Quants at Citadel and Jane Street are some of the smartest people in finance.
They also go home at 6pm.
My Grok Bot doesn't. It watches the market 24/7. No commute, no sleep, no weekend, no opinion about last week.
I gave it $2,324 and one task: make it 2x in 7 days.
It needed 9 hours.
Hour 0 - $2,324
Hour 9 - $4,648. 2x. target done, 6 days and 15 hours early
Hour 16 - $7,100
Hour 24 - $9,648
4.15x in a single day.
Here's the part that matters. During US desk hours, when the quants were at their screens, it made $1,910.
Everything else, $5,414, came while they were asleep.
The 3am wicks. The Asia open. The Sunday-night liquidations nobody is awake to catch. That's where most of the money sits, and that's exactly when a human desk is offline.
96 trades. 71 green. 25 red. The worst one cost $186, at 4:12am, and it was closed before anyone in New York woke up.
Citadel runs 330 pods to cover the market. My bot covers it alone, around the clock, and never asks for a bonus.
The whole build is in the post below ↓
PASTE THIS INTO CLAUDE CODE AND REA OPENS IT
This one-image prompt points it at one question: what did this update change?
What the prompt makes your agent do:
→ ask you 4 questions, then set REA up with a dry run first
→ analyze the old and new version once each
→ compare them and label every change OBSERVED, INFERRED or UNKNOWN
→ list the gaps instead of guessing, and run nothing
The tool calls are REA's own, including compare_application_versions
Works on software you own or may examine. Unknown file, use a VM.
REA is not a sandbox
Save the image, copy the prompt into an empty folder