diff --git a/maxim/dataset/dataset.py b/maxim/dataset/dataset.py
index dc07038..2878cbb 100644
--- a/maxim/dataset/dataset.py
+++ b/maxim/dataset/dataset.py
@@ -2,10 +2,10 @@
 from typing import Any, Dict, Optional
 
 from ..models.dataset import (
-    ContextToEvaluateColumn,
+    CONTEXT_TO_EVALUATE_COLUMN,
     DataStructure,
-    ExpectedOutputColumn,
-    InputColumn,
+    EXPECTED_OUTPUT_COLUMN,
+    INPUT_COLUMN,
 )
 
 
@@ -20,7 +20,7 @@ def sanitize_data_structure(data_structure: Optional[DataStructure]) -> None:
     encountered_context_to_evaluate = False
     if data_structure:
         for value in data_structure.values():
-            if value == InputColumn:
+            if value == INPUT_COLUMN:
                 if encountered_input:
                     raise Exception(
                         "Data structure contains more than one input",
@@ -28,7 +28,7 @@ def sanitize_data_structure(data_structure: Optional[DataStructure]) -> None:
                     )
                 else:
                     encountered_input = True
-            elif value == ExpectedOutputColumn:
+            elif value == EXPECTED_OUTPUT_COLUMN:
                 if encountered_expected_output:
                     raise Exception(
                         "Data structure contains more than one expectedOutput",
@@ -36,7 +36,7 @@ def sanitize_data_structure(data_structure: Optional[DataStructure]) -> None:
                     )
                 else:
                     encountered_expected_output = True
-            elif value == ContextToEvaluateColumn:
+            elif value == CONTEXT_TO_EVALUATE_COLUMN:
                 if encountered_context_to_evaluate:
                     raise Exception(
                         "Data structure contains more than one contextToEvaluate",
diff --git a/maxim/models/dataset.py b/maxim/models/dataset.py
index f3bc112..40a0668 100644
--- a/maxim/models/dataset.py
+++ b/maxim/models/dataset.py
@@ -63,6 +63,16 @@ def to_json(self):
         return return_dict
 
 
+# Column name constants used throughout the dataset utilities.
+INPUT_COLUMN: str = "INPUT"
+EXPECTED_OUTPUT_COLUMN: str = "EXPECTED_OUTPUT"
+CONTEXT_TO_EVALUATE_COLUMN: str = "CONTEXT_TO_EVALUATE"
+VARIABLE_COLUMN: str = "VARIABLE"
+NULLABLE_VARIABLE_COLUMN: str = "NULLABLE_VARIABLE"
+OUTPUT_COLUMN: str = "OUTPUT"
+
+# Type aliases for column names.  These are used for type checking but also
+# need to be concrete values at runtime for equality comparisons.
 InputColumn = Literal["INPUT"]
 ExpectedOutputColumn = Literal["EXPECTED_OUTPUT"]
 ContextToEvaluateColumn = Literal["CONTEXT_TO_EVALUATE"]