What do these systems actually do?
A martingale increases position size after a losing trade so that one eventual winner recovers the accumulated losses; a grid opens layered positions at fixed intervals as price moves against it, which produces the same escalation by a different route.
Both convert the DISTRIBUTION of outcomes without changing the expectation. A sequence of losses is not closed at a loss — it is held and added to until price returns far enough to close the basket profitably.
The result is a record of many small wins and almost no recorded losses, because a loss is only booked when the sequence is abandoned. Until then it is an open position, and open positions do not appear in a closed-trade equity curve.
That last point is the mechanism. The strategy has not avoided losses; it has moved them out of the metric people look at.
Why does the backtest look so good?
Three effects compound, and none of them require the tester to be wrong about anything.
The win rate is nearly perfect by construction. A 95% or 98% win rate is not evidence of an edge here; it is arithmetic. Sequences almost always eventually close in profit, and each one is recorded as a win regardless of how much drawdown it carried on the way.
The compensating event may be absent from the sample. A grid survives any move smaller than its total capacity. Test it over a period containing no move large enough and it never fails — and the periods that would break it are precisely the rare ones a five-year sample may not contain.
Closed-trade metrics hide the exposure. Profit factor, win rate and expectancy are computed on closed trades. A basket sitting 400 pips underwater with six open positions contributes nothing to any of them until it closes, so the risk is genuinely invisible in the standard summary.
Put together, these produce a curve that rises in a near-straight line, with a maximum drawdown figure that describes closed trades and not the account.
How do you see the real risk?
Plot equity including open positions, mark to market on every bar, and the smooth line becomes something else entirely.
This single change is the whole diagnosis. Closed-trade equity for a grid rises in small steps; mark-to-market equity shows the sequences — deep, long troughs where the basket was underwater, each one the actual risk that was carried.
Then measure the largest open exposure the system ever held, in lots and in account percentage. That number, not the closed-trade drawdown, is what the account was risking. For most retail grids it is a multiple of the balance.
Third, find the move size that would have broken it. Work out the total capacity — how far price can travel before margin is exhausted — and then check the instrument's history for moves of that size. If a comparable move exists anywhere in the record, the system's survival in your sample was a property of the sample.
Is there ever a legitimate use?
Scaling into a position is a legitimate technique with a defined total risk decided in advance, and that constraint is the entire difference between it and a martingale.
A trader who plans three entries, sizes all three so the combined loss at the invalidation level is one unit of risk, and honours that invalidation is averaging into a position. The maximum loss is known before the first entry and does not change.
A martingale has no invalidation. The position grows to whatever is required, the maximum loss is bounded only by the account, and the rule that would close it does not exist — which is why the strategy has a beautiful record until the single day it does not.
The test is one question asked before the first entry: what is the largest amount this sequence can lose, and did I decide it in advance? A defined answer describes scaling. "Whatever it takes" describes a martingale, however it is labelled.