1thou under cal has always worked well for me.
I would still prefer a hardened mandrel right at cal.
There is no formula for any of this. Just anecdotal setting of interference, given normal spring backs.
Normal spring back(either way) is ~1/2thou.
So for example, let's say a conservative reloader does not want to use his bullets for up-sizing of necks. And he wants high consistency reasonable neck tension, so he's not going to be annealing constantly, but minimally working his necks. He also will never size length of neck beyond seated bullet bearing.
In 22cal he measures loaded neck diameters of .248 with 12thou neck thickness. If his neck downsizing to desired, from fired dimension, is 5thou or less, he can go with a bushing stamped .246. The necks will spring back to ~.2465 OD which is ~1.5thou under cal interference.
If he were then to run a mandrel at cal (.224) through sized necks, they would up-size & spring back to ~.2475, leaving 1/2thou of interference.
With that, his seating bullets would expand necks but would not be up-sizing at all. A perfect pre-seating.
I know this would work fine, but you can add any effort you like from there. Maybe a .2455 bushing/.223 mandrel.
From there keep in mind that a circle of excess sizing (up/down, and lengths) and annealing, feeds on itself in the long run. You can be sure of that much.
If bushing sizing to desired has you downsizing >5thou, go with a slightly higher stamped bushing. This much sizing angle rolls the necks inward, past that stamped on the bushing. That's common but it works necks a lot.