I've been working with self improvement harness a little bit and one thing i've come to conclusion is harness task fit. The learning can be significantly improved if we understand the behaviour of task and how it should be learned. I'm pretty sure a general solution will definitely exist which will do fine, but we are yet to see one.