FX AI Lab's EA Evaluation Template | Not "Strong or Weak" - AI Reveals What Risk Tolerance an EA Suits

When you look at an EA’s performance record, it’s natural to want to know whether it is “strong” or “weak.” At our lab, though, we recommend reframing that question slightly. No EA is “strong for everyone” or “weak for everyone.” There is only the question of compatibility – “what level of risk tolerance does this EA suit?”

In this article, we publish, in a form you can directly copy, the evaluation framework our lab actually uses when reading EA performance records – the FX AI Lab EA Evaluation Template. We score and comment on nine axes – backtest, forward test, risk, martingale (averaging-down) danger level, capital efficiency, stability, market resilience, operational difficulty, and beginner-friendliness – and distill everything into a single closing sentence about “who this EA suits.” Reading How to Read EA Performance (Understanding PF and Max DD) first will make each axis of this evaluation template much easier to follow.

Why an “EA Evaluation Template” Is Necessary – “Strong or Weak” Leads to the Wrong Call

Summing up an EA’s quality with a single word – “strong” or “weak” – lets all the important information slip through the cracks. Take an EA that “temporarily draws the account down by 8% but trends upward over the long run”: that might be worth considering for someone running it with 1,000,000 yen of surplus funds, but it isn’t right for someone running it with 50,000 yen of living-expense money who absolutely cannot afford to lose any of it. The very same EA can be either “suitable” or “unsuitable” depending on who is using it.

That’s exactly why we believe an evaluation’s output shouldn’t be a total score or a star rating, but rather a single sentence describing what kind of risk, and how much of it, the target person can tolerate. There are three reasons we’ve turned this into a template. First, judging every EA on the same axes each time keeps the comparison fair and free of mood or gut feeling. Second, it guarantees a fixed place to check the “inconvenient numbers” that tend to get skipped – max DD, floating loss, and the number of martingale (averaging-down) steps. Third, it locks in a fixed input format for AI to read, which reduces inconsistency in how it’s interpreted.

This is really the flip side of How to Spot a Dangerous EA (9 Warning Signs). If the warning signs are “a net that filters out EAs to avoid,” this evaluation template is “a ruler for measuring the EAs that make it through against your own tolerance.” We recommend using the warning-sign checklist first as a screening step, and then evaluating whichever EAs pass it with this template.

The Nine Axes of the FX AI Lab EA Evaluation Template

We evaluate along the following nine axes. Each axis gets a three-tier rating – great (◎), good (○), or caution (△) – plus a one-line comment that even a beginner can follow. Think of ◎○△ not as “good/bad” but as a symbol showing which way that axis’s characteristic leans. For the martingale danger axis, for instance, “◎ means low danger level” – it doesn’t mean a high danger level is automatically disqualifying.

1. Backtest Soundness

How reliable is the verification performed on historical data? We look at the length of the test period, the number of trades, and modeling quality. Reliability rises when the period spans a long stretch – say, 2018 through 2026 – covering various market conditions (uptrends, downtrends, ranges), with several hundred trades or more. Conversely, a test covering only six months with just a few dozen trades may simply have gotten lucky. See What Is a Backtest? for more detail.

2. Forward-Test Consistency

How closely do the backtest results match the forward-test results (verification run on the actual live market)? A large gap here raises suspicion that the backtest was “curve-fit” – over-optimized to fit the past too closely. An EA whose backtest looks brilliant but which falls apart in forward testing gets a caution (△) on this axis. See also What Is a Forward Test?.

3. Risk Level (Max DD, Max Floating Loss)

How far is the account designed to temporarily drop? We look at maximum drawdown (DD) and maximum floating loss, in both percentage and monetary terms. A “max DD of 8.2%” means that if you’re running 1,000,000 yen, you need to be able to withstand a temporary paper loss of roughly 80,000 yen. Whether you personally can tolerate that number is the single biggest factor separating “suits you” from “doesn’t.”

4. Martingale (Averaging-Down) Danger Level

How much can the design of “martingale” (adding to a position when the market moves against it) inflate floating losses? We judge this by the number of steps (how many times it will add to the position), whether a hard stop-loss (SL) exists, and the interval (how many pips apart each addition is). An EA with many steps and no hard SL gets rated “high” danger on this axis, because floating losses can suddenly deepen during a range or a sharp one-directional move. We’ve run the numbers on just how frightening martingale can get in Stop-Losses and Money Management (The Danger of Martingale).

5. Capital Efficiency

How generous is the return relative to the recommended margin you need to set aside? We rate efficiency as higher when an EA can be run on a smaller amount of capital and still accumulate profit within a tolerable level of risk. That said, “efficient” does not mean “safe.” Raising the lot size to boost efficiency will naturally worsen the risk level (axis 3) as well. The trick is to always read this axis together with axis 3.

6. Stability (Losing Streaks, Monthly Variance)

How much do results swing from month to month, and what is the maximum losing streak? An EA is “stable” the smaller the gap between winning and losing months and the shorter its losing streaks run. Conversely, an EA that has explosively good months but occasionally suffers a big loss is psychologically hard to stick with even if its average profit is identical, so it scores lower on this axis.

7. Market Resilience (Favorable vs. Unfavorable Conditions)

What market conditions is the EA strong in, and where does it fall apart? A trend-following type that’s strong in trending markets tends to rack up small losses in directionless ranges. Martingale types tend to see floating losses deepen in ranges, and also tend to be weak against the sharp moves that follow economic releases. We work from the premise that “no EA is a jack-of-all-trades” and note its biases honestly.

8. Operational Difficulty

How much effort does setup and day-to-day monitoring take? Difficulty rises with the amount of overhead involved – needing a VPS (a server for continuous operation), needing to configure avoidance of economic releases, needing to monitor floating losses, and so on. See also EA Operating Environment (VPS and Setup).

9. Beginner-Friendliness

Is this a good fit as a first EA, or as a “small-scale slot for verifying behavior”? Beginner-friendliness rises the simpler the design, the easier the worst-case floating loss is to grasp, and the more observable its behavior is starting from a small amount. Conversely, an EA whose mechanics are complex and whose worst case is hard to read isn’t suited to being your first, no matter how good its track record looks.

Using It for Discretionary Trading – A “Personality Test” for Your Own Method

This evaluation template isn’t actually EA-exclusive. It applies directly to a method you’re using in discretionary trading too. Discretionary trading and EAs differ only in whether a human or a written rule makes the call – the yardsticks used to measure performance, such as PF, max DD, losing streaks, and market resilience, are the same either way.

For someone who consistently trades “pullback buying” (entering on a temporary retracement in the direction of the trend), for instance, axis 7 (market resilience) would read something like “strong in a clear trend, weak in a range.” Axis 6 (stability) can be filled in just by looking at the losing-streak record from your last several dozen trades. Axis 4 (martingale danger) becomes “high” if you have a habit of raising your lot size to try to win back a loss. Running your own method through this nine-axis personality test puts into words what kind of market tends to break you down, and it surfaces improvements to make to your entry patterns and stop-loss rules.

Terminology used in discretionary trading, SMC (Smart Money Concepts) terminology, and how EAs handle the same idea line up as shown in the table below. Once you get a feel for the fact that these are just different vocabularies for the same thing, you’ll start to see which world each axis of the evaluation template connects to.

Common Discretionary Term	SMC Term	How EAs (Automated Trading) Handle It
Waiting for a pullback/retracement	Entry in a discount/premium zone	Entry-condition logic (soundness rated on axes 1 and 2)
Placing a stop outside the stop-loss line	Outside liquidity (a spot that won’t get hunted)	Whether an SL is set (directly tied to axis 4 martingale danger and axis 3 risk level)
Adding to a position to win back a loss	— (a psychological trap in discretionary trading)	Martingale/averaging-down logic (danger rated on axis 4)
Suiting or not suiting the market	Trend/range, kill-zone time windows	Market resilience, active-hours filter (rated on axis 7)
Losing streaks that erode capital	—	Max DD, max losing streak (rated on axes 3 and 6)

Putting This to Use When Reviewing EAs – How AI Reads Each Axis (Input and Output)

Keeping these nine axes in mind when you look at an EA’s performance record or a library detail page pulls you out of the “I don’t know where to even look” state. At our lab we run this evaluation through AI as well as human eyes, so let’s lay out concretely what the input and output look like.

The input to the AI is, as much as possible, raw numbers: the backtest period, number of trades, modeling quality, PF, win rate, expected payoff, max DD (both percentage and amount), max losing streak, the forward test’s actual live-trading period and monthly results, the martingale step count, multiplier, interval, and whether an SL exists, the recommended margin, and the target instrument and timeframe. We hand all of this over as a single table. Wherever a number is missing, we have the AI explicitly mark it “unknown.” Never letting an unknown item look good by leaving it blank is the crux of this evaluation.

The output from the AI is a ◎○△ rating plus a one-line comment for each of the nine axes, and finally a single sentence on “who this suits.” What matters here is not asking the AI “is this EA strong?” What you should ask instead is “given a design with a max DD of X% and X martingale steps, what risk tolerance does this EA suit?” Simply changing how you frame the question changes the quality of the answer you get back. Red flags like “win rate emphasized in isolation” or “DD undisclosed,” which we listed in How to Spot a Dangerous EA, can also be caught by having the AI flag them as gaps in the input data.

Sample Evaluations – A Walkthrough with MAC v2.0 and SMC Gold Sniper

Words alone don’t convey this well, so let’s apply the template to two EAs our lab actually runs and verifies. Every figure shown here is our lab’s own first-party data – numbers we ran ourselves and publish. We list both the good numbers and the inconvenient ones.

Sample Entry 1: MAC v2.0 (Gold-Only, Martingale Type)

MAC v2.0 is an EA dedicated to gold (XAUUSD). It’s based on SMC concepts, but its design also adds to a losing position (martingale) at a 1.2x multiplier for up to a maximum of 15 steps when the market moves against it. The interval is 30 pips, take-profit (TP) is 15 pips, and there is no hard stop-loss (SL) – the EA manages the position as a whole instead. It’s being run in an HFM copy-trading slot at a guideline of 0.1 lots per 10,000 yen of capital, with a target monthly return of around +10%. That monthly return figure is only a guideline, however – it’s an average that includes months with floating losses and outright losing months, and it does not guarantee future profit.

Axis	Rating	Comment
1. Backtest Soundness	○	SMC-based logic, but the main judgment material is the transparency of the martingale design
2. Forward-Test Consistency	○	Running live via HFM copy trading. The +10% target monthly return is an average that includes “months with floating losses”
3. Risk Level (Max DD, Floating Loss)	△	No hard SL; floating losses deepen in stages if an adverse move continues. Max floating loss is disclosed as a real-world record
4. Martingale Danger Level	High	1.2x multiplier x up to 15 steps, no SL. Total lot size grows substantially by the time it stretches to step 15
5. Capital Efficiency	○	Guideline of 0.1 lots per 10,000 yen. Can be started small, but risk scales proportionally as you increase it
6. Stability	△	Stacks small wins with a 15-pip TP, but swings during occasional deep floating-loss periods
7. Market Resilience	△	Works in a market with retracements. Martingale deepens during a strong one-directional trend or a widening range
8. Operational Difficulty	△	Requires monitoring of floating losses. Copy trading keeps setup burden low but transfers provider risk onto you
9. Beginner-Friendliness	△	Understanding the mechanism is mandatory. Suitable only as a small, high-risk verification slot funded with surplus capital

▼ AI’s Overall Verdict (Who This Suits): “Suits someone who can psychologically withstand phases of carrying a fairly substantial floating loss and who wants to observe martingale-type behavior firsthand with a small amount of surplus capital. Because there’s no hard SL, pushing the lot size too high scales the worst-case floating loss proportionally, so sticking to the 0.1-lot-per-10,000-yen guideline is a precondition. Do not run this with money you absolutely cannot afford to lose.”

MAC v2.0’s detailed design and its max-floating-loss track record are published in MAC v2.0 Verification Data. A numerical walkthrough of the worst-case scenario at martingale step 15 is available in Stop-Losses and Money Management.

Sample Entry 2: SMC Gold Sniper (Gold, M30, Trend-Following Lean)

SMC Gold Sniper is an EA that trades gold (XAUUSD) on the M30 (30-minute) timeframe, using logic that combines SMC with Heikin-Ashi candles and the Parabolic SAR. Its backtest spans a long 2018-to-2026 period, with a profit factor (PF) of 1.87 and a max DD of 8.2%. It is currently undergoing forward-test verification.

Axis	Rating	Comment
1. Backtest Soundness	◎	Long-term 2018-2026 verification. Period covers uptrends, downtrends, and ranges, with a sufficient trade count
2. Forward-Test Consistency	In progress	Currently in forward testing. Presently confirming how far the backtest’s PF of 1.87 is reproduced in forward results
3. Risk Level (Max DD, Floating Loss)	◎	A max DD of 8.2% is comparatively shallow. Designed so that running 1,000,000 yen means temporarily withstanding about an 80,000 yen paper loss
4. Martingale Danger Level	Low-to-medium	Not martingale-driven; entries are logic-based. Floating-loss expansion is more limited than with MAC v2.0
5. Capital Efficiency	○	A PF of 1.87 paces out to “earning an average of 18,700 yen for every 10,000 yen lost.” Not above the 2.5 threshold that raises suspicion of over-optimization
6. Stability	○	Trends upward over the long-term backtest, but monthly variance in forward testing is still being verified
7. Market Resilience	○	Captures direction on gold’s M30 using SMC plus Heikin-Ashi. Relatively weak in directionless ranges
8. Operational Difficulty	○	M30-based, not extremely high-frequency. Stable VPS operation is advisable
9. Beginner-Friendliness	○	Shallow max DD makes the worst case easy to grasp. That said, understand it is still in forward-test verification

▼ AI’s Overall Verdict (Who This Suits): “Suits someone who can tolerate a phase where the account temporarily drops by around 8%, and who prefers a type that captures direction through logic rather than a design that inflates floating losses via martingale. The backtest is long-term with a realistic PF of 1.87, but it is still in forward-test verification, so this suits someone who wants to personally watch how well the backtest and forward-test results line up. Conversely, a cautious person who’d rather wait until a live track record accumulates should naturally hold off judgment until the forward-test results are in.”

