Exporting

Press Done, or ⌘/Ctrl+Enter. You do not have to finish every record first.

Two files

Both are written next to your input file, named after it.

File	Contains	In which schema
<name>-output.csv	Every complete record	Your output schema
<name>-remaining.csv	Everything else	The input schema

A record is complete when every required output field holds a valid value — see Labeling. Partial and untouched records both go to *-remaining.

*-remaining is written in the input schema precisely so it can be loaded straight back into MLabel with the same config. That is what lets you stop early, or hand the rest to someone else. If nothing is left over, no remaining file is written at all.

Every exported row carries your session answers and, if the config declares one, a timestamp — merged in the same way whichever file it lands in.

A second export is refused

Exporting again over existing files fails with:

These files already exist: data-output.csv, data-remaining.csv.

This is deliberate, and there is no “overwrite?” prompt. Re-exporting is easy to do by accident and the previous artifacts may be the only copy of hours of work.

To export again, move or delete the previous files first. Then press Done again.

Both files, or neither

If the second write fails — a full disk, a permission problem — the first is removed again. An export never leaves a half-finished pair on disk to be mistaken for a complete one. Writes are atomic, so an interrupted export cannot leave a truncated file either.

After exporting

• Reveal opens either file in Finder or Explorer.
• Keep editing returns you to labeling with everything still in place.
• Label another file goes back to the file picker.

A successful export clears your saved session — see Sessions for exactly what that means.

What round-trips, and what doesn’t

*-remaining is value-faithful, not byte-faithful. Reloading it gives you back exactly the same values, in the same column order, in the same dialect — that is guaranteed and property-tested.

Incidental formatting is not preserved:

• a byte-order mark,
• whitespace padding in headers,
• the original quoting style (quotes are applied where needed, not where they were),
• blank lines,
• whether the file ended with a newline.

For loading the file back into MLabel, or into anything that parses CSV, this is immaterial. If you are diffing against the original file byte-for-byte, expect noise.