I am revewing the messages form December (you see, I don't review the messages very 
often), and what follows is may point about argument specifiers.

It seems clear that the problem is what do we want the arg-spec to specify. Two things 
were proposed:

1. The original and current one. (Explanation following is form Frank)

>>\foo:abc  % abc are placeholders for real arg specifiers

>>is intended to be a short form for saying

>>- do "a" with the first argument do "b" with the second and do "c" with the
   third argument prior to passing the argument values to the base function

2. Let the specifiers tell what we can pass to the function according to how it will be 

While the last one is quite appealing, when analysed I see a *BIG* trouble. The way the 
argument is treated my change if the implementation of the function is changed (for 
example, by pdfTeXing the kernel, as it was said). One may always update de doc, but 
one cannot keep changing the names of kernel commands.
I see another trouble too. The casuistic (about manipulating arguments inside a macro) is 
very wide. Too many arg-specs would be needed, and even then they the thefinition of 
many of then could not be rigorous (or if they are, there will be macros that do not 
exactly stick to what their arg-spec tells). This does not happen if the arf-specs are what 
Frank's dewscripstion above tells.

<<do "a" with the first argument do "b" with the second and do "c" with the
   third argument prior to passing the argument values to the base function

is a very precise definition. One of the things I like more about TeX (seen as a 
programing language) and TeX primitives is, that with all its oddities, the rules are 
precisely documented and each primitive behaves exactly according to precese rules, 
even if they are weird at some points.

So, my point is that I like few arg-specs with their current meaning and have well 
documented in de doc the other part, i,e., what happens to the argument in the function 
and hence what can be passed to it.

Andreas Matthias wrote:

>>When I first used these functions I added a lot of \exp_args:... and
>>\exp_after:NN just to realize afterwards that they are not necessary
>>at all. An x arg-spec would have helped me a lot to get things right
>>from the beginning.

Yes, but missinterpreting the specifications, or simply forgotting the precise rules, is 
something that from time to time happens to everyone. I, for instansce, was continually 

\expandafter\toks@a\csname toks@ #1\endcsname

(or something similar, now I can't remember) tilll I realized that the \expandafter was not 

But I agree that many people will be confused as you were. To avoid this, the 
documentations that explains what arg-spec mean should also explain what they don't 
mean, for example, saying that an n secifier that _not_ mean that the argument will not 
get expanded inside the function.

p.s.: I defend the original model but I was _not_ one of the people coming up with it.

p.s. II: I also want the T and F specifiers to remain in existence.

Javier A.