Week 1: Setting Up a Paper Trading Bot
Setup notes and design lessons from the first stage of our paper trading experiment. This is a lessons post, not a results report.
Simulated paper trading, not live results; not financial advice. No real money was traded. This post makes no weekly performance claims. A later review of the ledger and decision logs we kept found that the simulated run (Feb 26 to Apr 8, 2026) ended net negative; the figures and their caveats are in the retired-project case study. The bot has since been retired; see why we retired it.
What the Bot Was Designed to Do
The bot was an experimental paper trading system for crypto and prediction markets. In paper mode it simulates trades against a $10,000 notional balance and never places real orders.
Its design put a critic agent between a trade signal and execution, so every proposed trade could be reviewed and vetoed against risk parameters. Circuit breakers were meant to pause trading after drawdown thresholds, and Telegram alerts were meant to keep a human informed.
Lessons From Setting It Up
- Start in simulation. Running with no real capital lets you test plumbing, alerts, and safety gates before money is involved.
- Put safety gates in before strategies. A veto layer and circuit breakers are easier to design first than to bolt on after something goes wrong.
- Keep a run record. Our ledger and logs were incomplete in places (see the data caveats in the case study), which is why this post does not report weekly figures. If you run your own paper test, log every decision from day one.
Where This Went
The bot ran as simulated paper trading and was later retired so we could focus on governed AI operations. The results, computed from the ledger and logs, are in the retired-project case study; the reasons for stopping are in the retirement post. For the design itself, see the architecture write-up and our risk rules post.
Related Reading
The trading bot has been retired. These posts cover the design and what we concluded.