diff --git a/cf_xarray/accessor.py b/cf_xarray/accessor.py
index 08bc28ea..29d63e31 100644
--- a/cf_xarray/accessor.py
+++ b/cf_xarray/accessor.py
@@ -24,7 +24,6 @@
 
 import xarray as xr
 from xarray import DataArray, Dataset
-from xarray.core.arithmetic import SupportsArithmetic
 from xarray.core.groupby import GroupBy
 from xarray.core.resample import Resample
 from xarray.core.rolling import Coarsen, Rolling
@@ -997,7 +996,7 @@ def _get_possible(accessor, criteria):
         return _get_possible(obj.cf, y_criteria)
 
 
-class _CFWrappedClass(SupportsArithmetic):
+class _CFWrappedClass:
     """
     This class is used to wrap any class in _WRAPPED_CLASSES.
     """
diff --git a/cf_xarray/tests/test_accessor.py b/cf_xarray/tests/test_accessor.py
index 8860b58a..c3f7005a 100644
--- a/cf_xarray/tests/test_accessor.py
+++ b/cf_xarray/tests/test_accessor.py
@@ -1524,15 +1524,14 @@ def test_groupby_special_ops() -> None:
     grouped = airds.groupby_bins("lat", np.arange(20, 50, 10))
 
     # __iter__
-    for (label, group), (cflabel, cfgroup) in zip(grouped, cfgrouped, strict=False):
+    for (label, group), (cflabel, cfgroup) in zip(grouped, cfgrouped, strict=True):
         assert label == cflabel
         assert_identical(group, cfgroup)
 
     # arithmetic
     # TODO: Extremely buggy!
     # 1. cfgrouped - cfgrouped.mean() raises RecursionError
-    # 2. use_flox=True doesn't preserve attributes for some reason
-    with xr.set_options(keep_attrs=True):
+    with xr.set_options(keep_attrs=True, use_flox=True):
         expected = grouped - grouped.mean()
         actual = grouped - cfgrouped.mean()
     assert_identical(expected, actual)