I also wrote a small article on it, not very detailed but maybe it can help you get a rough idea of the power of clojure.spec when combined with generative testing -- [http://abhirag.in/articles/spec_oracle.html]
There's more today than there was yesterday, but it's still not a system expected to be built, maintained, and extended for decades.
> derive algorithms mathematically
Of course, most of what computers do is not algorithms (or even "computing").
> can be awful for projects requiring maintenance
Or wonderful, see Smalltalk. Not that it is perfect and can't be improved, but it sure has been successfully maintained over a long period of time.
Pssst! Hey, buddy, that's what an algorithm is.
Both at the macro level, what they are used for, but also at the micro level. Actual computation tends to be incidental.
What is speech recognition for? Communication.
etc.
I think you keep arguing against something I never said, which is that computers do no computing whatsoever.
A large set of people in that community want to derive algorithms mathematically, using a deductive or proof based process, as in math.
Iterative development conflicts with a top down form of development, and also conflicts with the safety oriented culture of strong typing. While rapid iteration languages and environments are good for initial development, they can be awful for projects requiring maintenance.
About a decade ago, I built a small web app using Common Lisp (SBCL) and a lot of the development I did was done through adding new features and debugging in the REPL. While I saved the VM, reading the code months later to add a feature was terrible because the app was hacked together. I wonder if there's a way to fix this. Typed Racket looks like a promising move in that direction.
FWIW, ghci already interprets Haskell quite quickly for iterative development.