The big problem with VNs is, where exactly do they get the tremendous
diversity of compounds necessary to build themselves? Here is an
example of what I mean. Let's say you've got a shiny new VN: the
Ian_Parker_3000 mark 1. It's a very complex machine. It's not made
out of mud bricks and straw. It has some very sophisticated
constituent parts. For instance, many of its electrical connectors are
gold-plated to resist corrosion. It can build a copy of itself, sure,
but one of the things it needs (one in a long long list of things) is
gold.

As you mention, you can process lunar soil and obtain lots of different
compounds, but I seriously doubt you'd be able to find enough gold to
make the job worthwhile. If one Ian_Parker_3000 has to process so much
soil in the act of building one copy of itself that it digs a hole
that's visible from Earth, well I get the feeling it will be at or near
the end of it's life expectancy by the time it's done with that task.
It certainly isn't going to have time to do anything useful.

At any rate, long before we start talking about machines that build
themselves, how about we talk about machines that build *anything* in
space. If you could design a reliable solar panel that could be built
out of nothing more than lunar regolith, and if you build a robot that
builds those solar panels, then I think you'd have something a lot more
useful in the here and now than a VN Machine. If the robot could build
more solar panels (by mass) than the mass of the robot itself, then
you've saved the cost of transporting those solar panels to the Moon.

The robot I've described is complex, and quite possibly beyond our
present technology. Yet it's many order's of magnitude easier to build
than a VN. If you can't build this robot, then there's really no way
that you can build a VN machine.