Why standard trading tools fail on prediction markets

Researchers released a complete archive of Polymarket trades (2022–2026) with ground-truth buy/sell direction from blockchain settlement, then tested standard microstructure classifiers like the tick rule. These tools scored near 50%—random guessing—because prediction markets violate the mean-reversion patterns classical tools assume. The finding matters: inferred trade measures used in cost analysis and market quality assessment are systematically biased, and only real ground-truth metrics predict whether markets are calibrated correctly.