Hmm… I see. I was trying to get that alternative path to the destination node by calling “find” from the current root-cell, but I realise now that there was no individual node in each path on which to place separate transformation data.
It’s tempting to simply set all of my cell root-nodes to have a position of (0, 0, 0) (which is what I presume that you mean by having them be “without a position”–if I’m mistaken, please correct me!). Intuitively, it makes sense for the cell-roots to have a position, so that their level of the scene-graph corresponds roughly to the layout of the cells. However, since they’re being portal-culled, perhaps that’s superfluous, with the positions of their child-objects being enough for the remaining scene-culling…
As to “instanceTo”, since the approaches are so similar, is there any reason to not use that method rather than attaching them by hand via “addChild”? I imagine that, even if I did go through with the culling change discussed above, “instanceTo” would still produce the intended scene-graph, and so should still work for my purposes…