So it's not really a serialization format, it's a compact, modifiable untyped tree, that one can therefore send to another machine with the same architecture. Or deserialise into native language specific data structures.
Don't get me wrong, I find this type of data structures interesting and useful, but it's misleading to call it "serialization", unless my understanding is wrong.
I'm not sure what the distinction you are trying to make here is?
How does machine architecture play into it? It sounds like int sizes are the same regardless of word sizes of the machine, the choices made just happen to have high performance for common machine architectures. Or is it about endianess? Do big endian machines even exist anymore?
What is a serialization format, if not a data encoding "that one can therefore send to another machine" .. "Or deserialise into native language specific data structurs" ..?
Don't get me wrong, I find this type of data structures interesting and useful, but it's misleading to call it "serialization", unless my understanding is wrong.