diff --git a/ehrql/dummy_data_nextgen/query_info.py b/ehrql/dummy_data_nextgen/query_info.py
index ebc48315f..8447730fc 100644
--- a/ehrql/dummy_data_nextgen/query_info.py
+++ b/ehrql/dummy_data_nextgen/query_info.py
@@ -249,7 +249,7 @@ def specialize(query, column) -> Node | None:
             if rhs is None:
                 return lhs
             result = Function.And(lhs, rhs)
-            assert len(columns_for_query(result)) == 1
+            assert len(columns_for_query(result)) <= 1
             return result
         case Function.Or(lhs=lhs, rhs=rhs):
             lhs = specialize(lhs, column)
@@ -257,7 +257,7 @@ def specialize(query, column) -> Node | None:
             if lhs is None or rhs is None:
                 return None
             result = Function.Or(lhs=lhs, rhs=rhs)
-            assert len(columns_for_query(result)) == 1
+            assert len(columns_for_query(result)) <= 1
             return result
 
         # TODO: This could really use a nicer way of handling it.
diff --git a/tests/unit/dummy_data_nextgen/test_query_info.py b/tests/unit/dummy_data_nextgen/test_query_info.py
index 5a92e49c6..0b71f2dc1 100644
--- a/tests/unit/dummy_data_nextgen/test_query_info.py
+++ b/tests/unit/dummy_data_nextgen/test_query_info.py
@@ -1,6 +1,6 @@
 import datetime
 
-from ehrql import Dataset, days
+from ehrql import Dataset, days, maximum_of
 from ehrql.codes import CTV3Code
 from ehrql.dummy_data_nextgen.query_info import ColumnInfo, QueryInfo, TableInfo
 from ehrql.tables import (
@@ -147,3 +147,53 @@ def test_query_info_ignores_complex_comparisons():
     column_info = query_info.tables["patients"].columns["date_of_birth"]
 
     assert column_info.values_used == [datetime.date(2022, 10, 5)]
+
+
+def test_query_info_specialize_bug():
+    # This test reproduces an error encountered in real-world ehrQL which included
+    # queries that specialised down to comparisons between pure values and an
+    # unspecializable query involving a column, and resulted in a query with
+    # no column references that hit the assertion that len(columns_for_query(result)) == 1
+    dataset = Dataset()
+
+    # We can't create `Value(False)` directly in the query language, so this is the
+    # easiest way
+    pure_value = maximum_of(0, 0) != 0
+
+    # We need something which is not a pure value but which `specialize` can't currently
+    # handle. This is probably the simplest option but there are lots of others. I've
+    # listed a couple of others below.
+    unspecializable_thing = events.count_for_patient() == 0
+    # unspecializable_thing = events.sort_by(events.i).first_for_patient().i == 0
+    # unspecializable_thing = events.i.count_distinct_for_patient() == 0
+
+    # This is the core of the problematic construction. It also works with `&` instead
+    # to hit the other assertion.
+    query = pure_value | (pure_value & unspecializable_thing)
+
+    dataset.define_population(query)
+
+    # Ensure there's at least one column reference in the dataset (doesn't matter what
+    # it is) so that it always tries to specialise
+    dataset.dob = patients.date_of_birth
+    QueryInfo.from_dataset(dataset._compile())
+
+
+def test_query_info_specialize_bug_values_used():
+    dataset = Dataset()
+
+    pure_value = maximum_of(0, 0) != 0
+
+    # Define the unspecializable thing using a column reference, so we can confirm that
+    # the query info includes the expected values_used
+    unspecializable_thing = (
+        events.sort_by(events.date).first_for_patient().date == "2020-01-01"
+    )
+
+    query = pure_value | (pure_value & unspecializable_thing)
+
+    dataset.define_population(query)
+
+    query_info = QueryInfo.from_dataset(dataset._compile())
+    column_info = query_info.tables["events"].columns["date"]
+    assert column_info.values_used == [datetime.date(2020, 1, 1)]