[dogada]

I agree that identification is important issue, but IMO it's protocol level issue. RJSON is not protocol, it's algorithm. Someone will prefer to wrap all in {"rjson": ...}, someone like {"format": "rjson", "data": ...} and so on. I belive, algorithm itself should create as less limitations as possible.

[rubyruy]

The MIME type is allowed to add information about the nature of the data encoded on top of the encoding. This is why we have application/xhtml+xml. It means "this is encoded using xml, and btw the parsed xml structure is xhtml".

The ideal place to handle this is (as pointed out above) content negociation - specifically as 'application/rjson+json'

[gcb]

It's an algo. But the resulting data is a protocol.