It’s only doable if you know where to put the actors in the first place. For example, if you want a collectable to be on a mountain that exists in one map, you’d need to place a dummy object, copy its location axes individually, and paste them into the streamed map’s collectable transform. That’s simply not viable if you have more than a few game mode specific objects. For example, Banjo-Kazooie levels each contain 100 notes. If one wanted to make a 100-collectable game mode, that would require copying and pasting 300 separate values, and every time you wanted to make modifications to the location of the collectable, you’d have to go back and forth between the main and streamed levels placing dummy objects in the main level to find out where they should go and the copying and pasting the new axis values into the streamed level.
In my case, each map contains 33 collectables that are needed for the time attack game mode and anywhere between 50 and 100 for the sandbox game mode. Then, there’s race mode specific checkpoints and boosters. And, there’s a roller coaster track with boosters and additive visual elements for the coaster mode. And, another track for the tour mode. And, that’s only the modes for a single map! Iteration (or initial placement, for that matter) would be practically impossible if I couldn’t see the level geometry while I was placing all of that stuff for at least five separate game modes per map. Streaming from a separate level would only be viable if we could see another level overlaid on the to-be-streamed level we’re working on.