The spec says that the default output validator tries to compare floats (if a floating point tolerance is given) but does not define what a float is/which representation will be parsed as a float.