Lining these two up side by side shows that “even the same gold EA can have opposite personalities.” MAC v2.0 is the type that accepts floating losses through martingale; SMC Gold Sniper is the type that captures direction through logic. It’s not a question of which is “stronger” – it comes down to which shape of risk is closer to what you can tolerate. See SMC Gold Sniper Verification Data for details on SMC Gold Sniper, and check the Performance Dashboard (losing months disclosed too) for both EAs’ forward-test track records.

Putting This Into an AI Analysis – How to Use This Evaluation

A completed evaluation card isn’t meant to deliver a conclusion on its own – it’s material that feeds into your next decision. There are broadly three ways to use it.

The first is to check it against your own risk tolerance. Look first at whichever of the nine axes you can least compromise on – for most people that’s axis 3 (risk level) or axis 4 (martingale danger) – and confirm whether it falls within what you can tolerate. If that axis fails, passing is the sensible call no matter how good everything else looks. The second is to line up multiple EAs in the same table. Writing them out on the same nine axes, the way we did with MAC v2.0 and SMC Gold Sniper, lets you compare based on design differences rather than gut feeling. The third is to use it as material for deciding whether to join a copy trade. Copy trading reduces your setup effort in exchange for taking on the provider’s logic and risk, so weigh axes 4, 7, and 3 especially heavily. The mechanics are explained in What Is Copy Trading?.

Once you’re done evaluating, be sure to move on to EA Money Management. No matter how well an EA rates, ignoring the recommended margin and raising the lot size will send axis 3’s risk level downhill in a hurry. The evaluation template tells you “who this suits,” but “what will happen to your capital” needs to be confirmed with a money-management calculation. If you’re using an overseas broker as a verification slot, you can check the high-leverage assumptions involved in HFM’s Risks.

Summary

EAs can’t be measured with “strong” or “weak.” The same EA can be a good fit or a bad one, depending on who’s using it. The FX AI Lab EA Evaluation Template scores and comments on nine axes – backtest soundness, forward-test consistency, risk level, martingale danger, capital efficiency, stability, market resilience, operational difficulty, and beginner-friendliness – and distills it all into a single closing sentence about “what risk tolerance this suits.” The trick is not to total up the scores, but to use them to make sure you don’t miss the most painful weak point.

As we saw in the sample entries for MAC v2.0 and SMC Gold Sniper, even the same gold EA can have opposite personalities. Once you’ve finished evaluating, screen candidates with How to Spot a Dangerous EA, apply EA Money Management to your own capital, and check the real data on the Performance Dashboard or in the EA Library – following that order moves you from “this looks impressive somehow” toward being able to decide whether to adopt an EA based on your own tolerance. The full picture of our EA learning track is laid out in the EA & Automated Trading Hub.

Frequently Asked Questions

• Q. Is it OK to total up the nine axes’ scores into an overall score?
  A. We don’t recommend it. Totaling the scores buries the most painful information – how deep the max DD goes, how many martingale steps there are – in the average. This evaluation exists to show each weak point individually; even if eight axes come back ◎, a “high” martingale danger rating is what decides suitability. Judge by “does the axis I can least compromise on fall within my tolerance,” not by a total score.
• Q. Should I just ask the AI “is this EA strong?”
  A. We recommend changing how you frame the question. Instead of “is it strong,” ask “given a design with a max DD of X% and X martingale steps, what risk tolerance does this suit?” – that changes the quality of the answer you get back. The key is to feed in raw numbers as input (PF, max DD, trade count, martingale settings, etc.) and have the AI explicitly mark any missing item as “unknown.” See also How to Read EA Performance for more detail.
• Q. Can this template be used for discretionary trading too?
  A. Yes. Discretionary trading and EAs differ only in whether a human or a rule makes the call – the yardsticks for measuring performance (PF, max DD, losing streaks, market resilience) are the same. Running your last several dozen trades through this nine-axis personality test puts into words what kind of market tends to break you down, and surfaces improvements to make in Stop-Losses and Money Management.