Design a credible backtest
Goal
Build a reproducible experiment that makes its assumptions visible.
Lesson
Write down the date range, market universe, timeframe, exchange, fee assumptions, order model, startup history, stake model, and software version. Establish a simple baseline. Include multiple market regimes where data permits, and inspect results by instrument and time period rather than relying on one aggregate score.
Historical tests approximate execution. They may assume fills that would not occur, omit costs, or use a current pairlist that was not available historically. Treat each assumption as part of the result.
Reserve a later period before tuning begins. Freeze the strategy before evaluating that period. If the evaluation changes your design, begin a new experiment and reserve new data.
Practice
Take a previous backtest and list the information needed for another researcher to reproduce it. Identify any missing assumptions.