The insider vs. lucky forecaster problem is actually tractable statistically. In equity markets, informed trading detection uses a combo of signals: order size relative to market depth, timing proximity to the resolution event, and cross-market correlation (same entity appearing in related contracts).
For onchain prediction markets specifically, the pseudonymous addresses are actually more traceable than people assume - you can cluster wallets by funding source patterns and behavioral timing even when fresh addresses are used. Sophisticated actors know this and route through mixers, but most don't bother.
The deeper problem PollardsRho hints at: if known insiders crowd out calibrated forecasters (who rationally won't participate when they expect to be adversely selected against), you get a market that's accurate but thin and fragile. That's the classic adverse selection death spiral prediction market designers have been trying to solve. Polymarket's bet-sizing dynamics actually mitigate this somewhat - insiders can't take all the liquidity without moving price against themselves.
For onchain prediction markets specifically, the pseudonymous addresses are actually more traceable than people assume - you can cluster wallets by funding source patterns and behavioral timing even when fresh addresses are used. Sophisticated actors know this and route through mixers, but most don't bother.
The deeper problem PollardsRho hints at: if known insiders crowd out calibrated forecasters (who rationally won't participate when they expect to be adversely selected against), you get a market that's accurate but thin and fragile. That's the classic adverse selection death spiral prediction market designers have been trying to solve. Polymarket's bet-sizing dynamics actually mitigate this somewhat - insiders can't take all the liquidity without moving price against themselves.