An Expert Advisor should be evaluated like a system, not like a product screenshot. The right question is not "how much did it make?" but what produced the result, what risks were taken, and how likely is that behaviour to survive outside the test?
1. Understand the strategy before the statistics
You do not need the source code to evaluate an EA, but you should understand its basic behaviour:
- What market conditions trigger entries?
- How are exits handled?
- Are stop losses hard-coded or discretionary?
- Does it add to losing positions?
- Is position size fixed or risk-based?
- Can exposure increase after losses?
If those questions cannot be answered, you are evaluating a black box.
2. Check for hidden risk structures
Some systems create smooth-looking equity curves by accepting rare but very large losses. Look specifically for:
- Martingale sizing.
- Grid behaviour.
- Unlimited averaging down.
- No hard stop loss.
- Exposure that grows after losses.
- Correlated positions opened together.
None of those mechanisms automatically means a system is unusable, but they change the risk profile dramatically and must be disclosed and understood.
3. Treat backtests as evidence, not proof
A credible backtest should use realistic assumptions and enough data to include varied conditions. Questions to ask:
- Was out-of-sample data used?
- Were spreads and commissions realistic?
- Was slippage considered?
- How many trades are in the sample?
- How sensitive are results to small parameter changes?
- What is the maximum drawdown?
A system that only works on one exact parameter set is more fragile than one that performs reasonably across a range.
See Backtesting vs Forward Testing for the full validation framework.
4. Forward test the actual deployment environment
The EA should be forward tested on the same kind of environment in which you intend to run it: broker, symbols, spreads, VPS and platform settings all matter.
Forward testing helps reveal:
- Live spread behaviour.
- Slippage.
- Broker-specific execution differences.
- Session and symbol naming issues.
- Connectivity and restart behaviour.
- Market regimes not represented well in the optimisation period.
5. Evaluate drawdown before return
A system that returns 30% with a 5% drawdown has a very different profile from one that returns 30% after a 35% drawdown.
Look at:
- Maximum equity drawdown.
- Length of drawdown periods.
- Recovery time.
- Worst losing streak.
- Largest single loss.
- Exposure during the worst period.
Our Drawdown Management Playbook explains how to interpret and control these periods.
6. Check broker and infrastructure sensitivity
Some EAs are robust to execution differences; others are highly dependent on tight spreads and fast fills. Test whether modest changes in spread, slippage or latency destroy the result.
If they do, infrastructure becomes part of the strategy. Use the Trading VPS Selection Guide and Broker Due Diligence Guide before trusting live execution.
7. Look for regime dependence
Ask when the system tends to work and when it struggles. Trend systems, mean-reversion systems and breakout systems all have environments that favour them.
A robust evaluation asks:
- What happens in low volatility?
- What happens in high volatility?
- How does the EA behave around news?
- Does performance cluster in one market regime?
This is one reason short forward tests are weak evidence: they may capture only one favourable environment.
8. Demand a sensible failure plan
Every system can fail. A credible deployment plan should define:
- Maximum acceptable drawdown.
- Conditions for reducing risk.
- Conditions for pausing the EA.
- What data triggers a review.
- When a strategy is considered invalidated.
Without a failure plan, traders often keep changing settings or increasing risk while performance deteriorates.
A practical EA evaluation checklist
Before trusting an EA with meaningful capital, you should be able to say:
- I understand the broad strategy logic.
- I know whether it uses martingale, grids or averaging.
- Backtests use realistic assumptions.
- The sample is large enough to be informative.
- Forward testing exists on current market data.
- Drawdown is visible and acceptable for my risk tolerance.
- The EA has been tested with my broker/infrastructure.
- I know when I would reduce risk or stop using it.
If several of these are missing, you do not yet have enough evidence.
How this relates to Aurion
Aurion is being treated with the same principle: a backtest is not the finish line. The current product status remains forward testing, with public paid access not yet released. The goal is to build evidence before making stronger claims.
Risk disclaimer: This guide is educational content only and is not financial advice. Automated trading systems can fail, and historical or forward-test results do not guarantee future performance. Trading involves significant risk of loss.
Following our work on automation?
Aurion EA is our flagship MT5 Expert Advisor, built and tested in the open. See where it fits in the TTE ecosystem.
Frequently asked questions
What should I check first when evaluating an EA?
Start with the strategy logic and risk controls. If you cannot explain broadly what conditions make the EA enter, exit and reduce risk, a polished equity curve should not be enough to earn your trust.
Is a profitable backtest enough to validate an EA?
No. A backtest is a research tool, not proof of future performance. Forward testing, realistic costs and behaviour across different market regimes matter more than a single historical result.
What is a red flag in an EA?
Common red flags include opaque logic, martingale or uncontrolled averaging, unrealistic backtest assumptions, tiny samples, missing drawdown data and marketing that focuses on return while hiding risk.