The Mental Commands Detection Suite has a leveling system built in so it gets progressively harder as you add more actions. The best way to train cognitiv is to start with recording a neutral state (relaxed, act naturally). And then train 1 action. You need to maintain the thought consistently over the 8 seconds while training is in progress. Once you accept the training, the cube is live and you can try to see if you are getting results.
You should master 1 action and have good control before adding a 2nd action and so on...
Key things about Mental Commands training are (1) relax, so the algorithms don't have to cope with variable muscle signals from straining and (2) keep the visualisations as consistent and distinct as possible. Novices tend to strain hard for all actions, which is a challenge for the signal extraction systems, and they also tend to think "DO SOMETHING lift" or "DO SOMETHING push", which makes the signatures difficult to distinguish. More experienced users tend to be more comfortable and confident and can invoke the visualisations almost unconsciously, like moving an arm or leg. I tend to visualise a ball of fluid in the centre of my head which is squeezed forward through imaginary bars behind my eyes to make a PUSH, floats out through the top of my head for LIFT etc.
Training for an action can take up to 8 hours of training before the system recognizes your specific pattern.