diff --git a/docetl/operations/utils.py b/docetl/operations/utils.py
index c3e2545a..f4ce3c48 100644
--- a/docetl/operations/utils.py
+++ b/docetl/operations/utils.py
@@ -359,7 +359,11 @@ def safe_eval(expression: str, output: Dict) -> bool:
         # Safely evaluate the expression
         return bool(aeval(expression))
     except Exception:
-        return False
+        # try to evaluate with python eval
+        try:
+            return bool(eval(expression, locals={"output": output}))
+        except Exception:
+            return False
 
 
 class APIWrapper(object):
diff --git a/docs/concepts/operators.md b/docs/concepts/operators.md
index f81653c7..fc9070ef 100644
--- a/docs/concepts/operators.md
+++ b/docs/concepts/operators.md
@@ -135,7 +135,7 @@ validate:
   - all(len(insight["supporting_actions"]) >= 1 for insight in output["insights"])
 ```
 
-Access variables using dictionary syntax: `input["field"]` or `output["field"]`.
+Access variables using dictionary syntax: `output["field"]`. Note that you can't access `input` docs in validation, but the output docs should have all the fields from the input docs (for non-reduce operations), since fields pass through unchanged.
 
 The `num_retries_on_validate_failure` attribute specifies how many times to retry the LLM if any validation statements fail.
 
diff --git a/tests/test_validation.py b/tests/test_validation.py
new file mode 100644
index 00000000..ed64af4c
--- /dev/null
+++ b/tests/test_validation.py
@@ -0,0 +1,46 @@
+import pytest
+from docetl.operations.map import MapOperation
+from tests.conftest import api_wrapper, default_model, max_threads
+
+
+@pytest.fixture
+def map_config_with_validation():
+    return {
+        "name": "sentiment_analysis_with_validation",
+        "type": "map",
+        "prompt": "Analyze the sentiment of the following text: '{{ input.text }}'. Classify it as either positive, negative, or neutral.",
+        "output": {"schema": {"sentiment": "string", "confidence": "float"}},
+        "model": "gpt-4o-mini",
+        "validate": [
+            "output['sentiment'] in ['positive', 'negative', 'neutral']",
+            "0 <= output['confidence'] <= 1",
+        ],
+        "num_retries_on_validate_failure": 2,
+    }
+
+
+@pytest.fixture
+def sample_data():
+    return [
+        {"text": "I love this product! It's amazing."},
+        {"text": "This is the worst experience ever."},
+        {"text": "The weather is okay today."},
+    ]
+
+
+def test_map_operation_with_validation(
+    map_config_with_validation, sample_data, api_wrapper, default_model, max_threads
+):
+    operation = MapOperation(
+        api_wrapper, map_config_with_validation, default_model, max_threads
+    )
+    results, cost = operation.execute(sample_data)
+
+    assert len(results) == len(sample_data)
+    assert cost > 0
+
+    for result in results:
+        assert "sentiment" in result
+        assert "confidence" in result
+        assert result["sentiment"] in ["positive", "negative", "neutral"]
+        assert 0 <= result["confidence"] <= 1