We currently have a high-level description of the data model. It is good enough to understand what we do, but not detailed enough for an independent implementation with the same expressive power.

We also have a detailed description of the serialization format, but that is really low-level.

We need an in-between thing, maybe something like a formal ADT description, that allows for independent implementations (not necessary compatible, but capable of the same expressivity) without worrying too much about byte orders and the like.

Related T1257.