Grok, Opus 5.5, Jev and Dots tested on a four-agent trading desk
Four agents split scanning, stress testing, sandbox ordering and settlement, and Jev paper-traded first and blocked a position that would have lost $12.4K.
A CAMBRIDGE STUDENT LEFT GROK, OPUS 5.5, JEV AND DOTS ALONE ON A TRADING DESK FOR 30 SECONDS. THEY MADE $21,019 the agent that earned the most never placed a single real trade he built it between lectures. 1.2 million ticks hit the desk every second, 92% of them are noise, and no trade goes live until all four agents agree on it 01 / scan grokbot watches 14,280 markets at once and throws out 92% of the ticks before the other agents ever see them 02 / stress opus 5.5 runs 10,000 scenarios on every trade before a cent is committed. drawdown capped at 1.2%, every bet sized at ¼ kelly 03 / sandbox jev places every order on paper first. it killed one position that would have cost $12.4K, and failed orders never reach the real book 04 / settle dots verifies every trade and settles only the audited ones. 32 out of 32 checks passed after one session: > $21,019 net in 30 seconds > win rate up 20.3%, sharpe 2.41 > 487 bad orders rejected > 0 leaks to production grok made $1,479, opus $2,959, dots $4,179, and jev $12,399 without placing a single real trade the full setup for hiring all four is in my article below:
This post quotes another post: x.com/aimonreal/status/2107114107333697994