Much to their ultimate chagrin, I reviewed a book that 
a friend of mine had backed via patreon. The book 
mirrored a little bit of what Norvig wrote in the 90s, 
in Paradigms of Artificial Intelligence in Practice. 
Norvig's book is an interesting piece of historical 
history, only valuable now for his views on high value 
programming technique in a serious language.
One thing Norvig took pains to achieve throughout the 
book was a finished implementation of prolog in common
lisp which was then reattempted by Paul Graham, from the
more engineering-side lisp hackers. In every case, the
motivation given was that prolog is a perfect domain
specific language: Hard to improve on by substantively
changing, just like common lisp.
Prolog is basically a continuation-passing implementation
of a graph search on input facts. Prolog won't do anything
else; that's why you would want both it and your favourite
hacking language in the same program.
But with independently world-famous lisp authors 
taking prolog as a DSL in lisp so seriously, why did it
never catch on (except for at Allegro and Lispworks) ?
I think I stumbled across the answer. It's that 
#'continuation-based-prolog-graph-search 
is not in the standard nor in #:alexandria.
Re-imagining a complete prolog from the ground up both
deviates wildly out from common lisp, which we know is a 
hard place to find improvements from the last 40 years, and
trying to beat prolog at its own game, which would also be
a first(ish).

I'm getting cold feet as I finish writing this, but I think
reimaginings of what common lisp should be like die off.