Congrats To Polymarket, But I Still Think They Were Mispriced

Argues Polymarket's 60% Trump (vs Metaculus's 50%) was mispriced by ~10 cents, and that Trump actually winning barely vindicates it -- via the Bayesian coin argument (a single heads shifts you only ~2% between a fair-coin and a 60/40 hypothesis; even five heads leaves a fair coin at ~29%), so one election should barely move your prior over which forecaster is better. Scott's pre-election reasons for trusting non-money forecasters: they've beaten real-money markets historically; markets give weird results (structural costs, dumb money, banned smart money); and the specific Polymarket/Metaculus divergence was driven by ONE whale ('Theo', $30-75M on Trump) breaking the markets' usual synchrony. Handles objections (Theo was smart/confident -- but so was the guy who lost $5M; prediction markets assume roughly even money and enough minnows to digest a whale, which failed here given crypto/VPN access barriers and Kelly-insane bet sizes). Concludes prediction markets are still among our best truth sources, just not infallible.

Quality 74 · Strong. Strong, upper end. A sharp, rigorous forecasting-epistemics piece -- the don't-update-much-on-one-event Bayesian core and the one-whale-broke-the-market analysis are both clarifying and well-argued. Held just below Excellent as a focused, topical treatment.

Claude’s paradigm shift 52 · Moderate. Moderate. The Bayesian-updating point is standard; the fresh contribution is the specific Polymarket-vs-Metaculus application and the whale-mechanics analysis.

Real-world impact 2 · Minor. Minor. A notable intervention in the prediction-market-accuracy discourse; no material reach beyond it. Within-niche.