What your strategy has to survive
Any strategy can look good on old data. Working out whether that was luck is the whole job. Five steps, in order.
Every screenshot here is the real app. We do not publish mock-ups.
No skipping ahead
A strategy has to earn each step before it gets the next one. Build, backtest, gate, paper. Live comes last, and only when you switch it on yourself.
Step 01
Build
You write the rules: when to buy, when to sell, how much, and where you get out. No code.
The risk engine reads along as you type. Break one of your own limits and it says so there and then.
Step 02
Backtest
Your rules run against real market history, stamped with where the data came from.
If the engine is down, the run fails and says so. A made-up backtest is worse than no backtest.
Step 03
The gate
Three tests, all hunting the same thing: a result that was luck. Funds use them to throw strategies out. Here they run on your side of the table.
Most fail. That is not the product breaking. That is the product working.
Step 04
Paper
Whatever survives trades on paper first. Real prices, real market hours, fake money.
Every fill is labelled simulated and scored against your own limits. You watch it work in real time, not in hindsight.
Step 05
You switch it on
Live trading runs on your own computer, with your own broker keys. Nothing goes live until you switch it on, one strategy at a time. There is no setting that removes the human.
Plainly: today the live path is an experimental single-shot Alpaca order path, never tested against the live venue. Automated execution, multi-broker support and production execution safety do not exist today. We would rather you read that here than discover it later.
What the three tests check
No black box. Three published tests, and every verdict shows its numbers.
Try enough versions of a rule and one will look brilliant by luck alone. This asks how likely it is that yours is that one.
Probability of backtest overfitting
Bailey, Borwein, López de Prado & Zhu, Journal of Computational Finance 20(4), 2017.
A backtest's headline score flatters you. It ignores how many versions you tried first. This asks what is honestly left.
Deflated Sharpe ratio
Bailey & López de Prado, Journal of Portfolio Management 40(5), 2014.
Part of the history is hidden while you build. Your rules are scored on the part they never saw.
Out-of-sample holdout
Standard practice; the split is shown on every run.
Live runs on your machine
The website does the thinking. It never holds your broker keys. Real orders come from the app on your own computer.
Claude plugs into that app. Ask about your strategies in plain English. The conversation stays on your machine, like the keys.
Both halves come with the one plan. Your computer is the only place live trading exists.
Easier to say plainly
- Not a signals service. PTQ never tells you what to trade.
- Not advice. It is software. It has no opinion on markets.
- Not a bot marketplace. You cannot rent anyone's strategy here.
- Not a broker. Your money stays at your broker, in your name.
- Not a performance promise. We publish what the gate computes, never what you might make.