Lower confidence was to account for the fact that some people would read it and decline from voting either way; therefore P>50% of upvote would not imply P<50% of downvote. In hindsight it's a confusing scheme to parse into bets. Everyone who read (e.g.) the first prediction and didn't vote would count in the 80% who didn't downvote it for being self-conscious. The first five percentages were, in my mind, predictions concerning the distribution of the actions of those who read the comment, where 'didn't vote' also counts toward the union. But regarding betting, there's no way to get the data of how many actually read it.