It’s up to each developer to decide how they want to use LOD. Some may have decided that their models are not complex enough that it would make much difference.
Each object that I export (from Blender) - so this is a single object that I can place in scenery editor, might be 1 or 5 or 10 different meshes in Blender. For my highest LOD, I’m using pretty much the original high res Blender meshes that each have their own texture. To create the next level down, I might join (in Blender) 2 or 3 of these meshes together before decimating the geometry. It can be quite a bit of work and can take hours, but some of my terminal buildings have become crazy complex and it would be asking way too much of even a high end PC to have to draw all of this geometry from a distance where you wouldn’t be able to see the detail anyway.
Another example is that some of my objects have 6 or so lights dotted around them. From a distance I can reduce this to 1 or 2.
The simplest way to achieve things is probably to NOT change the content of each exported object, but to just half the vertex count and reduce the texture resolution by 50% for each step down in LOD. Again, if your models are relatively simple there might be no need to do this, or to only use different resolution textures for each LOD, or just use the same textures but decimate the geometry.
I’ve followed the rough guides given in the aircraft examples as to how mesh detail is reduced per level of detail and applied this to my buildings too, where it made sense.
The documentation from Asobo is still lacking in lots of detail, so it’s still probably a case of ■■■■ it and see. Perhaps there is more scientific methodology used by other content developers in games that use a similar engine.