For the same reason it was honestly really hard for me to do that test. It was like those personality surveys where I don't like any of the options, but I have to pick one
I think it mostly comes down to whether or not you call certain colours a "greeny blue" or a "bluey green" I'm much more inclined to say the former for the colours on between the two
I think that contrast is a big part of this that isn't really controlled for. During the test, the colors took up my whole screen, so against the black bezel of my phone in my dimly lit room they seemed to look more blue. But at the end when it says "For you, turquoise [color swatch] is blue", that color swatch was against a white background, and in that context it looked more green to me.
This is cool, however I don't like that the result is a fixed value. I don't think a person could take this test and reliably get the same result. This would be a good situation to use a logistic regression.