Likewise. I think for me there's quite a wide band of colours in the middle that I consider to be "neither/either", so I'm basically just picking a random answer for those.

A modified version of the test that finds two boundaries (green/neither/blue) could be interesting.

Or maybe it just needs to take more samples, in a more random order.