The 8 Questions Every Trader Must Ask Before Buying an Expert Advisor — And Why the Answers Change Everything
The MQL5 marketplace contains thousands of Expert Advisors. A few hundred of them are genuinely worth evaluating. A few dozen have real, verifiable live performance. And a small number are built on engineering solid enough to hold up in real market conditions over a meaningful period of time.
The problem is that from the outside, they all look similar. Impressive equity curves. Professional descriptions. Claims of neural networks, machine learning, and proprietary algorithms. The marketing language is nearly identical across the entire spectrum — from the system that will collapse in two weeks to the one that might actually trade your account profitably for years.
The difference is not visible on the product page. It is visible only when you know what to look for. This article gives you the framework. Eight specific questions, each targeting a different dimension of risk. If an Expert Advisor cannot pass all eight, it does not belong in a live account.
QUESTION 1: DOES THE SYSTEM USE GRID, MARTINGALE, OR POSITION STACKING?
This is the single most important question, and the answer disqualifies the majority of the marketplace immediately.
Grid systems open additional positions as the market moves against an open trade. Martingale systems increase position size after every loss. Position stacking opens multiple trades in the same direction simultaneously, amplifying exposure. All three strategies share the same structural flaw: they can produce consistent small wins followed by a single catastrophic loss that wipes out months or years of gains in a single drawdown event.
The equity curves of grid and martingale systems are recognizable. They climb steadily with almost no visible drawdown, then collapse vertically. The vertical collapse is not bad luck. It is the mathematical certainty built into the strategy from the beginning. The small wins are funded by accumulating a hidden liability that eventually comes due.
Any Expert Advisor that does not clearly state the absence of grid and martingale in its architecture should be treated as using them until proven otherwise.
What to require: An explicit statement that no grid, no martingale, and no position stacking is used, verifiable in the live performance data by the absence of correlated clusters of open trades moving in the same direction simultaneously.
ICONIC.FX standard: All products operate with a strict one position per symbol policy at all times. No grid. No martingale. No stacking. The constraint is structural, not configurable.
QUESTION 2: IS THERE A HARD STOP LOSS DEFINED BEFORE EVERY TRADE?
A trading system without a hard stop loss is not a trading system. It is a gamble with a delayed reckoning. The stop loss is not just risk management — it is the mechanism that gives every loss a ceiling. Without it, a single unexpected move can be account ending regardless of how many previous trades were profitable.
The stop loss must be defined before execution, not set afterward. A stop placed at entry cannot be moved against the position. If the system moves its stop loss away from price to avoid being stopped out, it is not using a stop loss at all — it is deferring the loss while increasing its size.
Examine the live trade history. Look at the worst trades. If the largest losses are significantly larger than the stated risk per trade, the stop loss is either not fixed or is being moved. Either scenario is disqualifying.
What to require: A defined stop loss on every trade, set before execution, never moved against the position, with maximum loss per trade consistent across the performance history.
ICONIC.FX standard: Every trade opens with a hard stop loss calculated before execution using ATR based dynamic sizing — calibrated to actual market volatility at the moment of entry. The stop does not move against the position. Automatic break even logic moves the stop in favor of the position as it develops, never against it.
QUESTION 3: IS THE PERFORMANCE LIVE, VERIFIED, AND UNINTERRUPTED?
Backtested performance is almost meaningless as a predictor of live performance. This is not an opinion — it is one of the most documented phenomena in quantitative finance. A strategy that has been optimized on historical data will show excellent historical performance almost regardless of whether it has any genuine predictive power. The optimization process finds patterns in past data that do not repeat in future data.
The specific failure mode is curve fitting: the strategy has learned the historical data, not the market. In live trading, when the data is no longer historical, the edge disappears. The backtest showed a 90% win rate. The live account shows 40%. The discrepancy was predictable from the beginning.
Additionally, look for gaps or resets in the live performance record. An account that was started, closed when it hit drawdown, and restarted fresh shows a different live history than one that has run continuously through multiple market conditions. A verified MQL5 signal account cannot be reset — the history is permanent and brokerage confirmed.
What to require: Live, brokerage confirmed performance on a recognized platform — MQL5 signals being the standard — running continuously for a meaningful period, with no resets and no cherry picked start dates.
QUESTION 4: DOES THE SYSTEM ADAPT TO CHANGING MARKET CONDITIONS?
