The NumberNear function currently just uses a fixed offset to test accuracy of floating point results, but this practice can produce inconsistent results because of the variable precision of IEEE representations. The std namespace provides floating point measurement functions that can be used to test single-bit precision changes to IEEE numbers. In particular std::nexttoward can be used to try to nudge 2 numbers towards each other, judging them equal if it takes less than some predetermined (kMaxULP?) number of nudges for them to match...