SWE-bench for trading agents. Dolores Research studies how AI agents behave when given a wallet and told to trade. Their benchmark replays thirteen recorded stretches of the BTC perpetuals market, from the COVID crash to the ETF approval, and runs each agent through them under identical rules, with every decision sealed so anyone can reproduce the result. The whole thing is open source under Apache-2.0, live on Virtuals Protocol.(Virtuals Protocol)