Run a backtest without look-ahead bias

Practice the sequence you already learned for “Run a backtest without look-ahead bias” until the order feels automatic. Practice quantitative-research order from data-science training. A strategy is not promoted until point-in-time data, costs, and a holdout period are applied.

1. Use point-in-time features only; no restated filings after the decision time.
2. Apply transaction costs and latency as the live system would see them.
3. Walk the backtest forward; do not fit on the whole sample.
4. Reserve a holdout period that was never used for research.
5. Compare to a simple baseline.
6. Do not drop losing days after seeing the equity curve.