Running a PCA on data from 64 node (wet contact, shaved head, lab environment), shows that you basically get one channel of data, there’s virtually zero useful information other than the main wave; you can tell whether the subject is asleep, awake or having a fit with high degree of accuracy, you can accurately classify whether the awake subject is relaxed vs aroused but that’s about it. The claims made for these systems are way beyond anything supported by the data. Would love to be proved wrong by standard, verifiable sourced data and a Jupyter notebook, but I won’t hold my breath.

I can't speak for this one, I gave the OCZ NIA a spin when it was released probably more than a decade ago now, and what I found was that mostly it picked up the electrical interference from all the low quality DC transformer plugpacks that powered all my monitors and chargers etc.

I'm not sure that "machine learning driven signal processing" is going to be enough to screen out the dirty EM environment you're gonna find in most peoples houses and offices.

I do wish these guys best of luck though, it's an interesting field.