Your actual equity curve is one ordering of your trades. Shuffle them and you get another equally valid history with the same expectancy and a completely different drawdown profile.
Run a few thousand shuffles and record the distribution of maximum drawdown, final return and longest losing streak. Typical result: a realised 14% maximum drawdown sits at the 30th percentile, with the 95th percentile near 26%. That upper figure is the one to size against, because nothing about your history made the mild ordering more likely than the harsh one. Resampling with replacement - the bootstrap version - goes further and also varies which trades occur.
Two limits. Shuffling assumes trades are independent, which breaks for strategies with serial correlation or regime clustering, and it cannot invent outcomes worse than any in your sample, so it understates the tail if your record has not met a crisis yet.
Related: sequence-risk, risk-of-drawdown, r-distribution, serial-correlation-of-returns