Static typing in a programming language is a mechanism for making certain types of non-terminating computations impossible to express in that programming language. The understanding "non-terminating" here includes errors as well (the idea of bottom ⊥ that you'll find in some earlier texts on programming languages semantics, such as Allen's Anatomy of Lisp). Non-terminating computations are equivalent to paradoxes in formal logic (see Curry-Howard correspondence), and the first type system was developed to deal with Curry's paradox (http://en.wikipedia.org/wiki/Curry%27s_paradox) in the first programming language (lambda calculus). Some type systems prevent more types of non-terminating computations, others less. But saying that run-time type assertions are equivalent to a typed programming language, or that types are like contracts or other things just leads to confusion.
As an experiment last year I started teaching myself machine learning by implementing the common algorithms in Clojure and Scala simultaneously. I expected Clojure to win out and worked a lot harder in the beginning but after a while I was forced to concede that my Scala code was cleaner, faster, and less buggy, partly because I felt comfortable changing things because I knew the compiler would help me.
You could only know the types of the input if the compiler was smart enough and define constraints on the types but as far as I know it's not possible in Clojure.
I don't get it.