Inverse middle-out

[candalfigomoro]

This is how the current MiddleOut method works:

It anchors the base predictions in a middle level. The levels above the base predictions use the Bottom-Up approach, while the levels below use a Top-Down.

Would it be possible to add an "inverse middle-out" where levels above the base predictions use the Top-Down approach and levels below use a Bottom-Up?

[candalfigomoro]

Thinking about it more, it's probably not really feasible :)