How to Choose a Safer MT5 Expert Advisor: A Practical Buyer's Checklist

Why most EA pitches fail before you even read the price

The MQL5 Market and the wider MetaTrader 5 ecosystem are full of expert advisors (EAs) promising smooth, near-vertical equity curves. Most of those curves are not records of what a robot did. They are renders of what the robot was tuned to do on a single slice of history. That distinction is the whole game.

A curve-fitted EA is one whose parameters have been optimized so tightly to past data that the result looks flawless in the tester and then falls apart the moment market conditions shift. The pitch sells the curve; the buyer inherits the drawdown. Your job, as a buyer, is to separate a genuinely engineered strategy from a marketing artifact before any money is at risk.

This guide is a practical checklist for doing exactly that. It is educational, not financial advice, and where it uses EudoraLab's own EAs as examples, every figure cited is a MetaTrader 5 Strategy Tester backtest, never live trading and never a forecast.

The single most useful filter: if a vendor shows you gains without showing you the matching drawdown, you are looking at a render, not a record. Walk away.

1. Demand real-tick (Model=4) backtests and reproducible set files

MetaTrader 5's Strategy Tester has several modeling modes. The most accurate is "Every tick based on real ticks" — internally referred to as Model=4 — which replays the broker's actual recorded Bid/Ask ticks rather than interpolating prices between bar opens and closes. Real-tick mode reflects spread widening, wicks, and spikes far more faithfully than the cheaper "every tick" or "open prices only" modes.

This matters because lower-fidelity modes systematically flatter certain strategy classes. A grid or scalping system that looks profitable on synthetic ticks can bleed out on real ones, where the spread and slippage it ignored actually exist.

What to ask for

• The modeling mode. If the vendor cannot say it was Model=4 / real ticks, treat the numbers as decorative.
• The exact set file, symbol, and timeframe. A serious vendor ships these so you can reproduce the same backtest on your own terminal. EudoraLab, for instance, ships the exact set file, symbol, and timeframe for each EA so a buyer can rerun the test rather than trust it.
• Spread and commission assumptions. Real-tick fidelity is wasted if costs were set to zero.

Reproducibility is the antidote to a faked screenshot. A static image of a perfect equity curve proves nothing — it can be edited in seconds. A set file you can run yourself cannot lie to you.

2. Look for drawdown shown beside every gain

Return without risk context is meaningless. A 900% backtest with a 70% drawdown and a 900% backtest with a 20% drawdown are completely different products, and only one of them is survivable for most accounts.

When you read a backtest, read the gain and the drawdown as a single pair, never in isolation. Pay attention to:

• Relative (equity) drawdown — the deepest peak-to-trough fall in account equity, including open floating positions.
• Profit factor (PF) — gross profit divided by gross loss. A PF comfortably above 1 across many trades is more reassuring than a huge headline return.
• Trade count and time span — a result built on tens of thousands of trades across many years is harder to fluke than one built on a few dozen.

Honest vendors put these side by side. As an illustration of that presentation style, here is how EudoraLab states its backtest pairs — again, all Model=4 Strategy Tester results, not live performance and not promises:

Notice how different those drawdowns are. The same brand, four very different risk profiles. A buyer who reads only the headline multiples and ignores the right-hand columns is setting themselves up for a shock.

3. Understand the strategy class — and its built-in risk

If a vendor cannot or will not explain how the EA actually opens and manages trades, that is itself a red flag. "Trend, momentum, and volatility" with no further detail is often a euphemism for a hidden grid or martingale.

But here is the part most buyer guides get wrong: grid and martingale are not automatically scams. They are a strategy class with a specific, knowable risk shape. What is dishonest is hiding it. What is responsible is disclosing it and bounding it.

The three classes you will meet most

• Grid / recovery-grid / controlled martingale. The EA adds positions on one side as price moves against it, aiming to exit the whole basket at a small average profit. These systems can show very high win rates and smooth curves precisely because losers are held and averaged — which means they carry floating drawdown by design and can incur large losses in a sustained adverse move. No honest vendor of such a system will ever claim it has "no grid" or "no martingale."
• Mean reversion. Enters against short-term extremes expecting a snap back. Risk concentrates when the market trends instead of oscillating.
• Momentum / trend. Enters in the direction of strength. Typically lower win rate, but losses are cut rather than averaged.

EudoraLab is openly built on controlled-grid mechanics for several of its EAs, and says so plainly. EudoraAegis builds a recovery grid on gold in 1.5x steps; EudoraAntaeus runs a leveraged controlled-martingale grid on the S&P 500, with lot per level scaling as 1.5^level. EudoraFuji adds positions in a spaced, martingale-style progression on the yen crosses. These are grids. The point is not that grids are forbidden — it is that the risk must be named, capped, and visible, never denied.

