Looking for public bank-export samples to try on the AI Bookkeeper, I imported a Revolut activity CSV from a public GitHub repository (made-up rows, layout only). It read fine, but the importer had not recognised it as Revolut: the detector wanted a trailing Balance column, and this layout has none.
That matters because the Revolut reader takes the Fee column off the amount, and only runs when the layout is recognised. A balance-less export with a fee would have booked the amount without it. The detector now accepts the layout with or without Balance.
Two tests pin it: the sample's layout (BOM, m/d/yy dates, Savings cashback beside Current card rows) and a 1.50 fee on a 100.00 exchange, booked as -101.50. Full suite: 2,340 pass. Not tested: a real Revolut account.