When your bunq statement arrives as a scan or a photo of the page rather than a clean export, the numbers on it are just pixels until they are extracted. A statement document that arrived as a scan or a photograph presents the same visual layout as a clean export, so the engine treats it the same way. Optical character recognition reads the table grid and the numeric columns; the printed running balance is then verified line by line against the closing figure on the document. Rows where a character was misread, a 3 mistaken for an 8 or a decimal misplaced, cause the running balance to break, and that row is flagged for review while the rest of the file proceeds.
Whichever order your bunq statement prints its columns in, each one still has to land in the right place once it reaches a spreadsheet. Column mapping is automatic because the engine identifies the statement layout on upload. Transaction dates, descriptions, amounts and balances land in the order the document uses, and those field positions are preserved straight through to the export. When you choose Excel, CSV or QBO as the output format, the columns you see in the preview are the columns you get in the file, with no column-shuffling step to run by hand.