I suppose rounding is solvable too. Depending of how value is used at the end it would be possible to deduce how much precision is needed to guarantee no cumulative error.