Every statistical test starts by assuming the thing you hope is true is false. For a trading result the null is normally that the expected return per trade is zero once costs are paid. The test then asks how often random data would produce a result at least as good as yours.
Picking the null honestly is most of the work. If your null is that returns are zero you are testing against cash. If your null is that returns match buy and hold you are testing against a much harder benchmark, and many trend systems fail that one while passing the first.
Example: 300 trades, average +0.08% each, standard deviation 1.4%. The standard-error of the mean is 1.4/sqrt(300) = 0.081%, so the observed mean is about one standard error from zero. You cannot reject the null. The strategy might be good; this sample simply does not show it.
Related: p-value, significance-level, standard-error, permutation-test