[brookst]

I suppose that’s fair, but I’d give credit for a 2.3% improvement in the test environment. For all we know it may be a net loss in other environments due to quirks (probably not, admittedly).