I'm having some rounding errors with the random tests - always a few pennies off. Could the error tolerance be relaxed, or perhaps the desired rounding scheme elucidated?

Haskell: I think there are some floating point issues. I'm experiencing the same problem as BalinKingOfMoria.

It seems more logical to use the type Rational instead of String, i.e., decompose :: Rational -> [Rational].