fix: normalize list score dicts in remote evals

py/src/braintrust/framework.py

@@ -1324,6 +1324,19 @@ def _build_classification_span_output(
     return grouped_output
 
 
+def _normalize_score(value: Any, error_message: str) -> ScoreLike:
+    original_value = value
+    if isinstance(value, dict):
+        try:
+            value = Score.from_dict(value)
+        except Exception as e:
+            raise ValueError(f"{error_message} Got: {original_value}") from e
+
+    if not is_score(value):
+        raise ValueError(f"{error_message} Got: {original_value}")
+    return value
+
+
 def _validate_classification_result(value: Any, classifier_name: str) -> Classification:
     if isinstance(value, Mapping):
         if not value:
@@ -1455,18 +1468,16 @@ async def await_or_run_scorer(root_span, scorer, name, **kwargs):
 
             result = await call_user_fn(event_loop, score, **kwargs)
             if isinstance(result, dict):
-                try:
-                    result = Score.from_dict(result)
-                except Exception as e:
-                    raise ValueError(f"When returning a dict, it must be a valid Score object. Got: {result}") from e
+                result = _normalize_score(result, "When returning a dict, it must be a valid Score object.")
 
             if isinstance(result, Iterable) and not isinstance(result, (str, bytes, Mapping)):
-                for s in result:
-                    if not is_score(s):
-                        raise ValueError(
-                            f"When returning an array of scores, each score must be a valid Score object. Got: {s}"
-                        )
-                result = list(result)
+                result = [
+                    _normalize_score(
+                        score,
+                        "When returning an array of scores, each score must be a valid Score object.",
+                    )
+                    for score in result
+                ]
             elif is_score(result):
                 result = [result]
             else:

py/src/braintrust/test_framework.py

@@ -296,6 +296,46 @@ def _run_eval_sync(self, *args, **kwargs):
         assert result.summary.scores[scorer_name].score == 1.0
 
 
+@pytest.mark.asyncio
+async def test_run_evaluator_normalizes_list_of_dict_scores():
+    data = [
+        EvalCase(input=1, expected=1),
+        EvalCase(input=2, expected=2),
+    ]
+
+    def simple_task(input_value):
+        return input_value
+
+    def list_dict_scorer(input_value, output, expected):
+        return [
+            {"name": "per_result", "score": 1.0},
+            {"name": "summary_only", "score": 1.0},
+        ]
+
+    evaluator = Evaluator(
+        project_name="test-project",
+        eval_name="test-list-dict-scorer",
+        data=data,
+        task=simple_task,
+        scores=[list_dict_scorer],
+        experiment_name=None,
+        metadata=None,
+    )
+
+    result = await run_evaluator(None, evaluator, None, [])
+
+    assert isinstance(result, EvalResultWithSummary)
+    assert len(result.results) == 2
+
+    for eval_result in result.results:
+        assert eval_result.scores["per_result"] == 1.0
+        assert eval_result.scores["summary_only"] == 1.0
+
+    assert result.summary.project_name == "test-project"
+    assert result.summary.scores["per_result"].score == 1.0
+    assert result.summary.scores["summary_only"].score == 1.0
+
+
 @pytest.mark.asyncio
 @pytest.mark.skipif(
     sys.version_info < (3, 14),