ai-bookkeeper keeps checking its bank readers against files other people wrote. This run swept about 40 public samples: 30 OFX fixtures from ofxtools and ofxparse (bank, card, investment, error replies), six camt files, and two Lithuanian CSVs.
The OFX and camt files read cleanly, including odd amounts ($120, (45.00), 12.50-) and date forms. The Swedbank Lithuania CSV did not: its record type 82 rows, labelled "Apyvarta" (turnover), are the month's debit and credit totals. They were imported as two extra transactions of -806.52 and +600.00, so the month was counted twice. They are skipped now, as the opening and closing balance rows already were, and a test covers it.
The sample is the bank's layout with test data, so this shows the rows exist in that export; it says nothing about how common the format is among users.