Edison：用 Jev 做交易入场闸门并审计 AI 是否真有用
黑客松项目 Edison 用 Jev 做 AI 入场过滤，回放同一策略对比有无闸门，结果回撤减半但年化收益从 21% 掉到 1.9%，判定 AI 改进未通过。
built edison for the @bitget_ai base camp hackathon. when an ai filter reduces drawdown, did it actually make better decisions or did it simply trade less? edison replays the same strategy with and without an ai entry gate, then audits the winners it missed, the losses it avoided, exposure, costs and the evidence behind the result. in our frozen study, jev cut max drawdown from 25.19% to 12.33%, but annualized return fell from 21.17% to 1.93% while exposure dropped to 22.33%. the ai improvement gate did not pass. that negative result is the point. see whether ai actually helped. demo: http://edison-fawn.vercel.app github: http://github.com/emmagh1/edison #BitgetHackathon
这条是引用着另一条写的：x.com/bitget_ai/status/2100519318824055159