Run a panel of AI forecasting bots over every open market, each given different prompts, source material, and personas, then blend their probability estimates into one number weighted by how accurate each bot has been in the past. You bet when that blended forecast disagrees with the market's price by more than normal noise. The edge is that superforecaster-style averaging beats the thin, lightly-traded crowd on under-followed Polymarket questions.
Markets on Polymarket where superforecaster ensemble disagreement engine is the natural play:
