My approach works not by making valid predictions but by allowing me to correct false ones.
or simply: