>>10
M-expressions aren't limited to explaining evalquote in this manual, they also describe the elementary functions. Elementary functions are used for ultimately implementing evalquote in m-expressions, if that's what you meant by this, m-expression to s-expression separation can be seen as explaining hosting bootstrapping.
The glossary eludes to part of why m-expressions are dropped.
M-expressions cannot be read by the machine at present but must be hand-translated into S-expressions
The machine can either be the literal ibm 7090 or a backplane using a theoretical hardware LISt Processor. This is partly useless for frankenstein harvards that don't have one and a ibm 7090 or LISt Processor that can't read it. A formal language was also required for the advice taker project, m-expressions are considered a functional notation while LISP is considered a formal mathematical language. The manual describes a formal mapping of m-expressions into s-expressions.
where are m-expressions today?
There was a hobby operating system I remember using two layers of languages, the top layer was a mix of m-expressions and asm, the bottom layer was a incomplete dialectic of LISP, common metal kernel and userland software architecture for operating systems. Even if it's considered decrepit in favour of pure s-expressions, it's still used.