Why Small Samples Need Modest Conclusions

A precise percentage can still describe very little evidence.

Count the observations

A win rate calculated from five trades is a description of those five outcomes. It offers limited evidence about future performance. Breaking the sample into instruments or sessions makes each subgroup smaller still.

Look beyond the win rate

Average win, average loss, costs and position size affect results. A high win rate can coexist with negative net performance if occasional losses are large enough. Metrics should be read together, not ranked in isolation.

Do not optimize every slice

With many possible time windows and setup labels, some groups will look unusually good by chance. Record your question before exploring the results and avoid turning every historical pattern into a new rule.

Keep collecting comparable records

Consistent definitions and complete fees improve the value of future reviews. More observations are helpful when they reflect a reasonably comparable process, not just a larger pile of inconsistent data.