I do believe that's quite intentional, due to the inexactness of floating-point values, it's not "right" to do that. There should be a `partial_min` function though, IMO, which has weaker guarantees.