I tend to prefer line delimited JSON myself, even if it's got redundant information. It will gzip pretty well in the data if you want to use less storage space.

I tend to prefer line delimited JSON myself, even if it's got redundant information. It will gzip pretty well in the data if you want to use less storage space. Either that or use the ASCII codes for field and row delimiters on a UTF-8 file without a BOM. Even then you're still stuck with data encoding issues with numbers and booleans. And that direct even cover all the holes I've seen in CSV in real world use by banks and govt agencies over the years. When I've had to deal with varying imports I push for a scripted (js/TS or Python) preprocessor that takes the vender/client format and normalized to line delimited JSON, then that output gets imported. It's far easier than trying to create a flexible importer application. Edit: I've also advocated for using SQLite3 files for import, export and archival work.