feat(evidence): evaluate retrieval with a small fixture

harness/tests/test_qdrant_vector_store.py

@@ -475,6 +475,35 @@ def test_evidence_search_uses_filtered_dense_and_bm25_prefetches_with_default_rr
     ]
 
 
+@pytest.mark.parametrize("retrieval_mode, expected_query", [
+    ("dense", None),
+    ("bm25", {"text": "cardiomiopatia", "model": "qdrant/bm25", "options": {"language": "italian"}}),
+])
+def test_evidence_diagnostic_branch_searches_use_the_runtime_filter(retrieval_mode, expected_query):
+    fake = FakeQdrantHttp()
+    _ready_collection_with_bm25(fake)
+    store = _store(fake)
+    generation = "gen:" + "1" * 32
+
+    store.search(
+        ["evidence"], [0.2] * 1024, limit=10, kinds=["evidence"],
+        query_text="cardiomiopatia", query_language="italian", retrieval_mode=retrieval_mode,
+        metadata_filter={"workspace_id": "demo", "vector_generation": generation, "document_ids": ["doc:abc"]},
+    )
+
+    query = next(call[2] for call in reversed(fake.calls) if call[1].endswith("/points/query"))
+    assert query["limit"] == 10
+    assert query["filter"]["must"][-2:] == [
+        {"key": "vector_generation", "match": {"value": generation}},
+        {"key": "document_id", "match": {"any": ["doc:abc"]}},
+    ]
+    if retrieval_mode == "dense":
+        assert query["vector"] == [0.2] * 1024
+    else:
+        assert query["query"] == expected_query
+        assert query["using"] == "bm25"
+
+
 def test_search_filters_by_workspace_and_allowed_record_kinds():
     fake = FakeQdrantHttp()
     store = _store(fake)