Reshuffling your trades thousands of times to see the range of outcomes luck could have handed you.
A Monte Carlo test takes your trades and reshuffles them thousands of times, building thousands of alternate equity curves from the same results in a different order. One backtest shows you one path. Monte Carlo shows you the cloud of paths you could just as easily have got, so you see the range of luck you got in your backtest before you trust in it.
It keeps your trades but scrambles the order, or resamples them, and replays the account thousands of times. Each run gives a different equity curve, a different worst drawdown, a different final number. Stack them all up and you get a distribution instead of one result. Your real backtest was just one version from that pile, and often not the middle one. The 95th percentile drawdown matters more than the one your backtest happened to show, because you might land there instead.
Order matters to your account even when it doesn't change the maths. Ten losers in a row early can bust you before the winners arrive, same trades, different sequence, dead account. A single backtest hides that risk because it shows one ordering. Monte Carlo shows it. If half the reshuffles blow past a drawdown you can't stomach, the strategy is riskier than the cute backtest you just got.
Monte Carlo can only reshuffle what you gave it, so it inherits every flaw in your sample. Feed it an overfit backtest and it hands you thousands of overfit curves, neatly quantified. Plain shuffling also assumes your trades are independent, which isn't quite true if your system clusters wins and losses. It's a brilliant reality check on sequence risk, but it can't invent the bad trade your data never contained. Garbage in, thousands of garbage curves out. So key is to always increase the quality and quantity of the samples you have in your bank of data.