If a transaction description on your Public Bank Berhad statement is long enough to wrap across two or three lines, it still needs to land in a single row once converted. When a transaction description runs long enough to wrap across two or three lines inside the statement, a generic parser often slices it at the line break and treats each fragment as a separate row. The engine does not rely on line breaks to guess where one transaction ends and the next begins. It reads the table structure as a whole, matching the description cell to its amount, date and running-balance column so the full text stays in a single record. A row that splits oddly in the source lands in the export as one clean entry, with every field in the right column.
When your Public Bank Berhad 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. Statements arrive as crisp digital exports and as photographed sheets captured under office lights, sometimes at an angle, with shadows or blur along the edges. The engine applies OCR to scanned and photographed pages, then runs the same running-balance check it uses on every other statement. The closing balance on the document is compared line by line against the extracted figures; a digit that the scan misread, like a 3 mistaken for an 8, breaks the arithmetic and flags that row for review before anything reaches the ledger.