But those full-blown Lisps have to have that full-blowness documented. It can't just be "here is the documentation for the ten primitives we used; to understand everything else, read the source".

It would be a bad doc if for every feature, it rambled on the details about how that feature is constructed out of the ten primitives, rather that specifying the relevant requirements for that feature.

Source? I’d love to read more about this

http://languagelog.ldc.upenn.edu/myl/llog/jmc.pdf The Roots of Lisp by Paul Graham.

Not car and cdr but close enough https://en.wikipedia.org/wiki/Church_encoding

Also, cons, car, and cdr need not be primitives. They can be built out of lambdas. In Scheme:

  (define (cons x y)
    (lambda (m) (m x y)))

  (define (car z)
    (z (lambda (p q) p)))

  (define (cdr z)
    (z (lambda (p q) q)))

(From https://mitpress.mit.edu/sites/default/files/sicp/full-text/..., exercise 2.4)

Sure. Now to implement consp, all you need is a global list which tracks all lambdas that came from the cons function; if the argument object is eq to anything on that list, then it is a cons! We also now need another primitive: weak pointers. We don't want our global list to prevent otherwise unreferenced cells from being reclaimed.

typecase is going to be fun to develop and optimize, with everything being a lambda.