Add production condition retrieval smoke test

apps/ai-service/tests/test_citation_and_intro.py

@@ -227,6 +227,41 @@ def test_list_mode_skips_the_sufficiency_clarify():
     assert g.generated is True
 
 
+def test_list_mode_rejects_a_generated_drug_outside_candidate_set():
+    evidence = Evidence(
+        evidence_id="a__chi_dinh__0",
+        matched_doc_id="a__chi_dinh__0",
+        kind="prose",
+        text="Thuốc A được chỉ định điều trị bệnh X.",
+        score=1.0,
+        source_refs=(SourceRef(physical_page=100, precision="exact", printed_page=101),),
+        hydrated_from_parent=False,
+        requires_visual_check=False,
+        drug_id="a",
+        drug_name="A",
+        section_key="chi_dinh",
+    )
+    result = _answerable(evidence)
+    gen = _Generator({
+        "claims": [{"text": "Thuốc D điều trị bệnh X.", "citations": [1], "drug_id": "d"}],
+        "evidence_sufficient": True,
+        "clarifying_question": None,
+        "quick_replies": [],
+    })
+    service = GroundedAnswerService(_Routing(result), gen)
+
+    grounded = service.answer_from_result(
+        "Bệnh X dùng thuốc gì?",
+        result,
+        list_mode=True,
+        candidate_drug_ids=("a",),
+    )
+
+    assert grounded.answer is None
+    assert grounded.result.decision == EvidenceDecision.ABSTAIN
+    assert grounded.result.reason == "unsupported_drug"
+
+
 def test_without_list_mode_the_same_evidence_does_ask_for_clarification():
     """Control for the test above: the same sufficiency payload, without
     `list_mode`, must actually clarify — proving the previous test's "not