Pin the over-fetch rule, and assert the type a lookup raises

`_fetch_limit` had no test: a text query over-fetches `limit * 10` only with a
reranker, and an image query keeps its vector ranking either way. The sandbox
test asserted `KeyError`, which `UnknownDatabaseError` subclasses, so it could
not tell the contract from a bare one.

`uses_configured_databases` documents a mapping of one as covered; the test
named for it passed no mapping at all. Its `config` parameter is an `AppConfig`.

`test_an_analysis_capability_mounts_the_configured_set` carried a VCR marker and
no cassette, making no HTTP calls.

CHANGELOG.md

@@ -22,12 +22,11 @@
 
 ### Fixed
 
-- `haiku-rag settings` prints YAML instead of Python dict reprs.
+- `haiku-rag settings` prints YAML.
-- The chat document filter pages results, searches the database for the rest, and
+- The chat document filter pages results and lists the selected separately.
-  lists the selected separately, instead of mounting a checkbox per document.
+  Selection is by document ID, and a typed search applies on enter.
-  Selection is by document ID.
 - `haiku-rag list` prints only the fields a document has.
-- `haiku-rag` and `haiku-ingester` print the message and exit when the configured embedder does not match the database, instead of raising a traceback.
+- `haiku-rag` and `haiku-ingester` exit with a message on an embedder mismatch.
 - Capabilities created without a client honor `lancedb.uri`.
 - A `lancedb.uri` without a scheme is treated as a local path. `--db PATH`
   overrides it.

evaluations/evaluations/config.py

@@ -7,6 +7,7 @@ from datasets import Dataset
 from pydantic import BaseModel, model_validator
 from pydantic_evals import Case
 from pydantic_evals.evaluators import Evaluator
+from haiku.rag.config.models import AppConfig
 
 
 class Turn(BaseModel):
@@ -84,7 +85,7 @@ class DatasetSpec:
     experiment_metadata: dict[str, Any] | None = None
 
     def uses_configured_databases(
-        self, config, override_path: Path | None = None
+        self, config: AppConfig, override_path: Path | None = None
     ) -> bool:
         """Whether `lancedb.databases` places the databases to evaluate over.
 

evaluations/tests/test_config.py

@@ -179,7 +179,18 @@ class TestCoversASet:
 
         assert spec.uses_configured_databases(config, _Path("/chosen.lancedb")) is False
 
-    def test_one_database_is_not_a_set(self):
+    def test_a_configured_set_of_one_is_still_configured(self):
+        """A mapping of one is a named database like any other."""
+        from haiku.rag.config.models import AppConfig, LanceDBConfig
+
+        from evaluations.datasets import DATASETS
+
+        spec = next(iter(DATASETS.values()))
+        config = AppConfig(lancedb=LanceDBConfig(databases={"a": "/a.lancedb"}))
+
+        assert spec.uses_configured_databases(config) is True
+
+    def test_naming_no_database_is_not_a_set(self):
         from haiku.rag.config.models import AppConfig
 
         from evaluations.datasets import DATASETS

tests/multi_db/test_capabilities.py

@@ -81,7 +81,6 @@ class TestStandaloneCapabilities:
         assert "beta document" in formatted
 
     @pytest.mark.asyncio
-    @pytest.mark.vcr()
     async def test_an_analysis_capability_mounts_the_configured_set(self, tmp_path):
         from haiku.rag.capabilities.analysis import (
             create_capability as create_analysis,

tests/multi_db/test_search.py

@@ -268,6 +268,84 @@ class TestRerankerFusion:
         assert stub.attached == {"alpha": b"bytes-alpha", "beta": b"bytes-beta"}
 
