One way to reason about this: instead of the score being one person’s numerical value it is instead the percentage of people who make a binary choice.

Put another way, you can calculate…

  1 rating between 0 to N

by sampling…

  N ratings of 0 or 1

So donuts-for-breakfast has a score of 5 not because it is quantifiably 5/100ths awful, but because only 1 in 20 people choose it.