Financial markets are not stationary. The volatility regime of two months ago is not the volatility regime of today. The correlation structure between assets shifts. Trends that persisted for weeks reverse. Ranges that held for months break. A fixed rule set that performed well in one regime will underperform or lose in a different one.
Most Expert Advisors on the market are fixed rule sets. The developer backtested them on a historical period, optimized the parameters for that period, and published the result. The system is a snapshot of one market moment, deployed into a continuously evolving market. Its performance will degrade as conditions diverge from the conditions it was built for.
An adaptive system has a mechanism to update its behavior in response to current conditions. This can take many forms: regime detection that switches strategy selection, reinforcement learning that updates policy from live trading outcomes, plastic neural layers that adjust to new correlation patterns, or allocation engines that rebalance in response to changing relative performance between instruments. The mechanism matters less than the fact that one exists and that it operates continuously, not just at startup.
What to require: A clear, verifiable mechanism for adaptation — not a marketing claim, but an architectural feature that is explained in enough detail to evaluate whether it is real.
ICONIC.FX standard: Every product adapts continuously. ICONIC BTC AI+ maintains an evolving archive of diverse elite strategies and selects from it based on current regime detection, with a neural layer that updates through live trading interactions. ICONIC NEUROCORE AI+ runs a reinforcement learning agent that updates its trading policy from every live outcome. ICONIC KYBERNETIC AI continuously recomputes causal information flow, reservoir features, and Nash equilibrium capital allocation on every tick.
QUESTION 5: HOW DOES THE SYSTEM BEHAVE DURING A LOSING STREAK?
Any trading system will experience losing streaks. The question is not whether they occur — they will — but how the system responds when they do. A system with no response mechanism to drawdown will continue trading at full size during a losing period, compounding the drawdown until it either recovers or destroys the account.
Look for three specific behaviors:
- Position sizing reduction during drawdown: Does the system scale down trade size when the account is in a drawdown? A system that reduces exposure when losing limits the depth of the drawdown automatically, without human intervention.
- Entry filter tightening: Does the system require higher confidence before entering a trade when recent performance has been poor? This reduces overtrading during periods when the edge may be temporarily absent.
- Maximum drawdown gate: Is there a hard limit at which the system stops trading entirely until the account is reviewed? This prevents a temporary losing streak from becoming an account ending event.
What to require: At least one of these three mechanisms, explicitly described and verifiable in the live performance behavior during the worst periods in the trade history.
ICONIC.FX standard: ICONIC BTC AI+ reduces its exploration temperature during drawdown — the AI becomes more conservative, requiring higher confidence before entry. A critical confidence gate can lock new entries for a defined period if the AI confidence score drops below threshold. ICONIC NEUROCORE AI+ and ICONIC KYBERNETIC AI both reduce position sizing through their dynamic capital allocation engines when portfolio performance deteriorates. All products incorporate a maximum spread filter that prevents trading in abnormal market conditions.
QUESTION 6: DOES THE SYSTEM REQUIRE DLLS, EXTERNAL SERVERS, OR THIRD PARTY DEPENDENCIES?
This is an operational risk question that most traders do not think to ask. An Expert Advisor that depends on an external Python server, a DLL library, a cloud API, or any component outside the MetaTrader 5 execution environment introduces failure modes that have nothing to do with the trading strategy itself.
The external server can go offline. The DLL can conflict with a broker's platform configuration. The API can have downtime during a critical market moment. The Python environment can crash without warning during a volatile session. Any of these events leaves the EA running without its core intelligence — or not running at all — at exactly the moment when reliable execution matters most.
Beyond reliability, DLL based EAs require explicit DLL permission from the trader, which itself introduces security considerations. An EA that runs entirely within the native MetaTrader 5 environment eliminates this entire class of risk.
What to require: A clear statement that the EA requires no DLLs and no external dependencies. All computation runs natively inside the MetaTrader 5 execution environment.
ICONIC.FX standard: Every product in the lineup runs entirely in native MQL5 — no DLLs, no external APIs, no Python bridges, no cloud dependencies. All AI computation executes in RAM using MQL5 native matrix and vector types. If MetaTrader 5 is running, the AI is running. No external component can take it offline.
QUESTION 7: IS THE STRATEGY EXPLAINED OR IS IT A BLACK BOX?
A legitimate trading system can describe what it does. Not in full implementation detail — intellectual property is a valid reason to protect code — but in architectural terms that allow a knowledgeable trader to evaluate whether the approach is sound.
