A backtest can do everything right and still be misleading. Realistic execution costs, enough trades to mean something, no lookahead bias, nothing curve-fit about the parameters — and the result can still be sitting at the fortunate edge of what the strategy's own trades could plausibly have produced, rather than near the middle of it. That's a different failure mode than any of the ones already covered, and it's specifically the one Monte Carlo exists to catch.
Passing every earlier check isn't the same as being robust
Each of the checks covered elsewhere in this loop rules out a specific way a result can be wrong — unrealistic fills, insufficient data, a fitted rather than a real pattern. A strategy can clear every one of them honestly and still have gotten a favorable roll of the dice in exactly which trades happened to occur, in exactly what order and mix. None of the earlier checks are built to catch that; they're answering different questions.
What this actually looks like
A strategy backtests well, the execution model is realistic, the sample size is reasonable, nothing about the parameters looks tuned — and then a Monte Carlo pass shows a wide, uncomfortable spread of plausible outcomes, with the actual historical result sitting well toward the favorable end of it. Nothing about the original backtest was dishonest. It just happened to be a good draw from a distribution that also contains meaningfully worse ones, and the single historical run never had a way to reveal that on its own.
Why this is worth stating on its own
It would be convenient if clearing every other check meant a result was automatically safe. It doesn't, and treating it as though it does is how a strategy that's been carefully, honestly tested can still turn out to be riskier than its one visible history suggested. The specific thing missing from an otherwise well-built backtest is a look at the spread of outcomes its own trades could have produced — not a flaw in how it was built, just a question it was never going to answer by itself.
Full reference: docs · The workflow this maps onto: reamerlabs.com