Add properties in Evaluation Result - Custom Evaluator extra fields.

sdk/evaluation/azure-ai-evaluation/CHANGELOG.md

@@ -1,6 +1,12 @@
 # Release History
 
-## 1.16.3 (Unreleased)
+## 1.16.4 (2026-04-03)
+
+### Features Added
+
+- Added support for evaluator `properties` passthrough in AOAI evaluation results. When an evaluator returns a `properties` dict, it is included alongside `score`, `label`, `reason`, `threshold`, and `passed` in the result object.
+
+## 1.16.3 (2026-04-01)
 
 ### Features Added
 

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_evaluate/_evaluate.py

@@ -2576,7 +2576,8 @@ def _extract_metric_values(
             "score": 4.5,
             "coherence_reason": "Good flow",
             "threshold": 3.0,
-            "sample": {...}
+            "sample": {...},
+            "properties": {"explanation": "Detailed analysis...", "confidence": 0.95}
         }
         expected_metrics = ["score"]
 
@@ -2586,13 +2587,29 @@ def _extract_metric_values(
                 "score": 4.5,
                 "reason": "Good flow",
                 "threshold": 3.0,
-                "sample": {...}
+                "sample": {...},
+                "properties": {"explanation": "Detailed analysis...", "confidence": 0.95}
             }
         }
+
+    Note: If a ``properties`` key is present in the metrics dict and its value is a dict,
+    it is extracted and attached to every per-metric result entry. This allows evaluators
+    to return additional output fields alongside standard score/reason/threshold values.
     """
     result_per_metric = {}
+    properties = None
 
     for metric_key, metric_value in metrics.items():
+        if metric_key == "properties":
+            if isinstance(metric_value, dict):
+                properties = metric_value
+            else:
+                logger.info(
+                    "Evaluator '%s' returned 'properties' as %s instead of dict; ignoring.",
+                    criteria_name,
+                    type(metric_value).__name__,
+                )
+            continue
         metric = _get_metric_from_criteria(criteria_name, metric_key, expected_metrics)
         temp_result_per_metric = {}
         if metric not in result_per_metric:
@@ -2612,6 +2629,11 @@ def _extract_metric_values(
         if result_name == "label" and criteria_type == "azure_ai_evaluator" and derived_passed is not None:
             _append_indirect_attachments_to_results(result_per_metric, "passed", metric, derived_passed, None, None)
 
+    if properties is not None:
+        for metric_dict in result_per_metric.values():
+            if metric_dict is not None and len(metric_dict) > 0:
+                metric_dict["properties"] = properties.copy()
+
     empty_metrics = []
     empty_metrics.extend(
         metric for metric, metric_dict in result_per_metric.items() if metric_dict is None or len(metric_dict) == 0
@@ -2855,7 +2877,8 @@ def _create_result_object(
             "score": 4.5,
             "reason": "Good logical flow",
             "threshold": 3.0,
-            "sample": {"input": "...", "output": "..."}
+            "sample": {"input": "...", "output": "..."},
+            "properties": {"explanation": "...", "confidence": 0.95}
         }
         criteria_type = "quality"
 
@@ -2869,8 +2892,13 @@ def _create_result_object(
             "reason": "Good logical flow",
             "threshold": 3.0,
             "passed": None,
-            "sample": {"input": "...", "output": "..."}
+            "sample": {"input": "...", "output": "..."},
+            "properties": {"explanation": "...", "confidence": 0.95}
         }
+
+    Note: The ``properties`` field is included only when the evaluator returned a
+    properties dict. It carries additional output fields beyond the standard
+    score/label/reason/threshold/passed values.
     """
     # Extract values
     score = metric_values.get("score")
@@ -2879,6 +2907,7 @@ def _create_result_object(
     threshold = metric_values.get("threshold")
     passed = metric_values.get("passed")
     sample = metric_values.get("sample")
+    properties = metric_values.get("properties")
 
     # Handle decrease boolean metrics
     if is_inverse:
@@ -2898,6 +2927,8 @@ def _create_result_object(
 
     if sample is not None:
         result_obj["sample"] = sample
+    if properties is not None:
+        result_obj["properties"] = properties
 
     return result_obj
 

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_evaluators/_common/_base_rai_svc_eval.py

@@ -321,7 +321,7 @@ def _parse_eval_result(self, eval_result) -> Dict[str, T]:
 