 
+class TestOverFetchingForAReranker:
+    """A reranker needs more candidates than it returns. Ranking without one does
+    not, and an image query keeps its vector ranking either way."""
+
+    @staticmethod
+    def _limits_asked(monkeypatch) -> list[int]:
+        from haiku.rag.store.repositories.chunk import ChunkRepository
+
+        asked: list[int] = []
+        search = ChunkRepository.search
+
+        async def spy(self, *args, **kwargs):
+            asked.append(kwargs["limit"])
+            return await search(self, *args, **kwargs)
+
+        monkeypatch.setattr(ChunkRepository, "search", spy)
+        return asked
+
+    @pytest.mark.asyncio
+    async def test_a_text_query_over_fetches_for_a_reranker(
+        self, tmp_path, monkeypatch
+    ):
+        config = _config(tmp_path, ["alpha", "beta"])
+        await _seed(config, "alpha", ["alpha document about cats"])
+        await _seed(config, "beta", ["beta document about cats"])
+        monkeypatch.setattr(HaikuRAG, "reranker", property(lambda self: StubReranker()))
+        asked = self._limits_asked(monkeypatch)
+
+        async with HaikuRAG(config=config) as rag:
+            await rag.search("cats", limit=3, search_type="fts")
+            per_database = list(asked)
+            asked.clear()
+            await rag.search("cats", limit=3, search_type="fts", sources=["alpha"])
+
+        assert per_database == [30, 30]
+        assert asked == [30]
+
+    @pytest.mark.asyncio
+    async def test_a_text_query_without_a_reranker_fetches_what_it_returns(
+        self, tmp_path, monkeypatch
+    ):
+        config = _config(tmp_path, ["alpha", "beta"])
+        await _seed(config, "alpha", ["alpha document about cats"])
+        await _seed(config, "beta", ["beta document about cats"])
+        monkeypatch.setattr(HaikuRAG, "reranker", property(lambda self: None))
+        asked = self._limits_asked(monkeypatch)
+
+        async with HaikuRAG(config=config) as rag:
+            await rag.search("cats", limit=3, search_type="fts")
+            per_database = list(asked)
+            asked.clear()
+            await rag.search("cats", limit=3, search_type="fts", sources=["alpha"])
+
+        assert per_database == [3, 3]
+        assert asked == [3]
+
+    @pytest.mark.asyncio
+    async def test_an_image_query_fetches_what_it_returns(self, tmp_path, monkeypatch):
+        config = _config(tmp_path, ["alpha", "beta"])
+        await _seed(config, "alpha", ["alpha document about cats"])
+        await _seed(config, "beta", ["beta document about cats"])
+        monkeypatch.setattr(HaikuRAG, "reranker", property(lambda self: StubReranker()))
+
+        dim = get_config().embeddings.model.vector_dim
+
+        async def embed_image(self, image):  # noqa: ARG001
+            return [0.1] * dim
+
+        monkeypatch.setattr(EmbedderWrapper, "supports_images", True)
+        monkeypatch.setattr(EmbedderWrapper, "embed_image", embed_image)
+        asked = self._limits_asked(monkeypatch)
+
+        async with HaikuRAG(config=config) as rag:
+            await rag.search(b"\x89PNG\r\n\x1a\n", limit=3)
+
+        assert asked == [3, 3]
+
+
 class TestOneReranker:
     @pytest.mark.asyncio
     async def test_the_set_builds_one_reranker_for_a_text_query(

tests/sandbox/test_sandbox_multi_db.py

@@ -5,6 +5,7 @@ import pytest
 from haiku.rag.client import HaikuRAG
 from haiku.rag.client.scope import DatabaseRef, DatabaseScope
 from haiku.rag.sandbox import AnalysisContext, Sandbox
+from haiku.rag.store.exceptions import UnknownDatabaseError
 from tests.multi_db.helpers import _config, _seed
 
 
@@ -300,7 +301,7 @@ class TestSelectionOnOneDatabase:
         await _seed(config, "alpha", ["alpha document about cats"])
 
         async with HaikuRAG(config=config) as rag:
-            with pytest.raises(KeyError, match="beta"):
+            with pytest.raises(UnknownDatabaseError, match="beta"):
                 await _mounted(rag, sources=["beta"])
 
     @pytest.mark.asyncio