fix(eval): handle unevaluated final response v2 results

src/google/adk/evaluation/final_response_match_v2.py

@@ -228,6 +228,14 @@ def aggregate_invocation_results(
         continue
       num_evaluated += 1
       num_valid += result.score
+
+    if num_evaluated == 0:
+      return EvaluationResult(
+          overall_score=None,
+          overall_eval_status=EvalStatus.NOT_EVALUATED,
+          per_invocation_results=per_invocation_results,
+      )
+
     overall_score = num_valid / num_evaluated
     return EvaluationResult(
         overall_score=overall_score,

tests/unittests/evaluation/test_final_response_match_v2.py

@@ -486,3 +486,34 @@ def test_aggregate_invocation_results():
   # Only 4 / 8 invocations are evaluated, and 2 / 4 are valid.
   assert aggregated_result.overall_score == 0.5
   assert aggregated_result.overall_eval_status == EvalStatus.PASSED
+
+
+def test_aggregate_invocation_results_none_evaluated():
+  evaluator = _create_test_evaluator_gemini(threshold=0.5)
+
+  actual_invocation, expected_invocation = _create_test_invocations(
+      "candidate text", "reference text"
+  )
+
+  per_invocation_results = [
+      PerInvocationResult(
+          actual_invocation=actual_invocation,
+          expected_invocation=expected_invocation,
+          score=None,
+          eval_status=EvalStatus.NOT_EVALUATED,
+      ),
+      PerInvocationResult(
+          actual_invocation=actual_invocation,
+          expected_invocation=expected_invocation,
+          score=1.0,
+          eval_status=EvalStatus.NOT_EVALUATED,
+      ),
+  ]
+
+  aggregated_result = evaluator.aggregate_invocation_results(
+      per_invocation_results
+  )
+
+  assert aggregated_result.overall_score is None
+  assert aggregated_result.overall_eval_status == EvalStatus.NOT_EVALUATED
+  assert aggregated_result.per_invocation_results == per_invocation_results