4. Check for account-level stops and structural gates

The difference between a controlled grid and an account-killer is almost always the presence of hard, structural limits that the strategy cannot override. When you evaluate a grid or recovery system, look specifically for:

• An account-level equity stop. A rule that closes a basket once drawdown hits a defined cap, rather than averaging forever. EudoraAegis, for example, uses an EquityStop that closes a basket at the profile's drawdown cap; EudoraAntaeus uses a forward, causal crash-shield that closes and pauses on defined conditions (such as a 10-day return below -10%).
• A trend or regime gate that gates entries. EudoraAntaeus arms a long basket only while the daily trend is up (D1 close at or above its 50-day SMA) — a load-bearing anti-ruin condition that keeps the grid from stacking into a falling market.
• Event and time filters. Central-bank blackout calendars, news pauses, and Friday-stops reduce exposure during the moments grids are most fragile.
• An AutoResume / breaker mechanism so the EA halts and restarts deliberately rather than fighting a crash.

The presence of these gates does not make any EA safe. It makes the risk bounded and intentional rather than open-ended. An uncapped grid with no equity stop is the classic account-blower; a capped one is a defined bet you can size around.

5. Beware of guaranteed, 100%-win, and risk-free language

This one is simple. Any of the following phrases should end the conversation:

• "Guaranteed profit" or "guaranteed returns"
• "100% win rate" or "never loses"
• "Passive income" or "risk-free"
• "Set and forget, withdraw monthly"

Markets do not offer guarantees, and no honest trading-systems vendor will pretend otherwise. A high backtested win rate is not the same as a guarantee — an 87% backtested win rate, like EudoraAegis shows on gold, still means losing baskets happen and the floating drawdown is real. The correct framing from any serious house is some version of: these are backtests, they are not forecasts, and you should trade only with risk capital you can afford to lose.

Treat the absence of risk language as a risk signal. The vendors most worth trusting are the ones who spend paragraphs telling you how their system can lose.

6. Forward-test on demo, then verify on MQL5

A backtest, even a perfect Model=4 one, describes the past. The only thing that describes the future is forward behavior. Before committing capital:

Run it on demo or forward first

• Load the vendor's exact set file on a demo account and let it run on live-incoming ticks for several weeks across different conditions.
• Confirm the live demo behavior broadly resembles the backtest's trade frequency and drawdown character. A system that traded 73 times a month in backtest but goes silent — or hyperactive — on demo deserves scrutiny.

Verify the vendor on MQL5

• Check the seller's MQL5 profile, product reviews, and any published Signals. Third-party-monitored forward results carry more weight than any screenshot.
• Confirm the listing details match the marketing. EudoraAegis, as an example, is MQL5 listing 174859 at $297 — details a buyer can independently confirm on the platform.
• Prefer vendors who buy through the MQL5 Market itself rather than off-platform checkouts, so the platform's protections and update mechanism apply.

If anything in the live/demo phase contradicts the backtest, believe the live data. Backtests inform; forward behavior decides.

7. Mind prop-firm drawdown rules

If you intend to run an EA on a proprietary-trading-firm (prop) account, the strategy class question becomes existential. Prop firms enforce hard daily and total drawdown limits, and a breach is usually an instant account failure — regardless of where the trade would have eventually closed.

This is fundamentally incompatible with most grid and recovery systems, which by design carry floating drawdown while a basket is open. A grid that would have recovered to a small profit can still trip a prop firm's intraday equity limit on the way there, ending the challenge before the recovery ever happens.

• EudoraAegis explicitly states it is not suitable for prop-firm daily/total drawdown rules — the floating basket drawdown can breach them.
• Any leveraged controlled-martingale system, such as EudoraAntaeus, carries the same incompatibility for the same reason.

If a vendor markets a grid EA as "prop-firm ready" without heavy qualification, be skeptical. Match the strategy class to the account rules before you ever press start. For more on how these mechanics interact, a vendor's FAQ is often the fastest place to confirm suitability.

Putting the checklist together

None of these checks, alone, certifies an EA as "safe" — no EA is risk-free, and grid and recovery systems in particular can incur large losses. But applied together, they reliably separate engineered systems from marketing artifacts.

Before you buy, confirm you can answer yes to each:

• Is the backtest Model=4 (real ticks), with a reproducible set file, symbol, and timeframe?
• Is drawdown shown beside every gain, with profit factor and trade count?
• Do you understand the strategy class and its specific risk — including any grid/martingale floating drawdown?
• Are there account-level equity stops and structural trend/regime gates?
• Is the vendor free of guaranteed / 100%-win / risk-free language?
• Have you forward-tested on demo and verified the seller on MQL5?
• Does the strategy class fit your account's rules, especially prop-firm limits?

EudoraLab's house standard — real-tick backtests, shipped set files, drawdown stated beside every gain, and open disclosure that several of its EAs are controlled grids carrying floating drawdown by design — is offered here as one example of that presentation discipline, not as a recommendation to buy. Every figure in this article is a backtest, not a forecast or a guarantee. Nothing here is financial advice. Whatever you choose, trade only with risk capital you can afford to lose.

The Operators · EudoraLab

Written by the EudoraLab desk — the operators who build, backtest and ship these MetaTrader 5 systems. Every figure we publish is a Model=4 real-tick backtest, packaged with reproducible set files so you can run it yourself. Not financial advice.