An average computed from ten trades is highly sensitive to one or two results. The uncertainty in an average falls roughly with the square root of the number of trades. If your results have a spread of around 1R, the average of 100 trades is uncertain by about 0.1R either side, so an observed +0.10R is not clearly different from zero.
The same applies to every subgroup. Expectancy for a single hour, a single setup or a single emotional state may rest on a dozen trades. A striking number from a small group is a prompt to keep recording, not a conclusion to act on.
Slicing makes the problem worse. The more ways you cut a record, the more likely one of the cuts looks remarkable by chance alone. Prefer a small number of questions chosen in advance over scanning for anything interesting.
Practical habits: state the number of trades beside every figure, set a minimum such as thirty trades in a group before you read anything into it, and wait before changing a rule because of a result.
Reflect shows how many trades each finding rests on and caps its confidence so that nothing reads as certain, which is the same discipline applied automatically.