Failure → Selection Criterion
Click any card to flip from failure pattern to selection criterion
Why Some XAUUSD EAs Fail —
And How to Use Failure to Pick Winners
Published 26 June 2026 · 12 min read
The fastest path to picking a winning EA is to know exactly how they fail — then use each failure mode as a due-diligence question. Six failure patterns account for most EA losses. Each one maps directly to a selection criterion: curve-fitting → demand out-of-sample validation. Wrong broker → confirm ECN requirement. No stop loss → verify every backtest trade has a hard SL. Regime blindness → check for trend filter. Developer abandonment → verify active changelog. No live evidence → require 6+ months of verified live results.
Using Failure as a Positive Framework
Most EA buyer guides are positive frameworks: "look for these good things." The problem is that good-looking things are easy to fake. A smooth equity curve can be manufactured through overfitting. A Myfxbook account can be opened last month with hand-picked trades. A "professional" website with five-star ratings costs a few hundred dollars to build. Positive evidence is easy to fake.
Failure evidence is much harder to fake because it requires either absence of bad traits or the presence of specific structural characteristics that are costly and time-consuming to produce. A genuine out-of-sample validation period requires months of forward testing. A 12-month live Myfxbook account requires 12 actual months of trading. An active changelog with specific update details requires actual development work. The negative-space framework — using failure modes to define what good looks like — cuts through marketing noise more effectively than any positive checklist.
What follows is each of the six failure-to-criterion mappings in depth: what the failure looks like, the specific question to ask, how to find the answer, and what a good answer versus a bad answer looks like in practice.
The 6 Failure-to-Criterion Mappings in Depth
Failure: Curve-fitting — EA backtested perfectly, failed live
Parameters were tuned to historical data until the backtest looked perfect. The EA has no real edge — only historical memory of specific past prices.
Selection criterion: Out-of-sample validation period in the backtest report
The developer tested parameters on a different period than the one used to optimise them. The out-of-sample results should be close to the optimisation results.
Red flag answer
"Here is the full 5-year backtest" with no distinction between optimisation and validation data
Green flag answer
"Optimised on 2019–2022, walk-forward tested on 2023–2025. Here are both results."
Failure: Wrong broker — EA profitable at ECN, losing at market maker
Spread differences between broker types erase the edge. A strategy backtested at 0.3 pip average has a different risk profile at 2 pip average.
Selection criterion: Developer specifies exact broker type required (ECN/STP)
The developer names the broker type and ideally the specific broker they use. They do not claim the EA works with any broker.
Red flag answer
"Works with any MT5 broker" — no acknowledgment that broker type affects results
Green flag answer
"Requires ECN/STP with raw spreads. We test on IC Markets and Pepperstone."
Failure: No stop loss — one bad trade wiped months of profit
Without a hard stop loss on every trade, a single news event, VPS outage, or market gap can produce an account-threatening loss with no limit.
Selection criterion: Hard stop loss visible on every trade in the backtest history
Every single trade in the backtest trade log shows a stop loss price. No trade has a zero or blank stop loss field. No martingale recovery is evident.
Red flag answer
Backtest shows trades with no SL or "virtual" SL that may not trigger in all broker conditions
Green flag answer
Every trade in the backtest history shows a stop loss within 15–25 pips of entry on XAUUSD
Failure: Regime blindness — EA only worked in trending markets
The EA was backtested during a trending regime and fails in choppy or high-volatility periods. No mechanism to detect or adapt to regime changes exists.
Selection criterion: Trend filter or regime detection in the strategy description
The developer explains what happens when the market is not trending — does the EA reduce trades? Filter entries? Use an EMA to align direction?
Red flag answer
"Trades in all market conditions at all times" — no discussion of what happens in choppy markets
Green flag answer
"H4 200 EMA filter ensures trades only in the dominant trend direction, reducing false breakouts"
Failure: Abandoned by developer — no updates, no support, silence
Developer moved on. No updates to address changing broker conditions, new MT5 builds, or market regime shifts. EA code grows increasingly outdated.
Selection criterion: Active changelog, recent live trades, developer available for support
The developer has released an update within the last 90 days. Live account shows trades in the last 30 days. Support responses within 48 hours.
Red flag answer
Last update 14 months ago. No Myfxbook activity in 6 months. No response to support enquiry.
Green flag answer
Changelog shows update 3 weeks ago with specific details. Live account showing trades this week.
Failure: No live evidence — only backtests provided
Backtests can be manipulated, cherry-picked, or simply overfitted. Without live forward-test results, there is no evidence the strategy works in real market conditions.
Selection criterion: Verified live account results running longer than 6 months
A Myfxbook or similar verified live account, showing at least 6 months of real trading with named broker and disclosed lot size. Not a demo account.
