Remove corpus counts from chat chrome

2026-08-10 17:26:58 +07:00
parent 46469468bb
commit 97cb6d16f4
31 changed files with 2192 additions and 424 deletions
@@ -124,6 +124,24 @@ def test_no_drug_named_asks_which_one():
    assert reply.reason == "no_drug"


+def test_drug_attribute_without_an_attribute_does_not_fall_into_overview_retrieval():
+    retrieval = _FixedRetrieval({})
+    answers = GroundedAnswerService(routing=None)
+    agent = RagAgent(
+        _FixedUnderstander(QueryFrame(
+            turn_type="drug_attribute", drugs=("paracetamol_acetaminophen",)
+        )),
+        retrieval,
+        answers,
+    )
+
+    reply = agent.handle("paracetamol thì sao?")
+
+    assert reply.decision == "clarify"
+    assert reply.reason == "missing_attribute"
+    assert retrieval.calls == []
+
+
 def test_needs_clarify_frame_is_surfaced_directly():
    agent = _agent(QueryFrame(
        turn_type="dosing_calc", drugs=("paracetamol",),
@@ -137,6 +155,22 @@ def test_needs_clarify_frame_is_surfaced_directly():
    assert reply.quick_replies == ()


+def test_understanding_provider_failure_is_an_abstain_not_a_fake_clarification():
+    agent = _agent(QueryFrame(
+        turn_type="out_of_scope",
+        needs_clarify=True,
+        clarify_reason="Dịch vụ đang gặp sự cố tạm thời.",
+        system_error="understanding_provider_unavailable",
+    ))
+
+    reply = agent.handle("liều paracetamol")
+
+    assert reply.decision == "abstain"
+    assert reply.reason == "understanding_provider_unavailable"
+    assert reply.answer == "Dịch vụ đang gặp sự cố tạm thời."
+    assert reply.clarification is None
+
+
 def test_needs_clarify_frame_carries_quick_replies_through():
    """This is the path real traffic actually hits (checked live): the
    understanding LLM call itself sets needs_clarify/clarify_reason before
@@ -153,6 +187,167 @@ def test_needs_clarify_frame_carries_quick_replies_through():
    assert reply.quick_replies == ("Người lớn", "Trẻ em")


