Seems to me that number 1 is the right one to test with. We are sure of the problems in 2 and 3, but 1 is a maybe problem of false inputs.
I know it would suck to invest the money and have it not work but that's part of r&d. Maybe you could do a small kickstarter-type thing where we sponsor the final step by sending you $10-20 with the idea of getting a small discount when we buy the final product. Just throwing it out there
