Research tool
SmartMoney
A local tool that tests market strategies and demands proof before a single order.
In research
5 agent roles
1 sealed year of data
0 real orders before proof
A local market research tool, run like a production project: five agent roles (auditor, analyst, risk, operator, review), shared registers and a written protocol. Each hypothesis is tested on history, then, if it holds, on a sealed year of data; no real order goes out before proof.
Stack: Python, Claude Code, Windows scheduled tasks, Hyperliquid API.
The method
- A protocol written before the trials: every trial counted, a placebo and benchmarks to compare against, the fees of each trade deducted.
- A sealed year of data, opened only for the trials that pass the first test.
- Every refuted hypothesis is recorded, just like a success.
The roles
- Five Claude Code skills: auditor, analyst, risk, operator and review, with shared registers (pitfalls, indicators, log, reviews).
- A review at the end of every session and a monthly review.
Production
- Hourly data collection and daily simulation as scheduled tasks, with no language model.
- Stages approved one at a time: test network, simulation, and only then real capital.