EEG binary classification / motor imagery performance

What sort of classification accuracy have you been getting for motor imagery or any 2-class classification problem with Ganglion or Cyton?

Any tips or tricks for improving performance?

My performance is not great with my Ganglion at the moment. It struck me, however, that when i get the 2 class motor imagery model working well enough that I can just use that for any 2 class BCI use case (I'll just have to think "right" or "left" even if it's being interpreted as "on", "off", "up", "down", or whatever the actual class labels are).

Also, is it just me or is the so-called 2 class motor imagery (or any similar problem) really 3 classes? Because aren't the classes implicitly "right", "left" and "neither"?