Error dataset
A separate dataset holding the rows that failed, with the reason for each, so a partial run is auditable rather than silently incomplete.
Keeping failures out of the main dataset is a data-quality decision; not charging for them is a billing one. Both matter, and they are independent.