Jev 驱动的交易 bot 几天赚 12 万美元的架构拆解
用 Jev 对同一市场状态并行评估 regime、方向、毒性流、流动性等六个问题，policy 转成下单动作，作者称每百万输入 token 0.042 美元、一万次决策约 42 美分。
a Jev-powered trading bot made $120k in a few days the AI behind its decisions can’t write a single sentence people spend years preparing for a seat at jane street. citadel. hudson river trading learning probability, studying market structure, practicing how to think under pressure then you look inside a trading system and find something you can start building yourself market data comes in. the state gets calculated. a decision gets made. an order goes out nobody touching a mouse FPGA engineers push that loop into hardware. Jev gives software builders a decision engine for an on-chain version regime. direction. toxic flow. liquidity. quote quality. inventory pressure six questions evaluated in parallel against the same market state code does the arithmetic. Jev supplies the judgments. policy turns them into actions. hard limits control what reaches the exchange same questions every update. same checks before every order. every result logged for the next cycle at $0.042 per million input tokens, 10,000 decisions using 1,000 input tokens each cost 42 cents in Jev inference forty-two cents for that layer of the operation the full blueprint is already public state engine. parallel questions. pricing logic. execution loop. fallback rules. kill switch you can start assembling it without a finance degree, a recruiter or somebody giving you permission to sit at their desk most people discover AI trading and open another chat window look one layer deeper and you start building the machinery save the article below and start with the first loop the career path everyone showed you ended at a trading desk this one starts with building your own
这条是引用着另一条写的：「用 Jev 从零搭高频交易 bot：每区块一次买卖决策」