Blog

Tech blog

Part 8 of 8 · Jev in trading, tested

I expected Jev to remember 2025. It doesn't know Trump was inaugurated

10 Oct 2026 · truongxx · 8 min read

I assumed a model trained on the internet would know how 2025's biggest markets resolved, so a backtest on old Polymarket data would be leaky. Jev scored the same on 17,000 old markets as on unseen ones.

#jev #polymarket #prediction markets #backtest #data leakage #machine learning

Tech blog

Part 7 of 8 · Jev in trading, tested

Can Jev beat Polymarket? I asked it about 26,174 resolved markets

10 Oct 2026 · truongxx · 12 min read

I asked Jev "will this resolve Yes?" for 26,174 resolved Polymarket markets. On the 9,138 resolved after its release it scored AUC 0.615. The market price a day earlier scored 0.844. Code included.

#jev #polymarket #prediction markets #ai trading #backtest #python