One of the reasons I would be happier if there were more emphasis in real math/geometry algorithms on "range" processing, i.e. treating each "number" as a scaled "range" instead and figuring out how to combine these ranges in every step to get some stable solution. Most often I see programmers just bashing analytic formulas like they were computing something precisely and then being surprised that results are way off. Imagine computing curve intersections by a combination of analytical and iterative method and then figuring out if such an intersection is unique or if it is some existing endpoint...