                         # Extract details from scoreProperties
                         if score_properties:
-                            parsed_result[f"{self._eval_metric. value}_details"] = _prepare_details(score_properties)
+                            parsed_result[f"{self._eval_metric.value}_details"] = _prepare_details(score_properties)
 
                         # Extract token counts from metrics
                         metrics = properties.get("metrics", {})
@@ -339,7 +339,7 @@ def _parse_eval_result(self, eval_result) -> Dict[str, T]:
                             total_tokens = ""
 
                         # Add token metadata (matching old format)
-                        parsed_result[f"{self._eval_metric. value}_total_tokens"] = total_tokens
+                        parsed_result[f"{self._eval_metric.value}_total_tokens"] = total_tokens
                         parsed_result[f"{self._eval_metric.value}_prompt_tokens"] = prompt_tokens
                         parsed_result[f"{self._eval_metric.value}_completion_tokens"] = completion_tokens
 

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_version.py

@@ -3,4 +3,4 @@
 # ---------------------------------------------------------
 # represents upcoming version
 
-VERSION = "1.16.3"
+VERSION = "1.16.4"

sdk/evaluation/azure-ai-evaluation/tests/unittests/test_evaluate.py

@@ -1353,6 +1353,97 @@ def run_test():
         assert len(empty_converted["_evaluation_results_list"]) == 0
         assert empty_converted["_evaluation_summary"]["result_counts"]["total"] == 0
 
+        # Test properties passthrough for custom evaluators
+        property_results = {
+            "metrics": {},
+            "rows": [
+                {
+                    "inputs.query": "test query",
+                    "outputs.friendly_eval.score": 4.5,
+                    "outputs.friendly_eval.score_threshold": 3,
+                    "outputs.friendly_eval.score_result": "Pass",
+                    "outputs.friendly_eval.score_reason": "The response was warm",
+                    "outputs.friendly_eval.properties": {
+                        "explanation": "Detailed reasoning about friendliness",
+                        "tone": "warm",
+                        "confidence": "high",
+                    },
+                }
+            ],
+            "studio_url": None,
+        }
+
+        _convert_results_to_aoai_evaluation_results(
+            results=property_results,
+            logger=logger,
+            eval_run_id=eval_run_id,
+            eval_id=eval_id,
+            evaluators={"friendly_eval": lambda **kwargs: {"score": 1}},
+            eval_meta_data={
+                "testing_criteria": [
+                    {
+                        "name": "friendly_eval",
+                        "type": "quality",
+                        "metrics": ["score"],
+                    }
+                ]
+            },
+        )
+
+        property_result = property_results["_evaluation_results_list"][0]["results"][0]
+        assert property_result["score"] == 4.5
+        assert property_result["label"] == "Pass"
+        assert property_result["reason"] == "The response was warm"
+        assert property_result["threshold"] == 3
+        assert property_result["properties"] == {
+            "explanation": "Detailed reasoning about friendliness",
+            "tone": "warm",
+            "confidence": "high",
+        }
+        assert "explanation" not in property_result
+
+        # Test that non-dict properties logs a warning and is omitted
+        non_dict_property_results = {
+            "metrics": {},
+            "rows": [
+                {
+                    "inputs.query": "test query",
+                    "outputs.friendly_eval.score": 3.0,
+                    "outputs.friendly_eval.score_threshold": 3,
+                    "outputs.friendly_eval.score_result": "Pass",
+                    "outputs.friendly_eval.score_reason": "Acceptable",
+                    "outputs.friendly_eval.properties": "not_a_dict",
+                }
+            ],
+            "studio_url": None,
+        }
+
+        with patch.object(logger, "info") as mock_info:
+            _convert_results_to_aoai_evaluation_results(
+                results=non_dict_property_results,
+                logger=logger,
+                eval_run_id=eval_run_id,
+                eval_id=eval_id,
+                evaluators={"friendly_eval": lambda **kwargs: {"score": 1}},
+                eval_meta_data={
+                    "testing_criteria": [
+                        {
+                            "name": "friendly_eval",
+                            "type": "quality",
+                            "metrics": ["score"],
+                        }
+                    ]
+                },
+            )
+
+        non_dict_result = non_dict_property_results["_evaluation_results_list"][0]["results"][0]
+        assert "properties" not in non_dict_result
+        mock_info.assert_any_call(
+            "Evaluator '%s' returned 'properties' as %s instead of dict; ignoring.",
+            "friendly_eval",
+            "str",
+        )
+
     @patch(
         "azure.ai.evaluation._evaluate._evaluate._map_names_to_builtins",
         return_value={},