"It seems that the point is to introduce their special-sauce black box, with an argument to authority about its methodology. I think the correlations you ask for are where the problems will lie, in that there is a value judgement that is being hidden. If I can put myself out on a limb here, I'd say that that measurement is going to be fundamentally unscientific."

Also, this seems like an important question, if you claim to be based in science:

"How can we falsify your claims? That is, what is a test we can perform that if it went a certain way, would show that your claims are false?"

It occurs to me that human color perception is another "subjective but scientific" field. You might want to research the experiments they performed for e.g. color matching in varying illumination levels, then come up with an experiment with a similar structure.

Specifically, if EffectCheck is accurate, then it should correlate with how 100 average people would classify e.g. a sample of 1,000 twitter messages.