Be skeptical of products that describe their technology only in vague terms: "proprietary algorithm," "advanced AI," "unique neural network." These descriptions are unfalsifiable. They cannot be evaluated and they cannot be compared. They are marketing language that substitutes for explanation.
A product that explains its architecture — even at a high level — is making a verifiable claim. The claim can be assessed against known principles of algorithmic trading, tested against the live performance data, and compared to alternatives. This is transparency. It reflects confidence in the engineering and respect for the buyer's intelligence.
What to require: A clear architectural description of what the system does, at sufficient depth to evaluate whether the approach is technically sound — even if implementation details remain proprietary.
ICONIC.FX standard: Every product in the lineup discloses its AI architecture explicitly. ICONIC BTC AI+ uses an evolutionary quality diversity archive, Hebbian neural plasticity, hindsight experience replay, and fractional calculus for long memory feature extraction. ICONIC NEUROCORE AI+ uses reinforcement learning with eligibility traces, Transfer Entropy directed coordination, and Echo State reservoir regime detection. ICONIC KYBERNETIC AI uses bidirectional Transfer Entropy causal gating, a 500-node Liquid State Machine reservoir, a Physics Informed Margin Axiom, and Stochastic Tunneling Nash Pareto capital allocation. The technical basis of each system is documented, publicly visible, and consistent with the code that runs in the live account.
QUESTION 8: DO THE FEE STRUCTURES ALIGN THE PROVIDER'S INCENTIVES WITH THE BUYER'S?
A subscription based pricing model creates an incentive misalignment. The developer earns the same monthly fee whether the EA performs or not. Their income is not connected to your trading outcomes. The worst case for them is losing a subscriber. The worst case for you is losing your account.
A performance fee model — where the provider earns only when the subscriber earns — reverses this misalignment. The provider's income depends on your profitable trading. If the system does not perform, neither party earns. This is the only fee structure that creates fully aligned incentives between provider and buyer.
The strongest form of a performance fee model applies the fee only to net new equity above the previous high water mark. This means the provider earns nothing on recovered drawdown — only on genuinely new profits. The subscriber pays only for performance that took them to a level they have never reached before.
What to require: Either a one time purchase with no ongoing subscription, or a performance only fee model applied to net new equity above the previous high water mark. Any subscription fee regardless of performance should be scrutinized carefully.
ICONIC.FX standard: The Expert Advisor products are sold on MQL5 under standard marketplace licensing. The ICONIC.FX copytrading signal operates on a performance fee only model — no subscription, no monthly cost, fees applied exclusively to net new equity above the previous high water mark. If the signal does not generate new profits for the subscriber, no fee is charged.
THE SCORECARD: APPLYING ALL EIGHT CRITERIA
The framework above gives every trader a structured basis for evaluation. Apply it consistently and the marketplace becomes significantly easier to navigate. Most products fail on question one. A substantial number fail on question three. Very few pass all eight.
For reference, here is how the ICONIC.FX product lineup scores against every criterion:
- Grid / Martingale / Stacking: None. One position per symbol, hard structural constraint.
- Hard stop loss before execution: Yes. ATR based dynamic placement on every trade, never moved against the position.
- Live verified performance: Yes. MQL5 brokerage confirmed signal, continuous and uninterrupted.
- Adaptation to market conditions: Yes. Three different AI architectures across three products, each adapting through a different mechanism.
- Drawdown behavior: Yes. Conservative scaling, confidence gating, and dynamic allocation reduction during losing periods.
- No external dependencies: Yes. Fully native MQL5, no DLLs, no external APIs, no Python, no cloud.
- Transparent architecture: Yes. Every product's AI architecture is publicly documented at the component level.
- Aligned fee structure: Yes. EA products via standard MQL5 licensing. Signal on performance only model with high water mark accounting.
WHERE TO GO FROM HERE
Use this framework on every EA you evaluate — including ours. The criteria above exist to protect you from bad products. Any system that holds up under rigorous evaluation is worth continued investigation. Any system that does not should be eliminated before it is given access to a live account.
The ICONIC.FX product lineup is available on MQL5 under developer profile mauriceprg.
For live market updates, AI system behavior in real trading conditions, and daily trading content, follow ICONIC.FX on Instagram: instagram.com/iconicfxofficial
Community discussions and live announcements on Telegram: t.me/iconicfxofficial
