yeah, the point isn't so much to make it readable, but to allow you to write JSON (as a python dev I love the interoperability with JSON) and then use the capabilities of the DataPower to do stuff with it.

I expect many people were doing this by hand since JSON has replaced XML in a lot of peoples minds, and IBM has created a way to standardize it to make it easier to work with other teams.

That doesn't make this mess any saner, if the point was to apply XML tools to JSON, why not convert to an existing XML dialect with equivalent expressive power, e.g. XMLRPC's serialisation format (which provides a strict superset of JSON capabilities (if you include the nil extension))?

The obvious problem being that you might not have control over the source of the data. JSONx means you can apply xml tools to json sources, period.

I'm not sure what you're trying to say. The source of the data is irrelevant, XMLRPC w/ nil (and other pre-existing formats I'm sure) are strict supersets of JSON, any JSON document can be converted back and forth between XMLRPC and JSON without data loss. So XMLRPC (or something else with similar capabilities) would give you exactly the same ability to apply XML tools to "JSON sources".

I've never understood the reasoning behind using a textual serialization format that's not human readable.

If you don't want it to be readable, make the darn thing binary and go for efficiency.

There's a difference between prioritizing readability and simply being readable.

Please explain the trade off, I can't see myself debugging an unindented XML file without an appropriate reader. Exactly like a binary file.