When your LfA Förderbank Bayern 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. Scanned and photographed statements add extra risk because the source is an image, not a clean data file. Every figure on the page has to be read by optical character recognition before it becomes a row in a table. A 3 misread as an 8, a decimal point that disappears in a crease, a column header shifted half an inch: any one of these will throw the numbers off. Convert·Into routes every uploaded image through an OCR pass tuned for financial documents. The engine looks for the statement's own printed closing balance, works backwards through every transaction, and flags anything that no longer adds up. You see the suspect cell, not just a rejected file.
If a transaction description on your LfA Förderbank Bayern statement is long enough to wrap across two or three lines, it still needs to land in a single row once converted. Transaction descriptions that run long and wrap across multiple lines are a structural problem. A converter that treats every new line as a new row will split one real entry into two or three phantom rows, each carrying a fragment of text and none carrying the correct amount. Convert·Into reads the layout grid first: column position, font metrics, ruling lines. It recognises that text continuing on the next line still belongs to the same transaction and re-joins it before export, so the description lands in one cell and the debit or credit amount stays in its proper column.
If the running balance on your LfA Förderbank Bayern 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 statement the engine processes goes through a running-balance check against the closing figure printed on the original document. The extracted transactions are summed from top to bottom, and the total is compared to the statement's own declared closing balance. A match means the extraction is arithmetic-grade clean. A mismatch means a row somewhere failed, and the engine flags it without rejecting the rest of the file. You review the one flagged entry, not the whole batch.