We wanted to avoid needing to spell out diffing mechanisms for every conceivable format in the core spec.
I get this, but why not just allow for format-specific diffing without specifying it? Show the atom one as an example and leave the specifics up to implementation.