Backtest validation checklist

Most strategies that look good are wrong in a way you can find. This is what I check before believing a result.

Data and leakage

• No look-ahead: every feature has to be computable at the moment of the decision, including higher-timeframe bars and indicator warm-up.
• Roll handling: futures continuous series must not create fake gaps or fake trades at the roll.
• Survivorship: equity universes must include delisted names.

Robustness

• Split-half replication: an edge found in the first half has to show up again in the second half.
• Concentration: remove the top trades or days. If the edge disappears, it was luck.
• Random-period sampling: many random windows instead of one flattering one.
• Costs: realistic commission and slippage per contract. Many edges are smaller than the spread.

Sizing and risk

• Measure drawdown the way the account is judged: intraday and marked to market.
• Size from bootstrap-resampled paths, not the single historical path.

Before going live

• A regression suite that replays the strategy trade for trade after every code change.