Checkpoint frontend UI/UX overhaul and ingestion embed benchmark work

apps/ai-service/tests/test_api.py

@@ -2,7 +2,9 @@ from fastapi.testclient import TestClient
 
 from config import Settings
 from main import create_app
-from rag.answer import GroundedAnswerService
+from rag.agent import AgentReply
+from rag.answer import Citation, GroundedAnswerService
+from rag.metrics import TRACE_WRITE_FAILED, InMemoryMetrics
 from rag.models import EvidenceDecision, QueryIntent, RetrievalResult, SubjectScope
 
 
@@ -51,3 +53,127 @@ def test_query_requires_structured_scope_and_intent():
     )
     response = TestClient(app).post("/v1/rag/query", json={"query": "Liều?"})
     assert response.status_code == 422
+
+
+# --- F-03: the app.state.conversational slot is the new RagAgent, wired live ---
+
+
+class FakeAgent:
+    """Stands in for `RagAgent` — proves `routers/rag.py` calls `.handle()`
+    and maps `AgentReply` correctly, not that the agent's own routing logic
+    is correct (that's `test_agent.py`)."""
+
+    def __init__(self, reply: AgentReply) -> None:
+        self._reply = reply
+        self.calls: list[tuple[str, str | None]] = []
+
+    def handle(self, turn: str, conversation_id: str | None = None) -> AgentReply:
+        self.calls.append((turn, conversation_id))
+        return self._reply
+
+    def complete(self, prefix: str, k: int = 8) -> list[str]:
+        return ["Metformin"] if prefix else []
+
+
+def _citation() -> Citation:
+    return Citation(
+        chunk_id="metformin::lieu::0", printed_page_start=714, printed_page_end=714,
+        physical_page=812,
+    )
+
+
+def test_query_routes_through_the_agent_when_one_is_configured():
+    agent = FakeAgent(AgentReply(
+        decision="answerable", reason="grounded_evidence_available",
+        answer="Liều 500 mg [1].", citations=(_citation(),),
+        drugs=("metformin",), turn_type="drug_attribute", generated=True,
+    ))
+    app = create_app(
+        settings=Settings(),
+        answer_service=GroundedAnswerService(FixedRouting()),
+        conversational=agent,
+        trace_writer=MemoryTraceWriter(),
+    )
+    response = TestClient(app).post("/v1/rag/query", json={
+        "query": "Liều metformin?", "subject_scope": "human",
+        "intent": "fact_lookup", "conversation_id": "c1",
+    })
+    body = response.json()
+    assert response.status_code == 200
+    assert body["decision"] == "answerable"
+    assert body["answer"] == "Liều 500 mg [1]."
+    assert body["resolved_drug_id"] == "metformin"
+    assert len(body["citations"]) == 1
+    assert agent.calls == [("Liều metformin?", "c1")]
+
+
+def test_query_agent_clarification_is_surfaced_as_the_answer():
+    agent = FakeAgent(AgentReply(
+        decision="clarify", reason="no_drug",
+        clarification="Anh/chị muốn tra thuốc nào?",
+    ))
+    app = create_app(
+        settings=Settings(),
+        answer_service=GroundedAnswerService(FixedRouting()),
+        conversational=agent,
+        trace_writer=MemoryTraceWriter(),
+    )
+    response = TestClient(app).post("/v1/rag/query", json={
+        "query": "liều dùng bao nhiêu", "subject_scope": "human",
+        "intent": "fact_lookup",
+    })
+    body = response.json()
+    assert body["decision"] == "clarify"
+    assert body["answer"] == "Anh/chị muốn tra thuốc nào?"
+    assert body["resolved_drug_id"] is None
+
+
+def test_suggest_delegates_to_the_agent_when_configured():
+    app = create_app(
+        settings=Settings(),
+        conversational=FakeAgent(AgentReply(decision="answerable", reason="x")),
+        trace_writer=MemoryTraceWriter(),
+    )
+    response = TestClient(app).get("/v1/rag/suggest", params={"q": "met"})
+    assert response.json() == {"suggestions": ["Metformin"]}
+
+
+def test_suggest_with_no_agent_configured_returns_empty():
+    app = create_app(settings=Settings(), trace_writer=MemoryTraceWriter())
+    response = TestClient(app).get("/v1/rag/suggest", params={"q": "met"})
+    assert response.json() == {"suggestions": []}
+
+
+# --- F-09: trace persistence is fail-open ------------------------------------
+
+
+class _RaisingTraceWriter:
+    """Stands in for a Postgres outage: `save()` always raises."""
+
+    def save(self, **fields):
+        raise ConnectionError("could not connect to postgres")
+
+
+def test_a_trace_write_failure_does_not_turn_a_good_answer_into_a_500():
+    agent = FakeAgent(AgentReply(
+        decision="answerable", reason="grounded_evidence_available",
+        answer="Liều 500 mg [1].", citations=(_citation(),),
+        drugs=("metformin",), turn_type="drug_attribute", generated=True,
+    ))
+    metrics = InMemoryMetrics()
+    app = create_app(
+        settings=Settings(),
+        answer_service=GroundedAnswerService(FixedRouting()),
+        conversational=agent,
+        trace_writer=_RaisingTraceWriter(),
+        metrics=metrics,
+    )
+    response = TestClient(app).post("/v1/rag/query", json={
+        "query": "Liều metformin?", "subject_scope": "human", "intent": "fact_lookup",
+    })
+    body = response.json()
+
+    assert response.status_code == 200
+    assert body["answer"] == "Liều 500 mg [1]."
+    assert body["trace_id"]  # a locally-generated fallback id, still present
+    assert metrics.total(TRACE_WRITE_FAILED) == 1