Some legacy evaluators do not support target output key

packages/uipath/src/uipath/eval/evaluators/base_legacy_evaluator.py

@@ -8,6 +8,7 @@
 
 from pydantic import ConfigDict, Field
 
+from .._helpers.output_path import resolve_output_path
 from ..models import EvaluationResult
 from ..models.models import (
     AgentExecution,
@@ -85,6 +86,19 @@ class BaseLegacyEvaluator(
 
     # Note: __init_subclass__ is inherited from BaseEvaluator and handles metrics tracking
 
+    def get_targeted_field(self, obj: Any) -> Any:
+        """Resolve the target output key path from the given object.
+
+        If target_output_key is set and not "*", resolves the dot-notation path.
+        Returns the original object if resolution fails or no key is configured.
+        """
+        if self.target_output_key and self.target_output_key != "*":
+            try:
+                return resolve_output_path(obj, self.target_output_key)
+            except (KeyError, IndexError, TypeError):
+                return obj
+        return obj
+
     def model_post_init(self, __context: Any):
         """Post-initialization hook for Pydantic models."""
         # Ensure config is set up for legacy evaluators

packages/uipath/src/uipath/eval/evaluators/legacy_exact_match_evaluator.py

@@ -2,7 +2,6 @@
 
 from uipath.eval.models import BooleanEvaluationResult, EvaluationResult
 
-from .._helpers.output_path import resolve_output_path
 from ..models.models import AgentExecution
 from .base_legacy_evaluator import LegacyEvaluationCriteria, LegacyEvaluatorConfig
 from .legacy_deterministic_evaluator_base import BaseLegacyDeterministicEvaluator
@@ -41,30 +40,8 @@ async def evaluate(
         Returns:
             EvaluationResult: Boolean result indicating exact match (True/False)
         """
-        actual_output = agent_execution.agent_output
-        expected_output = evaluation_criteria.expected_output
-
-        if self.target_output_key and self.target_output_key != "*":
-            if isinstance(actual_output, dict) and isinstance(expected_output, dict):
-                actual_resolved = True
-                expected_resolved = True
-
-                try:
-                    actual_output = resolve_output_path(
-                        actual_output, self.target_output_key
-                    )
-                except (KeyError, IndexError, TypeError):
-                    actual_resolved = False
-
-                try:
-                    expected_output = resolve_output_path(
-                        expected_output, self.target_output_key
-                    )
-                except (KeyError, IndexError, TypeError):
-                    expected_resolved = False
-
-                if not actual_resolved or not expected_resolved:
-                    actual_output = expected_output = {}
+        actual_output = self.get_targeted_field(agent_execution.agent_output)
+        expected_output = self.get_targeted_field(evaluation_criteria.expected_output)
 
         return BooleanEvaluationResult(
             score=self._canonical_json(actual_output)

packages/uipath/src/uipath/eval/evaluators/legacy_json_similarity_evaluator.py

@@ -3,7 +3,6 @@
 import math
 from typing import Any, Tuple, TypeVar
 
-from .._helpers.output_path import resolve_output_path
 from ..models import EvaluationResult, NumericEvaluationResult
 from ..models.models import AgentExecution
 from .base_legacy_evaluator import LegacyEvaluationCriteria, LegacyEvaluatorConfig
@@ -47,23 +46,8 @@ async def evaluate(
         Returns:
             EvaluationResult: Numerical score between 0-100 indicating similarity
         """
-        actual_output = agent_execution.agent_output
-        expected_output = evaluation_criteria.expected_output
-
-        if self.target_output_key and self.target_output_key != "*":
-            try:
-                actual_output = resolve_output_path(
-                    actual_output, self.target_output_key
-                )
-            except (KeyError, IndexError, TypeError):
-                actual_output = {}
-
-            try:
-                expected_output = resolve_output_path(
-                    expected_output, self.target_output_key
-                )
-            except (KeyError, IndexError, TypeError):
-                expected_output = {}
+        actual_output = self.get_targeted_field(agent_execution.agent_output)
+        expected_output = self.get_targeted_field(evaluation_criteria.expected_output)
 
         return NumericEvaluationResult(
             score=self._compare_json(expected_output, actual_output)

packages/uipath/src/uipath/eval/evaluators/legacy_llm_as_judge_evaluator.py

@@ -12,7 +12,6 @@
 from ..._utils.constants import COMMUNITY_agents_SUFFIX
 from .._execution_context import eval_set_run_id_context
 from .._helpers.helpers import is_empty_value
-from .._helpers.output_path import resolve_output_path
 from ..models import NumericEvaluationResult
 from ..models.models import (
     AgentExecution,
@@ -125,23 +124,8 @@ async def evaluate(
         if self.llm is None:
             self._initialize_llm()
 
-        actual_output = agent_execution.agent_output
-        expected_output = evaluation_criteria.expected_output
-
-        if self.target_output_key and self.target_output_key != "*":
-            try:
-                actual_output = resolve_output_path(
-                    actual_output, self.target_output_key
-                )
-            except (KeyError, IndexError, TypeError):
-                pass
-
-            try:
-                expected_output = resolve_output_path(
-                    expected_output, self.target_output_key
-                )
-            except (KeyError, IndexError, TypeError):
-                pass
+        actual_output = self.get_targeted_field(agent_execution.agent_output)
+        expected_output = self.get_targeted_field(evaluation_criteria.expected_output)
 
         # Create the evaluation prompt
         evaluation_prompt = self._create_evaluation_prompt(