Strategy building isn't a one-shot. It's a loop. The loop has four steps and most people skip two of them.
1. Design
Articulate the edge. "Buy dips in an uptrend" isn't an edge — it's a slogan. "On the 5-minute chart, when price is above the 50-EMA and RSI-14 drops below 30, buy one contract with a 10-tick stop and a 20-tick target" is an edge. It can be wrong. It can also be tested.
This is where the AI Strategist earns its keep. Talk to it like a collaborator, not a vending machine. It asks clarifying questions because the answers change the code.
2. Backtest
With real slippage and commissions, on enough history to cover multiple regimes. Read the Sortino, not just the total return. Read the profit factor — if gross losses are as big as gross wins, you're renting variance, not harvesting edge. Read the OOS drift — if it collapses on hold-out data, the strategy only works on data you already saw.
3. Forward-test
Paper-trade it live. For two weeks at minimum. This is the step most people skip, and it's the step that catches the things history can't show:
- Your backtest's slippage model was optimistic.
- The data feed has a 300ms lag you didn't account for.
- The "5-minute close" your indicator uses doesn't quite line up with your broker's 5-minute close.
- The strategy times out overnight on a corner case you never tested.
Runbook's forward-test uses the same engine that runs live — same slippage model, same fill logic, same error handling. A strategy that survives forward-test is a strategy that will survive real money.
4. Iterate
Now you have two data sets: the backtest and the forward-test. If they agree, you have a strategy. If they disagree, you have a puzzle. Read the disagreement:
- Backtest good, forward good → ship it.
- Backtest good, forward bad → overfit, bad slippage model, or data-feed quirk. Investigate before touching the logic.
- Backtest bad, forward good → suspicious. Probably a regime shift; check if the market changed under you.
- Both bad → the edge isn't real. Go back to step 1 with a different idea.
The trader who does this loop ten times with ten ideas finds one strategy that ships. The trader who does step 1 ten times without ever running steps 2-4 finds nothing, forever.