Audit Dimension Doubt

I'm currently studying the datawarehouse theory along with one of my friends from Kimball group's books and we're stuck at the part
where they talk about the audit dimension.
We understand that most of the fact records point to a single row in the audit dimension which represents a clean run for the whole ETL
process. However, we're not sure about the number of generated rows when we find errors during this process since the severity score on
each screen is an arbitrary number and each row might have a different total screen category score (which would be logged into the
audit dimension as completeness, validation, or out of bounds score).

We've had some ideas, like adding up the severity score for all the rows which had only validation errors, all which had only completeness
errors and all which had both kinds of errors and generate at most four rows per ETL run, but we're not sure if we're misunderstanding what
the book states about those values.

It is a little confusing. Try this article and see if it helps.
Architecture for Data Quality