Strategy lifecycle

You build strategies by talking to Claude, not by writing code in this app. The web app visualizes your strategies and runs backtests; authoring happens in Claude, guided by the build-strategy skill. Here is the whole path.

1. Describe a rule to Claude

Tell Claude the trading idea in plain English — entry, exit, and how big. Keep it simple: fewer rules and fewer knobs backtest more honestly. Claude picks the trading type that fits (Swing or Day) and writes a strategy that inherits the matching base class and declares its parameters (each tunable as a named, typed knob).

2. Save it

When you're happy, ask Claude to save the strategy. It proposes a name (or you give one) and stores it on your account. Saving is what lets you run it — backtests run by reference to a stored strategy, never against pasted code.

3. Refine — every change is a new version

Ask for a change ("use a 50-day average", "add a 2% stop", "make the threshold a parameter") and Claude regenerates and saves a new version. Versions are content-addressed and linked to their parent, so the full history is preserved and any backtest is reproducible. Asking for a brand-new idea instead creates a new strategy, not a version.

4. Run a backtest

Ask Claude to run a backtest, or open the strategy in the web app and hit Run backtest. A run needs a few things: the universe (which symbols), a date range, starting cash, and a value for every declared parameter. Claude asks for whatever is missing; the web form shows the parameters as required fields. The first run is also the real proof the strategy executes — if something is wrong, fix it with Claude and run again.

5. Read the result honestly

Read the numbers in the language of the type. A Day result is an additive per-day book with a daily distribution; a Swing result is one compounding curve. Don't compare them as if they were the same thing, and don't tune the strategy just to make the backtest look good — that's overfitting, and it lies.

Next: Your first backtest.