Less frequent context truncation, too, leading to better scores?