Randomly select votes that are actually recorded. Then add in metavoting that votes on the votes with random sampling. At Google's scale with a sufficiently ran

Randomly select votes that are actually recorded. Then add in metavoting that votes on the votes with random sampling. At Google's scale with a sufficiently random sampling you'd be extremely hard pressed to successfully brigade or spam the voting. Google could easily use its current fingerprinting to constrain (to an extent) multiple votes. Even knowing only a portion of the population will participate in the voting they can use a Wilson confidence interval[0] or similar to properly weight votes. Random sampling works here since you're not guaranteed one vote per user per page and the outcome in binomial, seen and downvoted or seen and not downvoted. [0] https://www.mikulskibartosz.name/wilson-score-in-python-example/ https://www.mikulskibartosz.name/wilson-score-in-python-exam...