Polymarket not only doesn't charge a commission for market makers (the people who put out offers), but actually gives them a percent of all fees collected from market takers to help ensure and incentivize market liquidity. So if you could create a model that offered even slightly worse than market accuracy, you'd have an infinite money machine. It does not seem to exist.

And I think the reason for that is because a lot of these factors are unquantifiable, subjective, and non-fixed. For instance determining the winning chances for a human in a chess position is surprisingly complex. There's even a huge chunk of profit to be made for somebody that could create such a model outside of printing money in prediction markets - it'd be an invaluable tool for players to use during opening preparation since positions where winning chances don't correlate with computer eval are sort of the money-shot in human prep that's often motivated by a desire to avoid computer prep. And that factor is just 1 amongst many that can weigh in on an expert bettor's opinion.