JSON flattener

Flattened JSON is what a lot of systems actually want: translation files, feature-flag stores, environment configuration and analytics payloads are all flat key-value maps, and a nested document has to be flattened to fit. The notation is the one those systems already use — dots between object keys, square brackets around array positions, so user.tags[0] reads back unambiguously.

Empty arrays and objects have no leaf to flatten into, and they are kept as explicit empty values rather than dropped. That is what makes the round trip work: a tags: [] that vanished on the way out would come back as "no such field", which is a different statement about the data.

Two shapes break the round trip, and both involve a key that looks like something else. A key that already contains a dot; {"a.b": 1}. Is indistinguishable from nesting once flattened, and comes back nested. Worse, a numeric object key becomes an array index: {"counts": {"1": 5}} flattens to counts.1 and unflattens to {"counts": [null, 5]}, because a numeric path segment can only be read as a position. If your keys are years, IDs or numeric codes, dotted notation is the wrong tool for them.

Flattening is also not the way to get JSON into a spreadsheet, though it looks close. It produces one object with many keys, not rows and columns: the JSON to CSV tool does that job properly, header row included.