Lets say that how to do a sum is pure abstract knowledge,  and this translates nicely to a declarative haskell sentence.

But to perform a sum of two concrete numbers is procedural, because either the program or the compiler or yourself have to extract from the available knowledge a  sequence of steps in order to obtain a new knowledge, that is , the  result. 

In imperative languages the sequentiation  is more explicit. in functional languages this is more implicit, because the compiler+ runtime do the sequentiation. 

In fact, a C compiler also perform an automatic sequentiation for this simple operation and generates a sequence of assembler code for either a sum or any mathematical expression, but the haskell compiler is way more powerful for extracting sequences of steps from declarative statements.   moreover, haskell promote a declarative style because laziness avoid to express abstract knowledge as sequences. Referential transparency avoid also dependencies of expressions from other expressions, and thus avoids artificial sequencing.

however every time the program has to interact with the external world, even for printing the result,, sequencing is necessary. That´s why  the compilers and interpreters exist!!!.   

If the program has to interact many times with the external world in a given sequence, the compiler can not guess such sequence if you don't write it explicitly.

 

The perfect declarative  haskell program has no main, no IO monad and no executable.