Importing a spreadsheet

You can import a CSV file from your dashboard to turn an existing spreadsheet into a collection. No assistant is needed.

  • One column names the thing each row is about.
  • The columns you keep become the collection's attributes.
  • You choose the collection's name. Use the name of a collection you already have to add the rows to it instead.

How rows are matched

If a row's name matches an entity you already have, anywhere on your account, the row is added to that entity rather than creating a copy. The entity joins the collection and keeps everything recorded about it before.

A row is matched in this order:

  1. By its name, exactly as written.
  2. By any name the entity had before a rename or a merge.
  3. By an identifying value, such as a ticker, website or DOI. Point that column at an attribute marked as identifying and the row joins whatever already holds that value, whatever it is called. This only works if the attribute is marked before you import.

Close names are not matched. "Open AI" and "OpenAI" arrive as two entities, and the pair is shown to you afterwards as a likely duplicate. Merging them takes one step.

If two of your entities share the same name, an identifying value in the row decides which one it means. With nothing to go on, the import creates a new entity rather than guess, and tells you so in its report.

Nothing is overwritten

Imported values are added, never written over existing ones.

  • If your file disagrees with a cell, the cell holds both claims and shows one of them, just as when two assistants disagree.
  • If your file agrees, it becomes a second source for the same value.

This means it is safe to import a newer version of the same sheet. Cells that changed will show more than one claim.

The file is kept

The file you upload is kept, and every value the import wrote lists it as its source by name. From a value, you can open the file itself and a report of what the import did:

  • how many rows joined entities you already had,
  • how many rows were skipped for having no name,
  • which attributes it had to create.

Months later, you can still tell which spreadsheet a figure came from.