From e3013914f6c842bb80b92435f9f955f3834d2871 Mon Sep 17 00:00:00 2001
From: Jesser Hamdaoui <hamdaouijesser2004@gmail.com>
Date: Mon, 20 Apr 2026 18:25:37 +0100
Subject: [PATCH 1/2] fix(optimization): handle None metric scores in
 LocalEvalSampler

---
 src/google/adk/optimization/local_eval_sampler.py | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/src/google/adk/optimization/local_eval_sampler.py b/src/google/adk/optimization/local_eval_sampler.py
index b00c34280f..f27d5a15fe 100644
--- a/src/google/adk/optimization/local_eval_sampler.py
+++ b/src/google/adk/optimization/local_eval_sampler.py
@@ -289,7 +289,11 @@ def _extract_eval_data(
         for eval_metric_result in per_invocation_result.eval_metric_results:
           eval_metric_results.append({
               "metric_name": eval_metric_result.metric_name,
-              "score": round(eval_metric_result.score, 2),  # accurate enough
+              "score": (
+                  round(eval_metric_result.score, 2)
+                  if eval_metric_result.score is not None
+                  else None
+              ),  # accurate enough
               "eval_status": eval_metric_result.eval_status.name,
           })
         per_invocation_result_dict = {

From 6a8c12522d845c8286a2c94049f260e7bc9dcdd5 Mon Sep 17 00:00:00 2001
From: Jesser Hamdaoui <hamdaouijesser2004@gmail.com>
Date: Mon, 20 Apr 2026 18:34:39 +0100
Subject: [PATCH 2/2] test(optimization): add regression test for None metric
 score extraction

---
 .../optimization/local_eval_sampler_test.py   | 42 +++++++++++++++++++
 1 file changed, 42 insertions(+)

diff --git a/tests/unittests/optimization/local_eval_sampler_test.py b/tests/unittests/optimization/local_eval_sampler_test.py
index 6ebd99cb58..e685256e09 100644
--- a/tests/unittests/optimization/local_eval_sampler_test.py
+++ b/tests/unittests/optimization/local_eval_sampler_test.py
@@ -338,6 +338,48 @@ async def test_extract_eval_data(mocker):
   ]
 
 
+def test_extract_eval_data_preserves_none_metric_score(mocker):
+  mock_eval_sets_manager = mocker.MagicMock(spec=EvalSetsManager)
+  mock_eval_case = mocker.MagicMock()
+  mock_eval_case.conversation_scenario = "test_scenario"
+  mock_eval_sets_manager.get_eval_case.return_value = mock_eval_case
+
+  mock_metric_result = mocker.MagicMock(spec=EvalMetricResult)
+  mock_metric_result.metric_name = "test_metric"
+  mock_metric_result.score = None
+  mock_metric_result.eval_status = EvalStatus.NOT_EVALUATED
+
+  mock_per_inv_result = mocker.MagicMock(spec=EvalMetricResultPerInvocation)
+  mock_per_inv_result.actual_invocation = mocker.MagicMock(spec=Invocation)
+  mock_per_inv_result.expected_invocation = mocker.MagicMock(spec=Invocation)
+  mock_per_inv_result.eval_metric_results = [mock_metric_result]
+
+  mock_eval_result = mocker.MagicMock(spec=EvalCaseResult)
+  mock_eval_result.eval_id = "t1"
+  mock_eval_result.eval_metric_result_per_invocation = [mock_per_inv_result]
+
+  mocker.patch(
+      "google.adk.optimization.local_eval_sampler.extract_single_invocation_info",
+      side_effect=[{"info": "actual"}, {"info": "expected"}],
+  )
+
+  config = LocalEvalSamplerConfig(
+      eval_config=EvalConfig(),
+      app_name="test_app",
+      train_eval_set="train_set",
+      train_eval_case_ids=["t1"],
+  )
+  interface = LocalEvalSampler(config, mock_eval_sets_manager)
+
+  eval_data = interface._extract_eval_data("train_set", [mock_eval_result])
+
+  assert eval_data["t1"]["invocations"][0]["eval_metric_results"] == [{
+      "metric_name": "test_metric",
+      "score": None,
+      "eval_status": "NOT_EVALUATED",
+  }]
+
+
 @pytest.mark.asyncio
 async def test_sample_and_score(mocker):
   # Mock results