So many gotchas in this area. Just for starters people are frequently surprised by the fact that adding a single null to a series of ints turns it into a series of floats (with a NaN). And that really is just the first of umpteen surprises that lie in store in pandas’ type system.
And sometimes we can add a single null inadvertently, like with shift or diff functions. There are so many footguns with pandas, that I always have to double-check the results.