Doug, thanks for responding. I thought my post got lost in the noise.

The final operation need not always be on the same class machine, but if,
working backwards from the finished part, even _one_ operation on _one_ part
requires a machine of the same class as the downed, part-removed machine, we
can reduce the problem to pairs but no farther.

I still don't accept your assertion that two of every tool is the minimum
requirement. I accept that two of some things establish a convenient
baseline for replication and repair (a broken lathe may not be able to
manufacture its own repair part). But "two of every tool" can not be a
universal minimum, especially when applied to tools like hex key sets and
collets.

...And this would be a pair of bench grinders, a pair of hex key sets,
pairs of all collets required, but not necessarily a full set, a pair of
collet closers, etc.

As an example, the still functional 1/2" collet does not participate in the
replication of another 1/2" collet to replace the damaged one. The same
situation exists with hex keys.

Another example that illustrates "fewer than two" being sufficient is folks
building the "Gingery lathe" - self generation through successive
approximation. The first spindle is used only until the second (improved)
spindle is made. As an example, you mention that at least one part of a
surface grinder is built using a surface grinder. If that part were to fail
on the solitary surface grinder and make the machine inoperative, one could
build several replacement parts using the lathe and mill. Install the first
crudely made part to restore limited operational capacity to the surface
grinder and grind the second replacement part. Now replace the first
generation crudely built part with the slightly improved part. Repeat this
successive approximation approach until the surface grinder is completely
restored.

If you have two lathes and both are in routine service, there is a possible
although improbable situation where both lathes could fail at the same time
through failure of identical lathe-built parts. In this situation, one would
need to rely on successive approximation or perhaps outside assistance. Two
of every tool is not a perfect guarantee by itself.

Bruce

snip