Not confused. Saying that new/top adds bias - that a blind sampling would help solve.

Reference to dups provides proof that "top" stories are being missed, since one dup might have 100+ upvotes, while another will have just a few upvotes and no comment saying it's a dup; meaning to me the only difference is the volume of comments/votes between the two.

Beyond that, I've seen research that shows seeding a submission with even just a few comments/upvotes greatly increases the odds of future traction.

Point is to reduce noise to signal ratio.