Exporting

Formats and scope

A table exports to XLSX or CSV, from the Export tab in its settings. Two things can be narrowed before it runs:

• A row range, which is how you take the first fifty rows of a long table rather than all of it.
• Which columns to include — useful when a working table carries scratch columns you would rather not hand over.

There is also a bulk export at the workspace level, which takes the tables as they are without the row and column scoping.

What the file contains

One row per document, one column per extraction column, and a leading Document column identifying the row. Where you have overridden a cell, the file carries your value, not the model’s — the export is the table as you have left it, which is what makes it safe to send.

Citations and confidence do not come out in the file. An exported table is the answers without their evidence, so the checking has to happen in the application, before the export, rather than by whoever receives the spreadsheet. This is the main reason to work through the failed and low-confidence cells first — see citations and confidence.

Nor do pictures. A cell from an image column exports as the description of what was found, because a spreadsheet cell is text. That is worth knowing before building a deliverable around one: the picture is a thing the table has and the export does not, which is an argument for sending the table rather than the file.

Exporting a bundle’s record

A bundle exports too, and it is a different document from the above. Three formats: CSV, XLSX, or a branded PDF report that leads with the bundle and signs off with a link back to it.

The shape is table, column, value — one line per question asked of that bundle — rather than the document-by-column grid a table exports. A bundle’s record has no such grid to flatten: each table asks it a different set of questions, and two tables’ identically named columns are different questions, so a matrix would need one column per name and would quietly merge them.

A question that has never been answered exports as an empty value rather than being left out. It was still asked of this bundle, and a report that silently omits the unanswered ones overstates how much is known.

One detail about spreadsheets

Any value that begins with an equals sign, a plus, a minus or an at-sign is exported with a leading apostrophe, so your spreadsheet shows it as text instead of evaluating it as a formula. A clause quoted from a document is text; it should not become a broken calculation because it happened to start with a minus sign.

The PDF report is the exception, and deliberately: it has no formulas to inject, so the apostrophe there would be a visible typo rather than a precaution.

What export is not

It is a snapshot, not a connection. A file exported today does not update when a cell is rerun tomorrow, and nothing in the file records which version of a column produced a value. If a table is going to be relied on repeatedly, keep the table as the record and export from it, rather than keeping the spreadsheet and hoping it is still current.