diff --git a/bigframes/core/compile/compiler.py b/bigframes/core/compile/compiler.py
index ed95dacf74..517c265cba 100644
--- a/bigframes/core/compile/compiler.py
+++ b/bigframes/core/compile/compiler.py
@@ -240,9 +240,7 @@ def compile_read_table_unordered(
         return compiled.UnorderedIR(
             ibis_table,
             tuple(
-                bigframes.core.compile.ibis_types.ibis_value_to_canonical_type(
-                    ibis_table[scan_item.source_id].name(scan_item.id.sql)
-                )
+                ibis_table[scan_item.source_id].name(scan_item.id.sql)
                 for scan_item in scan.items
             ),
         )
diff --git a/bigframes/core/compile/ibis_types.py b/bigframes/core/compile/ibis_types.py
index af2b7908ad..2dcc1b3c8a 100644
--- a/bigframes/core/compile/ibis_types.py
+++ b/bigframes/core/compile/ibis_types.py
@@ -113,7 +113,9 @@ def cast_ibis_value(
 
     Raises:
         TypeError: if the type cast cannot be executed"""
-    if value.type() == to_type:
+    # normalize to nullable, which doesn't impact compatibility
+    value_type = value.type().copy(nullable=True)
+    if value_type == to_type:
         return value
     # casts that just work
     # TODO(bmil): add to this as more casts are verified
@@ -189,52 +191,39 @@ def cast_ibis_value(
         ibis_dtypes.multipolygon: (IBIS_GEO_TYPE,),
     }
 
-    value = ibis_value_to_canonical_type(value)
-    if value.type() in good_casts:
-        if to_type in good_casts[value.type()]:
+    if value_type in good_casts:
+        if to_type in good_casts[value_type]:
             return value.try_cast(to_type) if safe else value.cast(to_type)
     else:
         # this should never happen
         raise TypeError(
-            f"Unexpected value type {value.type()}. {constants.FEEDBACK_LINK}"
+            f"Unexpected value type {value_type}. {constants.FEEDBACK_LINK}"
         )
 
     # casts that need some encouragement
 
     # BigQuery casts bools to lower case strings. Capitalize the result to match Pandas
     # TODO(bmil): remove this workaround after fixing Ibis
-    if value.type() == ibis_dtypes.bool and to_type == ibis_dtypes.string:
+    if value_type == ibis_dtypes.bool and to_type == ibis_dtypes.string:
         if safe:
             return cast(ibis_types.StringValue, value.try_cast(to_type)).capitalize()
         else:
             return cast(ibis_types.StringValue, value.cast(to_type)).capitalize()
 
-    if value.type() == ibis_dtypes.bool and to_type == ibis_dtypes.float64:
+    if value_type == ibis_dtypes.bool and to_type == ibis_dtypes.float64:
         if safe:
             return value.try_cast(ibis_dtypes.int64).try_cast(ibis_dtypes.float64)
         else:
             return value.cast(ibis_dtypes.int64).cast(ibis_dtypes.float64)
 
-    if value.type() == ibis_dtypes.float64 and to_type == ibis_dtypes.bool:
+    if value_type == ibis_dtypes.float64 and to_type == ibis_dtypes.bool:
         return value != ibis_types.literal(0)
 
     raise TypeError(
-        f"Unsupported cast {value.type()} to {to_type}. {constants.FEEDBACK_LINK}"
+        f"Unsupported cast {value_type} to {to_type}. {constants.FEEDBACK_LINK}"
     )
 
 
-def ibis_value_to_canonical_type(value: ibis_types.Value) -> ibis_types.Value:
-    """Converts an Ibis expression to canonical type.
-
-    This is useful in cases where multiple types correspond to the same BigFrames dtype.
-    """
-    ibis_type = value.type()
-    name = value.get_name()
-    # Allow REQUIRED fields to be joined with NULLABLE fields.
-    nullable_type = ibis_type.copy(nullable=True)
-    return value.cast(nullable_type).name(name)
-
-
 def bigframes_dtype_to_ibis_dtype(
     bigframes_dtype: bigframes.dtypes.Dtype,
 ) -> ibis_dtypes.DataType:
diff --git a/third_party/bigframes_vendored/ibis/expr/operations/relations.py b/third_party/bigframes_vendored/ibis/expr/operations/relations.py
index fccd4cbf58..ef45fdfc0d 100644
--- a/third_party/bigframes_vendored/ibis/expr/operations/relations.py
+++ b/third_party/bigframes_vendored/ibis/expr/operations/relations.py
@@ -10,11 +10,7 @@
 from typing import Annotated, Any, Literal, Optional, TypeVar
 
 from bigframes_vendored.ibis.common.annotations import attribute
-from bigframes_vendored.ibis.common.collections import (
-    ConflictingValuesError,
-    FrozenDict,
-    FrozenOrderedDict,
-)
+from bigframes_vendored.ibis.common.collections import FrozenDict, FrozenOrderedDict
 from bigframes_vendored.ibis.common.exceptions import (
     IbisTypeError,
     IntegrityError,
@@ -342,20 +338,6 @@ class Set(Relation):
     values = FrozenOrderedDict()
 
     def __init__(self, left, right, **kwargs):
-        err_msg = "Table schemas must be equal for set operations."
-        try:
-            missing_from_left = right.schema - left.schema
-            missing_from_right = left.schema - right.schema
-        except ConflictingValuesError as e:
-            raise RelationError(err_msg + "\n" + str(e)) from e
-        if missing_from_left or missing_from_right:
-            msgs = [err_msg]
-            if missing_from_left:
-                msgs.append(f"Columns missing from the left:\n{missing_from_left}.")
-            if missing_from_right:
-                msgs.append(f"Columns missing from the right:\n{missing_from_right}.")
-            raise RelationError("\n".join(msgs))
-
         if left.schema.names != right.schema.names:
             # rewrite so that both sides have the columns in the same order making it
             # easier for the backends to implement set operations
@@ -365,7 +347,15 @@ def __init__(self, left, right, **kwargs):
 
     @attribute
     def schema(self):
-        return self.left.schema
+        dtypes = (
+            dt.higher_precedence(ltype, rtype)
+            for ltype, rtype in zip(
+                self.left.schema.values(), self.right.schema.values()
+            )
+        )
+        return Schema.from_tuples(
+            (name, coltype) for name, coltype in zip(self.left.schema.names, dtypes)
+        )
 
 
 @public