Eight targets on a ring, the same reach the monkey makes. Four directions of cursor movement have been mapped onto combinations of six keys — and you are not told which. Find the mapping, then hit the targets.
The decoder is re-fitted every session, so nothing you learn carries over — which is the point. A little noise is added to the cursor, as there would be in a real readout.
A brain–computer interface has to write a program for each action it will call: left, right, up, down, click. None of those programs exist in advance. Every session begins by fitting a fresh map from recorded activity to intended movement, and that map is arbitrary — it depends on which units were picked up that day and how the electrodes settled.
So the user starts each session with a control scheme they have never used, and the only way to learn it is to try to move and watch what happens. That is what this task reproduces. The keys stand in for neural features, the hidden assignment stands in for the fitted decoder, and the ring of eight targets is the center-out reach from Fig. 2. The diagonals are the interesting case: they are not their own command, so you have to hold two directions at once and discover that their combination does something neither does alone.
Watch the acquisition times across blocks. The mapping never changes within a session, so any improvement is yours — and it is the same curve a subject climbs before a real interface starts to feel usable.