JockeyBench Benchmark for LLM Agent Decision-Making
October 13, 2026
JockeyBench evaluates LLM agents on repeated event forecasting using Japanese horse racing data. The study found that while some configurations show revision gains, all tested configurations resulted in negative net utility in betting returns.
HOW THIS AFFECTS YOU
●
researcherYou can use this protocol to test how agents handle time-gated evidence and noisy forecasting tasks.