Validation Lab: walk-forward & Monte Carlo

Two stress tests that separate a real edge from one lucky stretch.

A backtest that wins overall can still be carried by a single lucky period — or by the particular order its trades happened to land in. The Validation Lab, on every result page, runs two stress tests that attack exactly those two kinds of luck. The Reality check stays free for everyone; this deeper layer is Pro and up.

Walk-forward: does the edge hold across time?

Run validation re-tests your saved strategy across up to 5 consecutive time windows instead of one big one (short histories drop to 3–4 windows so each keeps enough bars). A strategy that only made money in one bull run gets exposed: it will be green in one window and red in the rest.

1. “Holds up across time” — at least ~70% of windows profitable, with reasonably consistent results. The edge held across periods.
2. “Mixed across periods” — in between: profitable in some windows but not most (40–70%), or mostly profitable yet with results swinging so widely between windows that the average hides the instability.
3. “Regime-dependent” — profit concentrated in one stretch. Expect it to disappear when conditions change. When nothing made money at all the card says “Loses across all windows” instead, because a strategy that lost everywhere did not lean on a regime — it just lost.

Monte Carlo: how much was luck?

Run Monte Carlo re-shuffles the run's actual trades 1,000 times (bootstrap resampling) to ask: given these trades, what range of outcomes could you realistically have gotten? Instead of one equity curve, you get a distribution.

1. Return confidence interval — where most re-shuffled outcomes land. A wide interval means the headline number is fragile.
2. Probability of profit — the share of resamples that ended positive.
3. Worst-case drawdown — how bad the same trades can look in an unlucky order.

How to use them together

1. Read the free Reality check first — if it already says the sample is tiny or costs are missing, fix that before validating.
2. Walk-forward answers "does it survive different periods?"; Monte Carlo answers "how much of this number is trade-order luck?".
3. A strategy worth trusting should pass both: robust across windows and a tight, mostly-positive Monte Carlo distribution.