I haven't read or understood the author's proof, but the "standard" proof probably goes like this (any corrections? anyways though, I have to admit this approach sucks all of the joy out of the blog post)
We take the standard definition of the reals as equivalence classes of Cauchy sequences.
Let a_n be the mixed-base number up to n digits, e.g. a_4 = 0.1234. Clearly it's Cauchy
Let b_n = 1 - a_n, e.g. b_4 - 0.00001
Obviously, lim n->inf (a_n + b_n) = 1
But since the sequences a and b are Cauchy, we can exchange limit and sum
1 = lim n->inf (a_n) + lim n -> inf (b_n)
But the second limit = 0, therefore the first = 1