TL;DR. JustinHarris.AI, the Las Vegas AI Consultant, built Odr, an autonomous trading system that refuses to touch real money until it earns the right. Every strategy runs in shadow mode against a paper account for 60 trading days, every order clears four code-enforced risk checks, and capital only turns on when a human says go live.
What an autonomous trading system has to get right before it gets to be fast
Most trading bots are confident long before they are correct. They run live on a backtest that looked great and a risk policy that lives in a README, and the first bad day takes a real chunk out of a real account before anything stops it. We learned this the hard way: on June 9th, an earlier hand-rolled version of this system lost 7.4% intraday with no circuit breaker, because the four risk rules it claimed to follow were written in documentation and never enforced in code. That single day is why the whole thing was rebuilt around one rule. An autonomous trading system does not get to be fast until it has proven it can be safe. Odr is the answer to that lesson: every strategy earns the right to touch money, and a human is the only switch that turns money on.
How shadow mode lets a strategy prove itself with nothing at risk
Shadow mode is the load-bearing idea. When a new strategy goes in, it does not start trading real capital and it does not get judged on a backtest. It runs against a live paper account, scoring real trades on real prices, but its results are tracked separately and it cannot deploy a single dollar. It has to survive the market as the market actually moves, not as a backtest replays it. Only after a long, boring proving period does it become eligible to graduate, and graduating still does not mean money. It means the system has earned the right to ask. Here is what a strategy has to clear:
- Beat the public benchmark ETF by a Sharpe margin of at least 0.3.
- Hold maximum drawdown to 8% or lower across the whole window.
- Place at least 30 trades, so the result is not luck on a handful.
- Run at least 60 trading days in shadow before it is even considered.
- Log zero risk-invariant violations, not one.
- Earn an operator conviction score of 80 out of 100 or higher.
The four risk rules that live in code, not in a document
The biggest lesson from the 7.4% day was that a risk rule you only wrote down is not a risk rule at all. So in this autonomous trading system, four hard limits are enforced as actual code that every order has to pass before it can reach the broker. No order may risk more than 2% of the account if its stop is hit. No single name may grow past 25% of the account. The account may never spend its way below a 10% cash floor. And if the account is down 3% on the day, a breaker trips and blocks any new buying, while still always allowing the system to sell and reduce risk. These checks are a single, self-tested function with no side effects, deliberately built so it can never be quietly skipped. There is also a manual pause switch a human can flip to freeze all buying instantly. The point is simple: the rules that protect the money are not a promise in a file, they are a gate in the path.
Where the signal comes from, and why it is legal and public
Odr runs two strategies. One follows price momentum. The other follows the stock trades of members of Congress, which are required by law to be disclosed publicly under the STOCK Act. That data is not a secret feed and it is not bought from anyone; it is scraped directly from the U.S. government disclosure sites for zero dollars, the same source that paid apps and subscription services quietly resell. The edge here is not speed. By law these disclosures can lag the actual trade by up to 45 days, so no one is front-running anything. The edge is that most retail investors never act on the signal at all, and that the system ranks each lawmaker by their own historical track record so a disclosed buy from a consistently strong filer carries more conviction than a noisy one. It is a slow, legal, public edge, captured with discipline.
Why we built it the hard way on purpose
It would have been faster to keep hand-rolling. Instead, after the bad day, we ripped roughly 76 kilobytes of our own custom trading code and adopted freqtrade, the most widely deployed open-source retail trading framework on the internet, with a battle-tested risk module the homemade version only pretended to have. We kept the two things that were actually ours: the congressional-trade signal feed and the reliability rails that alert a human the moment anything fails. That is the same judgment we bring to client work. Do not reinvent the engine; adopt the proven one, and spend your effort on the part that is genuinely your edge. An autonomous trading system is the highest-stakes version of that discipline, because here the cost of skipping a step is measured directly in dollars.
The outcome
Odr is held in shadow mode by design. It is a personal system, paper-only, and it will not move real capital until it has cleared every condition and a human has personally said go live. That restraint is the whole point. The same instinct that makes us refuse to let a bot touch money on a hunch is the instinct we point at every system we build: prove it before you trust it, enforce the guardrails in code, and keep a person on the switch that matters.
Related work
- The AI marketing system we run our own business on
- Valhalla: the operating system behind the work
- AI Managed Services
- See all of our work
Want a system built with this kind of discipline pointed at your business? Get your free AI Audit and we will show you exactly where to start. Get your free AI Audit.