If you are working with Primo VE and not Primo, see Understanding the Dedup and FRBR Processes (Primo VE).
There are several scenarios in which you may want to run the dedup process for the pipe, but you want to prevent specific records from being deduped and/or you want to ensure that dedup only works within a specific institution.
As mentioned previously in Duplicate Detection Vector, it is possible to suppress specific records from taking part in the dedup process by assigning them type 99 via the normalization rules (dedup/t).
If you want to prevent cross-institution deduplication, you can consider using the Harvesting, NEP, Dedup & FRBR parallel processing mode. In this mode, dedup and FRBR run as part of the pipe, and only records that belong to the institution of the pipe are matched with each other. Note that the institution in this case is the owner of the data source and pipe. If the data source includes multiple institutions, they only dedup among themselves.
Another method to prevent dedup across institutions or across any type of group is to add a key (such as the institution code) to the candidate fields. This ensures that candidates, and hence the match, are only found within the group.