As the Daggerfall paper doll uses image output to click on items to unequip, there must be a way to sample which pixels the player is actually clicking on. It's possible to click on items through other items (e.g. click on pants or shirt though gaps in armor) so broad-phase rect and layer order alone isn't adequate. The final test needs to be done on a per-pixel basis to remain accurate to the Daggerfall way of doing things.
It would be possible to use the classic 8-bit image for pixel tests, provided the replacement item is only slightly different in shape. But it would no longer be pixel perfect due to resolution differences, and modded items might have a different topology to classic items. So a simple pixel test is the best method to preserve Daggerfall's way of doing things.
I agree with what you're saying though. I might be able to refine shader to output mask information to a second RenderTexture, which can then be sampled for hit tests under the mouse without requiring modded texture to be readable. I've added a note to my internal tracking and will look at this when I can.
This whole setup would be a lot easier if Daggerfall used equipment slots everywhere like it does for jewellery, etc. Then it's just a simple rect check on the slot. Alas, this game loves to challenge me.
