For now, fake the sample by using the current output of OCR, which needs to be improved in the future