A sample size of 3 happy users, even with review bias, is still way too small to refute a supposed defect rate of 17%.

And I bet that comments on HN are going to be less affected by negative review bias than actual reviews.