Perhaps you are thinking about the conditional number of a matrix or more simple for f(x) a function with derivative f' the inverse g(x) has derivative g'(x)= 1/f'(g(x)), so using an encoding with a function with little variability means that recovering the original from the encoded value is not robust, any small error is amplied. The condition number of a matrix is a way to measure the difficulty of solving a linear problem. For a non linear problem one usually apply the above using a linear approximation near a point, so you have the jacobian matrix and the condition number of the jacobian matrix is a good measure of the difficulty of recovering a value encoded when there are errors, obviously one way to enhance the precision is to use redundancy or error recovery techniques.