Freeze the rules first

Specify entries, exits, size and data frequency before calculating results. Repeated changes after seeing performance can fit historical noise. Preserve a revision log that separates original hypotheses from later adjustments.

Model execution limitations

Costs and trading hours matter. If a bar touches both a stop and a target, coarse data may not show the order. Do not automatically select the favourable execution; document uncertainty or use finer data.

Keep an independent sample

Reserve data that was not used for tuning, then observe forward behaviour. Compare drawdowns, trade counts and dependence on particular periods. Historical evidence does not establish future effectiveness.

Check concentration of results

Compare a simulation's total with a version excluding its few best outcomes as a sensitivity exercise. A large difference indicates reliance on a small number of events. Keep the original record; the purpose is to understand concentration, not to rewrite the actual results.

Source and reading notes

This is an independently written educational overview based on the ATFX reading linked below. Examples are illustrative, not live quotes or forecasts. It does not reproduce the original article or imply endorsement by its author.