Thank you, Toby et al for your answers, opinions, advice, criticism, etc.

I almost think I'm beginning to get it.

Let me regurgitate this to be sure:

Given &&&&&&mlv&mlin., I resolve &mlin. = 5 and reduce the &&&&&& to &&& to
get &&&mlv5.

Then I resolve &mlv5. = DAISY and reduce the && to & to get &DAISY.

Finally I resolve &DAISY. = YELLOW.

So, it looks to me like once I reach a stage where I have a string of
ampersands followed by a "pure" name (i.e., one with no ampersands in it),
that string of ampersands should have length 2**n - 1 for some n, e.g.
&&&&&&&x.  Then, I would always resolve the last "ampersand plus pure name"
combination, and halve the remaining ampersands.

Yes, getting into a six-ampersand situation probably does point to a poor
initial design.  But this was one of those cases where I built the data for
one purpose with a set of dummy variables, and then had to cull some of the
data for a subsequent analysis, depending upon which dummies were non-
zero.  The details of that are not important here; suffice to say that I
was able to do what I needed to do, and you helped me unscramble the &
mess.  By the way, I found Chang's suggestion about using an intermediate
macro variable to take care of the first resolution stage very helpful.

--  TMK  --
"The Macro Klutz"