We want to store amendments as a normalized data structure, as opposed to freeform AsciiDoc contents.

This can be completed in stages. Full completion may require us to normalize datasets as well, though it’s not clear whether it’ll be part of this milestone.