Should we also mention the idea of "informationally equivalent". I suggest adding this to the Note:

"In some cases all distributions of a dataset will be fully informationally equivalent, in the sense that lossless transformations between the representations are possible. An example would be different serializations of an RDF graph using RDF/XML, Turtle, N3, JSON-LD. However, in other cases the distributions might have different levels of fidelity to the underlying data. For example, a graphical representation alongside a CSV file. The question of whether different representations can be understood to be distributions of the same dataset is use-case specific, so the judgement is the responsibility of the provider."