diff --git a/eval_protocol/models.py b/eval_protocol/models.py
index 40bb1e44..2ac254a4 100644
--- a/eval_protocol/models.py
+++ b/eval_protocol/models.py
@@ -12,7 +12,7 @@
 from openai.types.chat.chat_completion_message_tool_call import (
     ChatCompletionMessageToolCall,
 )
-from pydantic import BaseModel, ConfigDict, Field
+from pydantic import BaseModel, ConfigDict, Field, field_validator
 
 from eval_protocol.get_pep440_version import get_pep440_version
 from eval_protocol.human_id import generate_id
@@ -595,7 +595,7 @@ class EvaluationRow(BaseModel):
     supporting both row-wise batch evaluation and trajectory-based RL evaluation.
     """
 
-    model_config = ConfigDict(extra="allow")
+    model_config = ConfigDict(extra="allow", validate_assignment=True)
 
     # Core OpenAI ChatCompletion compatible conversation data
     messages: List[Message] = Field(description="List of messages in the conversation. Also known as a trajectory.")
@@ -626,6 +626,17 @@ class EvaluationRow(BaseModel):
         default=None, description="The evaluation result for this row/trajectory."
     )
 
+    @field_validator("evaluation_result", mode="before")
+    @classmethod
+    def _coerce_evaluation_result(
+        cls, value: EvaluateResult | dict[str, Any] | None
+    ) -> EvaluateResult | None:
+        if value is None or isinstance(value, EvaluateResult):
+            return value
+        if isinstance(value, dict):
+            return EvaluateResult(**value)
+        return value
+
     execution_metadata: ExecutionMetadata = Field(
         default_factory=lambda: ExecutionMetadata(run_id=None),
         description="Metadata about the execution of the evaluation.",
diff --git a/tests/test_models.py b/tests/test_models.py
index 723685b8..bdcd5d9b 100644
--- a/tests/test_models.py
+++ b/tests/test_models.py
@@ -198,8 +198,14 @@ def test_metric_result_dict_access():
     }
     assert set(metric.items()) == expected_items
 
-    # __iter__
-    assert set(list(metric)) == {"score", "reason", "is_score_valid"}
+
+def test_evaluation_row_accepts_dict_assignment_for_evaluation_result():
+    row = dummy_row()
+    row.evaluation_result = {"score": 0.6}
+
+    assert isinstance(row.evaluation_result, EvaluateResult)
+    assert row.evaluation_result.score == 0.6
+    assert row.evaluation_result.is_score_valid is True
 
 
 def test_evaluate_result_dict_access():