Your first strategy
1. Create it
Click Add strategy in the sidebar. Give it a name ("RSI Scalper SPY"). If you already have code or notes, drop the .py or .md files in the same dialog. Then Create & Start Chat.
2. Describe the idea
In the strategy's chat, write the idea as you would tell it to a person:
Trade SPY. Buy when RSI(14) on 15-minute bars falls under 30. Sell when it goes over 55, or at the close. Use 20% of the account per trade.
The AI writes the strategy, and checks that it runs and that the data is there.
3. Plan the walk-forward
Before the first backtest, decide how the strategy will be judged. Open Walk-Forward at the top of the chat and create a validation plan, or say yes when the AI proposes one: In-sample months, Out-of-sample months, Years of history. The AI suggests 12 and 3.
The plan cuts the history into folds. In each fold you tune on the in-sample months and are judged on the out-of-sample months after them. See What is a walk-forward.
Why first: a backtest over the whole history lets you see the months the plan wants unseen. A fold seen that way is marked in the report, and its verdict is worth less.
4. Backtest on the in-sample months
Now run backtests, as many as you need, on the first fold's in-sample months. Say it in the chat, or use Backtest → Run Backtest.
The Backtest panel shows the total P&L, the win rate, the number of trades and the equity curve. Open Data to see every trade on the price chart. Ask the AI: "Why does it lose on Mondays?", "Add a stop at 1%". The runs are numbered, so you can compare them. See Backtests.
Nothing here is a verdict: these are the months you tuned on.
5. One try out-of-sample
Pick the best run as the fold's winner and press Go OOS. It is tested once on the months it never saw, and the fold closes. Then the next fold.
6. Read the verdict
The Edge Report joins the out-of-sample months of every closed fold and says what they are worth. That, not a backtest, is what to trust before going live.