I'll keep that in mind the next time I have to work around pandas automatically converting strings to floats. Or ints to floats (nope, no NaNs in either the inputs or the output). I'm sure it'll make me feel better about my data getting corrupted.