I did something like this a couple years ago, I'm not sure if this is best practice, but here's how I did it (this is off the top of my head, I don't have the script handy, may be forgetting or misremembering a bit):
- I identified a subnetwork that had minimal geographical overlap with other subnetworks. I didn't use any UN/AP attribution for this, it was a rural area that I knew (from the business) that was mostly on one circuit.
- I created a new polygon feature class with one feature, a manually digitized polygon to encompass this area.
- I wrote a Python script that used the polygon feature class to do the following:
- For each of the domain/structure classes in the AP, delete anything outside the polygon
- Use a Python set and a series of search cursors to compile all Global IDs that remained (this also included the tables for non-spatial objects)
- Open an update cursor on the C_Associations table and delete anything that had a "to global id" or "from global id" that was missing from the Global IDs set
- Build the AP
- I added two subnetwork controllers for the subnetwork manually after the build, rather than use the C* table for those
This did result in some leftover nearby features that were parts of other subnetworks, but it didn't really matter for what I was trying to do (identifying the most common errors to refine the ETL process, determining whether there were rules missing I needed to add, picking up ETL omissions like missing J:J associations, etc.) I reused this process a few times as I made further changes to the ETL process, sometimes for the same region I used the first run, and sometimes for different regions.
