ebook_search.api.routes.search.search_ebooks", fake_search_ebooks)
+    monkeypatch.setattr("python.ebook_search.api.routes.search.answer_query", fake_answer_query)
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+    app.state.config = EbookSearchConfig(rerank=RerankConfig(enabled=False), answer_enabled=False)
+
+    with TestClient(app) as client:
+        response = client.post("/search", data={"query": "where is the quote?"})
+
+    assert response.status_code == 200
+    assert called is False
+    assert "Answer generation is disabled" in response.text
+
+
+def test_ui_shows_component_scores(monkeypatch) -> None:
+    def fake_search_ebooks(_engine, query, _config, *, rerank=False):
+        del rerank
+        return SearchResponse(
+            query=query,
+            rank_label="Hybrid + rerank",
+            results=[
+                SearchResult(
+                    chunk_id=1,
+                    text="source text",
+                    source_title="Book",
+                    score=0.9,
+                    rerank_score=0.9,
+                    vector_score=0.8,
+                    bm25_score=2.5,
+                    fused_score=0.03,
+                )
+            ],
+        )
+
+    monkeypatch.setattr("python.ebook_search.api.routes.search.search_ebooks", fake_search_ebooks)
+    monkeypatch.setattr(
+        "python.ebook_search.api.routes.search.answer_query",
+        lambda _query, _results, _config: "answer",
+    )
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+    app.state.config = EbookSearchConfig(rerank=RerankConfig(enabled=False), answer_enabled=True)
+
+    with TestClient(app) as client:
+        response = client.post("/search", data={"query": "where is the quote?"})
+
+    assert response.status_code == 200
+    assert "rerank" in response.text
+    assert "vector cosine" in response.text
+    assert "BM25" in response.text
+    assert "RRF" in response.text
+
+
+def test_ui_shows_search_runtime_chart(monkeypatch) -> None:
+    def fake_search_ebooks(_engine, query, _config, *, rerank=False):
+        del rerank
+        return SearchResponse(
+            query=query,
+            rank_label="Hybrid",
+            results=[],
+            timings=(
+                RuntimeStep(name="Embedding + vector search", duration_ms=12.5),
+                RuntimeStep(name="BM25 search", duration_ms=4.0),
+            ),
+        )
+
+    monkeypatch.setattr("python.ebook_search.api.routes.search.search_ebooks", fake_search_ebooks)
+    monkeypatch.setattr(
+        "python.ebook_search.api.routes.search.answer_query",
+        lambda _query, _results, _config: "answer",
+    )
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+    app.state.config = EbookSearchConfig(rerank=RerankConfig(enabled=False), answer_enabled=True)
+
+    with TestClient(app) as client:
+        response = client.post("/search", data={"query": "where is the quote?"})
+
+    assert response.status_code == 200
+    assert "Runtime" in response.text
+    assert "Total" in response.text
+    assert "Embedding + vector search" in response.text
+    assert "BM25 search" in response.text
+    assert "Answer generation" in response.text
+    assert "ms left" in response.text
+
+
+def test_ui_embed_all_batches_until_complete(monkeypatch) -> None:
+    counts = iter([32, 32, 5, 0])
+    batch_sizes: list[int] = []
+
+    def fake_embed_missing_chunks(_session, config):
+        batch_sizes.append(config.embedding_batch_size)
+        return next(counts)
+
+    monkeypatch.setattr("python.ebook_search.api.routes.admin.embed_missing_chunks", fake_embed_missing_chunks)
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+
+    with TestClient(app) as client:
+        response = client.post("/admin/embed-all")
+
+    assert response.status_code == 200
+    assert "Embedded 69 chunks in 3 batches of 32" in response.text
+    assert batch_sizes == [32, 32, 32, 32]
+
+
+def test_ui_scan_schedules_bm25_refresh_after_database_change(monkeypatch) -> None:
+    scheduled = False
+
+    def fake_ingest_configured_paths(_session, _config):
+        return 1
+
+    def fake_schedule_bm25_refresh(_app):
+        nonlocal scheduled
+        scheduled = True
+
+    monkeypatch.setattr("python.ebook_search.api.routes.admin.ingest_configured_paths", fake_ingest_configured_paths)
+    monkeypatch.setattr("python.ebook_search.api.routes.admin.schedule_bm25_refresh", fake_schedule_bm25_refresh)
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+
+    with TestClient(app) as client:
+        response = client.post("/admin/scan")
+
+    assert response.status_code == 200
+    assert "Indexed 1 EPUBs" in response.text
+    assert scheduled is True
+
+
+def test_admin_page_shows_embedding_counts_by_model(monkeypatch) -> None:
+    def fake_embedding_model_stats(_session):
+        return [
+            EmbeddingModelStats(
+                model_name="qwen3-embedding-0.6b",
+                dimension=1024,
+                embedded_chunks=40,
+                total_chunks=64,
+            ),
+            EmbeddingModelStats(
+                model_name="qwen3-embedding-4b",
+                dimension=2560,
+                embedded_chunks=8,
+                total_chunks=64,
+            ),
+        ]
+
+    monkeypatch.setattr("python.ebook_search.api.routes.admin.embedding_model_stats", fake_embedding_model_stats)
+    patch_app_runtime(monkeypatch)
+    app = create_app()
+
+    with TestClient(app) as client:
+        response = client.get("/admin")
+
+    assert response.status_code == 200
+    assert "qwen3-embedding-0.6b" in response.text
+    assert "1024" in response.text
+    assert "40" in response.text
+    assert "24" in response.text
+    assert "qwen3-embedding-4b" in response.text
+    assert "2560" in response.text