My version used strings as well, and I thought that as I was comparing small integers either way, it made sense to stay in ASCII as the strings were already easy to index, and it meant I could skip parsing input numbers, only needing to parse output numbers so they could be summed.
I did start with numbers so I could convert it back to compare, but it's so fast (the whole thing takes 1ms - and that's reading/parsing the input twice) that it's almost a micro benchmark.