Hi, It seems like it expected behavior. From the demo I can assume that it's not for those type of "document". Imagine that you have a normal document, where you have a 2 columns of text. You wouldn't want to read it line by line like you want to do with receipts Unfortunately I try to do the same thing, so I have to stay with simple previous iOS OCR solution and trying to connect those in lines by my own (not very good) algorithm.

Maybe putting the raw ocr result in apple intelligence ml model with wort of try?