Intraday vs long
The SDK has two backtest models. You don't pick one with a flag — the data resolution decides:
| Resolution | Model | Capital | Overnight |
|---|---|---|---|
minute | Intraday | Resets to the same amount every day | Flat — all positions closed at the session close |
daily | Long | Funded once | Held; compounded across the whole period |
Intraday (minute)
This models how intraday and medium-frequency trading actually runs. Each trading day is an independent trial:
- It starts from the same fixed capital — no compounding from yesterday's gains.
- Your strategy is rebuilt fresh each day, so indicators start cold (the overnight gap is a hard reset, not a rolling window).
- Every open position is liquidated at the session close — you are never holding overnight. Resting orders are cancelled.
Because every day is the same size, performance is a constant-size book: the equity curve is the running sum of each day's profit and loss over the fixed base, and the headline return is total P&L / starting capital. Risk-adjusted return uses the daily P&L series: mean(daily) / std(daily) × √252. You also get the full daily distribution — win rate, best/worst day, mean and standard deviation.
Long (daily)
A single funded book over the whole period — positions are held and the account compounds, the classic buy-and-hold style. At the end of the period all positions are liquidated once so the final profit and loss is realized.
Intraday is exact when you run locally. Cloud submit of a minute strategy currently evaluates it with the long-style engine — full cloud intraday parity is in progress. For honest intraday numbers today, use local run.
Next: Fills.