Let $x \in [0, 1]$ be irrational and let $$x = \cfrac1{a_1 + \cfrac1{a_2 + \cfrac1{a_3 + \frac1\ddots}}}$$ be the regular continued fraction expansion of $x$. If you truncate the continued fraction expansion at $n$ you get the $n$th convergent $A_n/B_n$ to $x$.
Let $R_x(n) = x - A_n/B_n$. I know that $$R_x(n) \leq \frac1{B_n B_{n+1}}$$ For $\phi = \frac{\sqrt{5} - 1}{2}$ we have that $R_\phi(n) = O(\phi^{2n})$ and this holds for any $x$ so it is the worst-case approximation to the $R_x$.
Are there any such approximation for the "average-case"?
EDIT: I don't have a rigorous definition of "average-case". But I would be interested in approximations that work for almost all $x$, for example.
Lochs' Theorem implies, that
$$R(n) = \Theta \left(\exp\left(- \frac{\pi^2 n}{6 \ln 3}\right)\right)$$
for almost all $x \in \mathbb{R}$.