+def test_dosing_without_route_asks_only_when_evidence_is_ambiguous():
+    result = RetrievalResult(
+        EvidenceDecision.ANSWERABLE,
+        "grounded_evidence_available",
+        (
+            _evidence("Đường uống, người lớn: 500 mg mỗi lần."),
+            _evidence("Đặt trực tràng, người lớn: 500 mg mỗi lần."),
+        ),
+        resolved_drug_id="paracetamol_acetaminophen",
+    )
+
+    class _Generator:
+        def generate(self, system, user, schema):
+            return (
+                '{"claims": [], "evidence_sufficient": false, '
+                '"clarifying_question": "Anh/chị muốn dùng đường nào?", '
+                '"quick_replies": ["Uống", "Đặt trực tràng"]}'
+            )
+
+    retrieval = _FixedRetrieval({"paracetamol_acetaminophen": result})
+    answers = GroundedAnswerService(routing=None, generator=_Generator())
+    agent = RagAgent(
+        _FixedUnderstander(QueryFrame(
+            turn_type="dosing_calc",
+            drugs=("paracetamol_acetaminophen",),
+            population="nguoi_lon",
+        )),
+        retrieval,
+        answers,
+    )
+
+    reply = agent.handle("Người lớn", conversation_id="dose-route")
+
+    assert reply.decision == "clarify"
+    assert reply.reason == "needs_more_info"
+    assert reply.quick_replies == ("Uống", "Đặt trực tràng")
+    assert len(retrieval.calls) == 1
+    remembered = agent._last_frame["dose-route"]
+    assert remembered.needs_clarify is True
+    assert remembered.population == "nguoi_lon"
+    assert remembered.clarify_reason == reply.clarification
+
+
+def test_dosing_without_route_answers_directly_when_evidence_has_one_route():
+    result = RetrievalResult(
+        EvidenceDecision.ANSWERABLE,
+        "grounded_evidence_available",
+        (_evidence("Đường uống, người lớn: 500 mg mỗi lần."),),
+        resolved_drug_id="paracetamol_acetaminophen",
+    )
+
+    class _Generator:
+        def generate(self, system, user, schema):
+            if "entailed" in schema.get("properties", {}):
+                return '{"entailed": true, "unsupported": []}'
+            return (
+                '{"claims": [{"text": "Đường uống, người lớn: 500 mg mỗi lần.", '
+                '"citations": [1]}], "evidence_sufficient": true, '
+                '"clarifying_question": null, "quick_replies": []}'
+            )
+
+    retrieval = _FixedRetrieval({"paracetamol_acetaminophen": result})
+    answers = GroundedAnswerService(routing=None, generator=_Generator())
+    agent = RagAgent(
+        _FixedUnderstander(QueryFrame(
+            turn_type="dosing_calc",
+            drugs=("paracetamol_acetaminophen",),
+            population="nguoi_lon",
+        )),
+        retrieval,
+        answers,
+    )
+
+    reply = agent.handle("Liều Paracetamol cho người lớn")
+
+    assert reply.decision == "answerable"
+    assert reply.quick_replies == ()
+    assert "500 mg" in reply.answer
+    assert len(retrieval.calls) == 1
+
+
+def test_general_dosage_section_survey_does_not_force_population_chip():
+    result = RetrievalResult(
+        EvidenceDecision.ANSWERABLE,
+        "grounded_evidence_available",
+        (_evidence("Người lớn: uống 10 mg. Trẻ em: liều theo cân nặng."),),
+        resolved_drug_id="example",
+    )
+    retrieval = _FixedRetrieval({"example": result})
+    agent = RagAgent(
+        _FixedUnderstander(QueryFrame(
+            turn_type="dosing_calc",
+            drugs=("example",),
+            needs_clarify=True,
+            clarify_reason="Người lớn hay trẻ em?",
+        )),
+        retrieval,
+        GroundedAnswerService(routing=None),
+    )
+
+    reply = agent.handle(
+        "Dược thư hướng dẫn dùng Example thế nào: đường dùng và các liều nếu có?"
+    )
+
+    assert reply.decision == "answerable"
+    assert reply.quick_replies == ()
+    assert "tra cứu tổng quan toàn mục" in retrieval.calls[0][2]
+
+
+def test_clear_precaution_section_survey_overrides_model_overclarification():
+    result = RetrievalResult(
+        EvidenceDecision.ANSWERABLE,
+        "grounded_evidence_available",
+        (_evidence("Theo dõi chức năng thận và điện giải."),),
+        resolved_drug_id="example",
+    )
+    agent = _agent(
+        QueryFrame(
+            turn_type="drug_attribute",
+            drugs=("example",),
+            attribute="than_trong",
+            needs_clarify=True,
+            clarify_reason="Muốn hỏi thận trọng hay chống chỉ định?",
+        ),
+        {"example": result},
+    )
+
+    reply = agent.handle("Những tình huống nào cần thận trọng khi dùng Example?")
+
+    assert reply.decision == "answerable"
+    assert reply.quick_replies == ()
+
+
+def test_complete_adult_dosing_core_ignores_an_irrelevant_weight_reask():
+    result = RetrievalResult(
+        EvidenceDecision.ANSWERABLE, "grounded_evidence_available",
+        (_evidence("Đường uống, người lớn: 500 mg mỗi lần."),),
+        resolved_drug_id="paracetamol_acetaminophen",
+    )
+    retrieval = _FixedRetrieval({"paracetamol_acetaminophen": result})
+    answers = GroundedAnswerService(routing=None)
+    agent = RagAgent(
+        _FixedUnderstander(QueryFrame(
+            turn_type="dosing_calc",
+            drugs=("paracetamol_acetaminophen",),
+            population="nguoi_lon",
+            route="uong",
+            needs_clarify=True,
+            clarify_reason="Cân nặng của người lớn là bao nhiêu kg?",
+        )),
+        retrieval,
+        answers,
+    )
+
+    reply = agent.handle("Uống")
+
+    assert reply.decision == "answerable"
+    assert len(retrieval.calls) == 1
+    assert retrieval.calls[0][1] == "lieu_luong_va_cach_dung"
+
+
 def test_single_drug_attribute_retrieves_and_answers():
    result = RetrievalResult(
        EvidenceDecision.ANSWERABLE, "grounded_evidence_available",
@@ -490,7 +685,10 @@ def test_a_generous_budget_does_not_change_normal_behaviour():
                return '{"sufficient": true, "clarifying_question": null, "quick_replies": []}'
            if "entailed" in schema.get("properties", {}):
                return '{"entailed": true, "unsupported": []}'
-            return '{"answer": "Liều 500 mg [1].", "evidence_sufficient": true, "clarifying_question": null}'
+            return (
+                '{"claims": [{"text": "Liều 500 mg", "citations": [1]}], '
+                '"evidence_sufficient": true, "clarifying_question": null}'
+            )

    answers = GroundedAnswerService(routing=None, generator=_Generator())
    agent = RagAgent(
@@ -500,7 +698,8 @@ def test_a_generous_budget_does_not_change_normal_behaviour():
    reply = agent.handle("liều metformin")

    assert reply.decision == "answerable"
-    assert reply.answer == "Liều 500 mg [1]."
+    assert reply.answer == "Liều 500 mg"
+    assert reply.blocks[0].claims[0].source_ids


 # --- durable conversation history (ADR 0008's named gap): an optional