fix: wrap diagnosis run_log with StepsJsonRunLog (restore algo #7 traces)

2026-07-16 06:05:55 -04:00
parent e1cac644c8
commit 7d02cded99
2 changed files with 67 additions and 1 deletions
@@ -147,6 +147,70 @@ def _fake_question(question_id: str, video_id: str) -> object:
    )


+@pytest.mark.asyncio
+async def test_diagnosis_reads_traces_from_steps_json(
+    runner_with_real_store: Runner,
+) -> None:
+    """traces 表为空但 predictions.steps_json 有轨迹时，诊断仍拿到非空 traces。
+
+    构造一条只写 steps_json、不写 traces 表的 predictions 行；patch run_diagnosis
+    捕获传入的 run_log，直接 await 其 get_traces 断言经 StepsJsonRunLog 从
+    steps_json 重建出非空轨迹（算法 #7 恢复）。
+    """
+    from app.harness.inference import PREDICTIONS_SCHEMA
+    from app.harness.log import HarnessLog
+
+    steps_json = json.dumps(
+        [
+            {
+                "thought": "先看整体",
+                "tool_call": {"tool": "search_tree", "args": {"query": "开场"}},
+                "tool_output": "命中 L2 节点 A",
+            }
+        ],
+        ensure_ascii=False,
+    )
+    with HarnessLog(str(runner_with_real_store._paths.db_path), "infer_adhoc") as log:
+        log.create_table("predictions", PREDICTIONS_SCHEMA)
+        log.create_table("traces", {"video_id": "TEXT", "question_id": "TEXT", "step": "INTEGER"})
+        log.insert(
+            "predictions",
+            {
+                "video_id": _REAL_VIDEO_ID,
+                "question_id": _REAL_QUESTION_ID,
+                "task_type": "Action Reasoning",
+                "prediction": "A",
+                "answer": "B",
+                "evidence": "",
+                "reasoning": "",
+                "steps_used": 1,
+                "prompt_tokens": 0,
+                "completion_tokens": 0,
+                "stop_reason": "finished",
+                "steps_json": steps_json,
+            },
+        )
+
+    captured: dict[str, object] = {}
+
+    async def _fake_run_diagnosis(**kwargs: object) -> DiagnosisResult:
+        captured["run_log"] = kwargs["run_log"]
+        return _empty_diagnosis_result()
+
+    with patch(
+        "core.evolution.diagnose.run_diagnosis",
+        new=AsyncMock(side_effect=_fake_run_diagnosis),
+    ):
+        await runner_with_real_store._run_diagnosis(
+            "infer_adhoc", question_ids=[_REAL_QUESTION_ID]
+        )
+
+    run_log = captured["run_log"]
+    traces = await run_log.get_traces("infer_adhoc", question_ids=[_REAL_QUESTION_ID])
+    assert traces, "traces 表空时应从 steps_json 重建出非空轨迹"
+    assert traces[0]["tool_name"] == "search_tree"
+
+
@pytest.mark.asyncio
 async def test_run_diagnosis_full_scan_loads_all_video_trees(
    runner_with_real_store: Runner,