feat: persist single evaluation history

2026-07-23 19:46:15 +08:00
parent 5be9434ba3
commit 6116d10e8f
5 changed files with 398 additions and 13 deletions
@@ -333,6 +333,15 @@ class EvaluationRuntimeTests(unittest.TestCase):
                self.assertEqual(result['model'], 'test-model')
                self.assertEqual(runtime.list_datasets('alice'), [])
                self.assertEqual(runtime.list_experiments('alice'), [])
+                history = runtime.list_single_analyses('alice')
+                self.assertEqual(len(history), 1)
+                self.assertEqual(history[0]['id'], result['id'])
+                self.assertEqual(history[0]['query'], '导航去公司')
+                self.assertEqual(history[0]['prediction'], '地图导航')
+                self.assertNotIn('raw_output', history[0])
+                detail = runtime.get_single_analysis(result['id'], 'alice')
+                self.assertEqual(detail['prediction'], '地图导航')
+                self.assertEqual(detail['canonical']['query'], '导航去公司')
            finally:
                runtime.shutdown()