Standard error is the honest width around every number a backtest reports. The formula sd/sqrt(n) has one brutal implication: to halve the uncertainty you need four times the data. Going from 100 to 400 trades buys you the same precision gain as going from 400 to 1,600.
Worked example: 250 trades, mean +0.12%, standard deviation 1.5%. Standard error = 1.5/sqrt(250) = 0.095%. A rough 95% confidence-interval is 0.12 +/- 1.96 x 0.095, so from -0.07% to +0.31%. The strategy could easily be a loser after costs.
The same arithmetic applies to the sharpe-ratio: its standard error is approximately sqrt((1 + S^2/2)/n) in the same units as the measurement period, which is why a 1.8 Sharpe over one year is close to meaningless.
Related: confidence-interval, sample-size, central-limit-theorem, signal-to-noise