Regression in hash diffing logic

[mcmire]

It seems like some recent changes have negatively impacted how diffs between hashes are represented. Unfortunately I've had to hide this behind a secret gist since this is for work, but regardless, it's creating confusion: https://gist.github.com/mcmire/a4d0381909b479491dd4ec23c42192e4

[mcmire]

This also might be related to #99.