Red flag answer
Only a PDF backtest report. No live account. "Live results coming soon."
Green flag answer
Myfxbook verified live account, 18 months of data, identified broker, consistent performance across multiple market conditions.
What Good Actually Looks Like — A Concrete Example
Abstract criteria are useful but concrete examples are clearer. Consider what an EA that passes all six criteria actually looks like in practice. The developer describes a specific, named strategy: M15 range breakout entries with an H4 200 EMA trend filter that confirms direction before allowing the trade. They show backtests with a clearly labelled out-of-sample period. They specify ECN/STP broker requirement. Every trade in the backtest history shows a hard stop loss — no exceptions. The strategy description explains how the trend filter prevents false breakouts against the dominant direction.
A changelog exists and was updated 6 weeks ago — specific changes documented. A Myfxbook account shows 14 months of live trading across multiple market conditions, with a drawdown period in there that was subsequently recovered. Customer support responds within 24 hours.
Goldie Razor V2.8.4 is an example of a system where these structural elements are present and verifiable — the H4 200 EMA regime filter, the hard stop loss on every trade, the documented out-of-sample testing, the ECN broker requirement. This is not a guarantee of future performance, but it is the structural profile that distinguishes systems built for durability from those built for impressive screenshots.
Score Your Shortlisted EA — 6 Questions
Apply these six questions to the specific EA you are evaluating right now.
1. Does the backtest include an out-of-sample validation period?
2. Does the developer specify ECN broker requirement?
3. Does every backtest trade show a hard stop loss?
4. Does the strategy description mention trend or regime filtering?
5. Is there an active changelog updated in the last 90 days?
6. Are there verified live results running longer than 6 months?
Related Reading
Questions to ask before buying a gold EA
The full pre-purchase interview guide — the questions behind each selection criterion.
Documentation you need before buying an EA
The specific documents that prove each of the 6 selection criteria above.
Why do most gold trading EAs fail?
The statistical failure picture that motivates this buyer's framework.
Should you trust gold trading bot backtests?
How to evaluate the backtest document once you have it — signs of curve-fitting vs. genuine validation.
How to choose between different gold EAs
The broader EA selection framework, including strategy type comparison.
Frequently Asked Questions
Out-of-sample validation means testing the EA on historical data that was not used to optimise its parameters. In practice: if the developer optimised parameters on 2019–2022 data, they should have tested the resulting settings on 2023–2025 data without further adjustment. You verify this by asking the developer directly: "What period was used for optimisation and what period was used for out-of-sample testing?" A good answer names both periods specifically. A bad answer is any version of "the backtest covers the full history" — that means the same data was used for both optimisation and testing, which provides no meaningful validation.
Myfxbook is verified — the trades are real, from a real broker account — but the framing can still be misleading. A developer who started their Myfxbook account 3 months ago after a 2-year backtest period is presenting a very short live track record. A developer who shows a 2-week period that happened to coincide with perfect conditions is cherry-picking. What you want to see: a Myfxbook account that has been running through at least two different market conditions, including at least one drawdown period that was recovered. Consistent performance across varied conditions is substantially more meaningful than a short streak of wins.
It tells you the developer understands that their EA's profitability depends on specific execution conditions, and they are honest about this rather than claiming it works everywhere. This specificity is positive. By contrast, a developer who says "works with any broker" either has not tested their EA across different broker types, or is running a strategy with such wide margins that spread type does not matter (which usually means low-frequency, higher target trades — a different risk profile). For scalping EAs specifically, ECN/STP with raw spreads is not optional — it is where the edge is.
If the EA is sold through MQL5 Market, the product page has a "Comments" section where updates are often posted, and some developers maintain a dedicated changelog post. Ask the developer directly: "Do you have a changelog?" and "When was the last update?" If sold through a personal website, look for a dedicated changelog page or a release notes section. If the EA has no changelog anywhere and the developer cannot tell you when they last updated it, treat this as a red flag — it suggests the product is not being actively maintained. An actively maintained EA should have at least 3–4 updates per year.
Technically possible but extremely unlikely for sustained profitability. An EA with no out-of-sample validation (criterion 1), running at a market-maker broker (criterion 2), with no visible stop losses in the backtest (criterion 3), no trend filter (criterion 4), no updates in 2 years (criterion 5), and only backtests as evidence (criterion 6) is stacking risk factors that independently each reduce the probability of live profitability. The compounding effect of all six absent criteria makes sustained profitability highly unlikely — though an EA might get lucky in the right market regime for a few months before the structural weaknesses become visible.
Goldie Razor V2.8.4
M15 breakout + H4 EMA filter — built for XAUUSD on MT5