DataFrame.where/mask with a column-indexed Series wipes the frame

df - s aligns a Series on the columns, but df.where(s, 0) and df.mask(s, 0) align a Series condition on the row index. With a per-column boolean mask like (df.max() - df.min()) != 0, no labels match, the condition becomes all False, and where() replaces every value. mask() also replaced every value in the test. There is no warning or error (pandas 2.2.3 and 3.1.0, 2x2 frame). The documented fix, where(s, other, axis=1), works only from pandas 3.1.0: on 2.1.4, 2.2.3 and 3.0.0 it raises TypeError: 'int' object is not subscriptable, also with a float or default other. Portable options are to broadcast the mask yourself, df.where(np.broadcast_to(s.values, df.shape), 0), or column assignment df.loc[:, ~s] = 0.