That still does not explain how you get several 300% and 1% values in the same category like attack for example, because you cannot train just one value in a sub category.