generally you need an approach like Separating records but in this case you split firstly the records per subfield without any cleansing and checked in a second step the content for "LFS-Nr." and/or further chars to identity if this record is garbage or not and in a third step could be the cleasing from the valid record follow.
Well, for now I can make it - I go by the nr. of letters. The maximum that can usually be expected to be there in a "clean" record is 6 ("Stueck" - pcs.) - so whenever there's more than 6, I say that record is garbage and needs post_processing.
However - you know the one about the race between developers and nature ... There's no limit to the creativity of users ...