Consistency doesn't match research.
Sometimes language features trump usability.
Hard to scale to enough participants.
Hard to change designs after v1.
