new feature: Multi-Intent

tests/migration/test_judge_context_compaction.py

@@ -51,3 +51,82 @@ async def test_pipeline_compacts_context_before_external_judge():
     assert len(rendered) < 40000
     assert 'TIM Fashion Mensal' in rendered
     assert 'session' not in rendered
+
+
+@pytest.mark.asyncio
+async def test_response_quality_uses_original_request_for_transaction_confirmation():
+    seen = {}
+
+    class ResponseQualityJudge:
+        name = 'tim_response_quality'
+
+        async def evaluate(self, question, answer, context):
+            seen['question'] = question
+            return JudgeResult(name=self.name, score=1.0, passed=True, reason='ok')
+
+    pipeline = JudgePipeline(judges=[ResponseQualityJudge()], enabled=True)
+    context = {
+        'transaction_status': 'COMPLETED',
+        'confirmation_received': True,
+        'transaction_evidence': [{
+            'arguments': {
+                'operator_instructions': 'cancelar o serviço adicional e enviar a segunda via',
+            },
+        }],
+    }
+
+    await pipeline.evaluate_all(
+        'autorizo que prossiga conforme solicitado',
+        'Cancelamento concluído com sucesso.',
+        context,
+    )
+
+    assert 'cancelar o serviço adicional e enviar a segunda via' in seen['question']
+    assert 'Confirmação do usuário neste turno: autorizo que prossiga conforme solicitado' in seen['question']
+
+
+@pytest.mark.asyncio
+async def test_non_quality_judge_keeps_literal_confirmation_question():
+    seen = {}
+
+    class GroundednessJudge:
+        name = 'tim_groundedness'
+
+        async def evaluate(self, question, answer, context):
+            seen['question'] = question
+            return JudgeResult(name=self.name, score=1.0, passed=True, reason='ok')
+
+    pipeline = JudgePipeline(judges=[GroundednessJudge()], enabled=True)
+    context = {
+        'transaction_status': 'COMPLETED',
+        'confirmation_received': True,
+        'transaction_evidence': [{'query': 'cancelar um serviço'}],
+    }
+
+    await pipeline.evaluate_all('sim', 'Cancelamento concluído.', context)
+
+    assert seen['question'] == 'sim'
+
+
+@pytest.mark.asyncio
+async def test_quality_judge_does_not_infer_confirmation_from_text():
+    seen = {}
+
+    class ResponseQualityJudge:
+        name = 'tim_response_quality'
+
+        async def evaluate(self, question, answer, context):
+            seen['question'] = question
+            return JudgeResult(name=self.name, score=1.0, passed=True, reason='ok')
+
+    pipeline = JudgePipeline(judges=[ResponseQualityJudge()], enabled=True)
+    await pipeline.evaluate_all(
+        'sim',
+        'Resposta comum.',
+        {
+            'confirmation_received': False,
+            'transaction_evidence': [{'query': 'solicitação anterior'}],
+        },
+    )
+
+    assert seen['question'] == 'sim'