From d0014b848c9d8455138b51c1d814718f47c2234f Mon Sep 17 00:00:00 2001
From: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
Date: Sat, 30 May 2026 21:22:08 +0800
Subject: [PATCH] fix(eval): preserve custom eval metadata

---
 src/google/adk/evaluation/common.py          |  2 +-
 tests/unittests/evaluation/test_eval_case.py | 24 ++++++++++++++++++++
 2 files changed, 25 insertions(+), 1 deletion(-)

diff --git a/src/google/adk/evaluation/common.py b/src/google/adk/evaluation/common.py
index 34fbae195d..3b45d5450b 100644
--- a/src/google/adk/evaluation/common.py
+++ b/src/google/adk/evaluation/common.py
@@ -22,6 +22,6 @@ class EvalBaseModel(pydantic.BaseModel):
   model_config = pydantic.ConfigDict(
       alias_generator=alias_generators.to_camel,
       populate_by_name=True,
-      extra="forbid",
+      extra="allow",
       arbitrary_types_allowed=True,
   )
diff --git a/tests/unittests/evaluation/test_eval_case.py b/tests/unittests/evaluation/test_eval_case.py
index acf21f6a35..51a1a2521d 100644
--- a/tests/unittests/evaluation/test_eval_case.py
+++ b/tests/unittests/evaluation/test_eval_case.py
@@ -22,6 +22,7 @@
 from google.adk.evaluation.eval_case import IntermediateData
 from google.adk.evaluation.eval_case import InvocationEvent
 from google.adk.evaluation.eval_case import InvocationEvents
+from google.adk.evaluation.eval_case import SessionInput
 from google.genai import types as genai_types
 import pytest
 
@@ -281,3 +282,26 @@ def test_conversation_and_conversation_scenario_mutual_exclusion():
   # these two should not cause exceptions
   EvalCase(eval_id='test_id', conversation=[])
   EvalCase(eval_id='test_id', conversation_scenario=test_conversation_scenario)
+
+
+def test_eval_models_preserve_extra_metadata():
+  session_input = SessionInput(
+      app_name='app',
+      user_id='user',
+      category='search',
+      tier='low',
+  )
+  eval_case = EvalCase(
+      eval_id='case',
+      conversation=[],
+      session_input=session_input,
+      benchmark='smoke',
+  )
+
+  assert session_input.model_extra == {'category': 'search', 'tier': 'low'}
+  assert eval_case.model_extra == {'benchmark': 'smoke'}
+
+  dumped = eval_case.model_dump(by_alias=True)
+  assert dumped['benchmark'] == 'smoke'
+  assert dumped['sessionInput']['category'] == 'search'
+  assert dumped['sessionInput']['tier'] == 'low'