> Asm is simple enough

The general conceptual model of "asm" is simple.

Some instruction sets and architectures are hideous, though.

> merely being a human compiler is not particularly enlightening nor useful.

I don't think I can agree with that. At least it teaches you what the compiler is doing. And abiding by conventions (HLL-esque control flow, but also things like "put the return value in r0" and "put constant pools after the function") can definitely make it easier to make sense of the code. (Although you might share a constant pool across a module or something, if the instructions reach far enough.)

Not to say that you can't do interesting things, and can't ever beat the compiler. One of the things I most enjoyed discovering, in mid-00s era THUMB (i.e. 16-bit ARM) code, is that the compiler was implementing switch statements with tables of 32-bit constants that it would load into an indirect jump. I didn't get around to it, but I figured I could mechanically replace these with a computed jump into a "table" of 16-bit unconditional branches (except for very long functions, but this helped bring the branch distances under thresholds).