So that kind of approach would require additional logic outside of the DLT pipeline itself? Sounds doable, but the added complexity worries me, a lot. 

Also, I don't understand why would I transfer data back to the main table because it's already populated in the end of refresh. Could you elaborate?