The name of the collision node is often extremely important, because the collision handler is often configured to generate an event in the case of a collision, based on the name of the collision node. Thus, the egg loader is always careful to preserve the name of the collision node whenever it appears in the egg file.
There is no such guarantee about geom nodes. The egg loader is free to rename these or even combine them with siblings, as it feels it needs to, in order to optimize the scene. We do this because, for 99% of the named geometry nodes within an egg file, you don’t care about the name or even keeping it separate from other nodes, and you’d rather have it render as fast as it can even if it means losing some of that hierarchy.
There are exceptions, of course; and when you do care about preserving the name of a geom node, you should tag it with the flag, which exists for this very reason.
I’m not 100% sure what the egg loader will do if you use the “keep” collide tag in conjunction with the flag. That’s kind of an awkward combination. The right thing would be to name both of them, I suppose, but then you’d still have to be careful with your find statement, since it is now ambiguous. Really, if you want to be able to get a handle to your geometry node, you should probably separate it completely from the collision geometry.
In many cases, keeping the collision geometry separate is best anyway. Often you want to have many thousands of polygons in a visible model, but only a few dozen in the collision model.
David