If the running balance on your BancoPosta statement does not add up row by row, that is exactly the kind of mismatch Convert·Into is built to catch rather than pass through silently. Every converted statement is checked against a single, non-negotiable source of truth: the closing balance printed on the document itself. The engine reads that final figure from its fixed position on the page, then steps backward line by line, verifying that the running balance after every transaction adds up. A deposit of 150 against an opening balance of 1,200 must yield 1,350, and that sum must match whatever the next line already states. When the arithmetic holds all the way to the top, the entire file is proven internally consistent; the export matches the statement exactly, with no silent drift. If a single row does not balance, that row is flagged for you to review, while the rest of the file stays intact. A 3 misread as an 8 breaks the chain, and that is precisely the sort of failure the check is designed to catch.
If the BancoPosta statement you have is a scanned image or a phone photo instead of an exported file, that alone does not stop Convert·Into from reading the transaction table off it. Scanned and photographed documents pass through the same pipeline as any other statement: the engine applies layout detection to the image, extracts the transaction table, and then verifies the running balance line by line. That verification step matters most here, because OCR can mistake a digit or merge two adjacent fields. The same balance check that catches a formatting anomaly in a clean file also catches a misread digit in a scan, so you know the export is reliable whether the original arrived by email or through a camera lens.
Should a description on your BancoPosta statement break across multiple lines of the printed page, that line break is not the same thing as a new transaction. Long transaction descriptions that wrap onto a second or third line often confuse a simple parser, splitting what should be a single row into fragmented pieces. The layout engine identifies the column boundaries first and treats the description field as a single span, so multiline text stays inside its own cell rather than bleeding into the amount column. No transaction is lost to a line break, and the export preserves the full description text in a single, clean field.