Martingale and Grid EAs on Prop Firm Accounts: Why the Curve Lies

A martingale EA answers a losing trade by opening a larger one in the same direction, usually double the previous size, so a single winner clears the whole string. A grid EA does the same with geometry: orders at fixed intervals, adding exposure every time price moves against the open basket. Many robots combine both, and that combination fails fastest.

Both share the one property that matters to a prop firm: risk per idea is not fixed. It is a function of how wrong the system already is, so the more wrong it gets, the more it bets. That is the exact inverse of what a funded account rulebook enforces.

It is also why the screenshots look so good. The curve is a staircase of small wins with almost nothing in between, because losses are not realised, they are held open and averaged. Floating drawdown never shows up in a win rate. It shows up once, at the end.

An evaluation is not a profit contest. It is a survival contest with a profit condition attached. Firms typically pair a profit target with a daily loss limit and a maximum drawdown, and the drawdown rule tends to end the account first. Many firms measure that drawdown on equity rather than closed balance, so an unrealised loss counts against you at its worst tick.

That detail is fatal for grid and martingale. The strategy's premise is that the open loss is temporary. The rulebook's premise is that it is real, right now. Which is correct over a long horizon does not matter: the account is judged against the rule, not the outcome. If the distinction is still fuzzy, read [daily loss versus max drawdown](/blog/daily-loss-vs-max-drawdown) first.

There is a second collision, and it arrives later. Many firms enforce a consistency or best day condition, and a martingale run produces the exact profile those rules filter for: a long tail of tiny wins and one outlier close worth many multiples of the average. An account can survive the drawdown rule and still stall at payout review. Terms vary between firms and get revised often, so check your own rulebook.

The arithmetic is worth doing once by hand, because the shape is not intuitive. A ladder that opens at 0.10 lots and doubles on each adverse move is at 0.80 lots by the fourth entry and 6.40 by the seventh, with the basket carrying 12.70 lots. Two rungs later the newest order alone is 25.60 lots. One trending session that does not retrace is enough to get there.

The second half of the arithmetic is the part traders miss. A doubling ladder stays alive until it runs out of margin, and on a retail broker account margin is usually the binding constraint. On a prop account it is not. The drawdown rule bites long before a margin call would, so the firm stops the sequence while your own account would still have room. Traders who ran the same EA profitably outside a prop firm are not imagining it. They were playing under a constraint that let the recovery finish.

Risk teams are not watching your chart. The practical method is a pattern query against trade history, and martingale is one of the easier signatures to query for: the evidence sits in the volume column of every ticket you opened.

None of these checks require the firm to identify your software. The behaviour is the fingerprint, so renaming the file or changing the magic number achieves nothing. A discretionary trader almost never scales into a loser at exactly double the previous size, on the same symbol, minutes apart. We covered the wider mechanics in [how prop firms detect rule violations](/blog/how-prop-firms-detect-rule-violations). Automation itself is not the issue: many firms permit expert advisors, and those that do are policing behaviour rather than tooling, as [are trading bots allowed on prop firms](/blog/are-trading-bots-allowed-prop-firms) explains.

Many firms never write the word martingale into their terms. That does not make it safe. The prohibition is spread across clauses that catch the behaviour indirectly, and any one is enough.

The catch-all is the one traders discover late, because it is invoked at payout rather than at breach. An account can pass, trade for weeks, and only meet the clause when a withdrawal request triggers a manual read of the history. The money was never at risk in the market, it was at risk of never being approved. [Why prop firms deny payouts](/blog/why-prop-firms-deny-payouts) walks through that review.

Firms revise these documents often, so a clause that did not exist when you bought the robot may exist by the time you ask for the money. Ask a vendor how fast the software is updated when a firm changes a rule, and treat a vague answer as the answer.

Most vendors will not tell you their robot martingales. Some do not think of it that way: internally it is called grid recovery, smart averaging, or a hedged basket. You do not need their answer. You can establish it yourself in an afternoon.

Test the settings file the vendor actually recommends, not a default preset. Recovery logic is often shipped switched off and enabled by the aggressive or challenge set, which is the one people reach for when an evaluation clock is short. Treat the tester as a floor, not a verdict: it shows the worst sequence inside the data you gave it, not the one that will end the account.

The difference between an EA that manages risk and an EA that hides losses is not sophistication. It is which variable is allowed to move.

In a martingale, position size is the free variable and the loss is the constant that is never accepted. In a risk managed system it is reversed: the loss is accepted trade by trade, and size comes from a fixed fraction of the account and the distance to the stop. A losing streak makes the next position smaller, because the account itself got smaller. That is the logic behind [the half percent rule](/blog/position-sizing-half-percent-rule).

A system built for prop accounts has to go further, because the firm's limits are not your own risk model. It has to know the daily loss limit, the maximum drawdown, and how the firm measures each, then stand the strategy down before a boundary is reached, not after. That is the job PraxAI GUARD does, reading the rules in real time while the strategy trades. PraxAI FUNDED takes over in the funded phase.

The two engines exist because passing and keeping are different games. Most bots pass challenges, the easier half. Only 1 to 3% of funded traders keep the account long term, and a doubling ladder is a reliable way to end up outside that group.

If you recognised your own robot above and the account is currently funded, treat it as a live incident rather than a research project. The order of operations matters, because the worst moment to decide is in the middle of an open sequence.

Sequence matters more than speed. Closing a large ladder at the wrong moment can realise more in one click than the daily loss limit leaves room for, turning a survivable problem into a breach you caused yourself. Check how much of today's allowance is already spent, size the exit against what is left, and close in parts across sessions if the basket is too big to clear at once.

There is no version of martingale that sits comfortably inside an equity based drawdown limit. Cap the ladder and put a hard stop on the whole basket and you have removed the mechanism that made the curve look smooth, keeping only the exposure that made it dangerous. It is not a tuning problem, it is a category problem.

The workable path is the boring one: fixed fractional sizing, a stop on every position, a hard cutoff before the firm's own limits, and a strategy that books its losers instead of storing them. Repeated evaluations are what make the alternative expensive: the average trader spends $2,400 or more a year on repeat challenges. [Why funded traders lose the account](/blog/why-funded-traders-lose-the-account) covers the wider pattern.

PraxAI is built around that constraint rather than against it. The drawdown lock lives in the code and stands the system down before it reaches the firm's daily limit, and every result shown is published on a live, public Myfxbook. None of that says what your account will do. It says how the software is allowed to behave, which is the only part anyone can honestly promise. Bots do not die in the challenge. They die in the rulebook.