FAQ

Three usual causes: symbol spelling (Hyperliquid core markets use bare names while HIP-3 keeps its prefix, such as xyz:AAPL); the freq is not declared in extra_freqs; window × frequency exceeds the lookback_days warm-up.

Backtest uses the frozen HISTORICAL_BAR_EXECUTION model: the current real completed-bar close logical BBO plus configured fees/slippage after the strategy decision. It never uses an unseen next open. Paper never uses that bar as a quote; submit-time fresh OnePort BBO determines its isolated simulated execution. Neither mode claims venue queue or fill equivalence.

on_funding is an optional legacy hook: Backtest and Paper call it once per verified UPERP settlement with non-zero exposure, up to 256 legacy event callbacks per boundary. New strategies should process ctx.funding_events once from handle_bar; omitting on_funding adds no strategy callback. Live does not estimate settlements or emit this hook.

Runs freeze the project version and dataset references so the inputs can be identified and reviewed. Exact repeatability also depends on the recorded engine and runtime environment.

The strategy sandbox disables network access and uses a read-only root filesystem. Supported research data access is exposed through the ctx data APIs.