The problem

With rectangular data you always know how to start. A data frame has already answered “what is a row” before you arrive, so the same verbs work on every table you will ever meet. Technique transfers. That is why those verbs are worth learning once.

JSON never answers that question. Every document makes you work it out from scratch, which is why the code you wrote for one file is useless on the next one, even when the two are nominally the same kind of thing.

That is not a skill gap. It is a property of the format, and it is why this is a different animal from every other data-cleaning problem.

This part is about the document rather than the tool. The first chapter is why JSON defeats technique, in the author’s own account. The second names the properties that make one document harder than another, as separate axes — and records which of them turned out not to be axes at all.