This is the part the coverage skips, and it’s measurable: the “probability” is usually set by almost nobody.
I pulled the 1,000 most recently created Manifold markets: median number of bettors is 5, 8.1% have zero bets, 79% have fewer than 10. A five-person room reported to three decimal places reads as consensus but is closer to your “pure guesses” — the decimal point is doing all the work.
It also fragments: the 500 “currently open” Polymarket markets are actually 155 questions, because one event is split into one binary leg per candidate (median 2 legs, max 54). So a lot of the apparent breadth is one event wearing many hats, and the legs disagree with each other often enough that the aggregation isn’t free.
This is the part the coverage skips, and it’s measurable: the “probability” is usually set by almost nobody.
I pulled the 1,000 most recently created Manifold markets: median number of bettors is 5, 8.1% have zero bets, 79% have fewer than 10. A five-person room reported to three decimal places reads as consensus but is closer to your “pure guesses” — the decimal point is doing all the work.
It also fragments: the 500 “currently open” Polymarket markets are actually 155 questions, because one event is split into one binary leg per candidate (median 2 legs, max 54). So a lot of the apparent breadth is one event wearing many hats, and the legs disagree with each other often enough that the aggregation isn’t free.