Checklist before listing EA — Nine verifications to identify EA that will fall apart after listing
Introduction: EAs that collapse after listing share a common trait
After listing an EA, its forward performance often differs completely from backtests. Buyers reach out with questions. With more people creating EAs with AI, this is happening more frequently.
Many EAs that collapse can be identified before listing, just from reports on hand. You don’t need special tools or the ability to read inside the EA. A list of backtest trades and a spreadsheet are enough.
This time, I will add a new verification to the checks introduced in Parts 1 and 2 and compile it into a pre-listing checklist. The example will continue to be the MT5 sample EA “Moving Average” (EURUSD, 1-hour, Jan–Sep 2026, 193 trades).
I have run about 200 backtests in six months, but none have run in real funds after passing this checklist. That's how often luck can produce favorable results.
Check 1–4: The four numbers from last time
We rearrange the four numbers introduced in Part 2 as a pre-listing criterion. For details,Part 2please refer to.
| Check | Pre-listing guideline | Sample EA |
|---|---|---|
| 1. Number of trades | At least 30 trades. Preferably 100 or more | 193 trades. Pass |
| 2. Median per trade | If negative, be sure to check item 3 | −1.40 USD. Caution |
| 3. Total excluding biggest win | Even excluding top 5, still positive | 5 trades with −34.27 USD. Fails |
| 4. 95% confidence interval of the mean | Lower bound is positive | −1.50〜+11.80 USD. Fails |
At this point, the sample EA hits two checks. If you are listing, you should extend the period and increase the number of trades and verify again.
Check 5: Split into first half and second half
This is the check announced last time. Arrange trades in chronological order and split by half based on the number of trades. Recalculate the figures for each half as in Part 2.
| Item | First half | Second half |
|---|---|---|
| Period | 2026-01-02 to 2026-05-21 | 2026-05-21 to 2026-09-30 |
| Number of trades | 96 | 97 |
| Total P/L | +846.16 USD | +112.71 USD |
| Average per trade | +8.81 USD | +1.16 USD |
| Median | −1.82 USD | |
| PF | 1.84 | 1.15 |
About 88% of the profit is in the first half, and the second half averages only about 13% of the first half. Looking at the total, it seems “positive over nine months,” but in reality it may have merely matched the first-half market conditions by chance.
This check is important because buyers typically move after listing to continue the second half. An EA with a weak second half is likely to remain weak for buyers as well.
Guidelines are as follows:
- Both halves have a positive total
- The second half’s average per trade is at least half of the first half’s
If the optimization period and the non-optimized period are separated, treat the non-optimized period’s performance as the “second half.”
Check 6: Split into buy and sell
An EA that earns only from buying or only from selling may simply have followed the market direction during that period. If the market direction changes, performance changes too.
Sample EA: buys 93 trades for +448.54 USD, sells 100 trades for +510.33 USD. This shows no bias and passes.
Guideline: If one side accounts for most of the total, consider either eliminating the other side or clearly justify it as a dedicated EA for that side.
Check 7: Include actual account costs
The sample EA report showed total fees as zero. MT5 Tester computes without fees unless you enable them. The buyer’s account will have fees and spreads may be wider than the tester.
What to confirm is how much more cost can be added before profit becomes zero. For the sample EA it’s about 46 USD per lot, or about 4.6 pips round trip for EURUSD.
Guideline: Even when adding fees and spreads likely used by buyers, keep this under half of that cushion. EA with many trades and small per-trade profit tends to fail here.
Check 8: View drawdown by “order sequence”
The maximum drawdown in the report is the result of a particular order of past trades. If losing streaks occur in a different sequence, it can deepen.
The actual maximum drawdown for the sample EA was 337.03 USD. If you reorder trades 5,000 times, the worst 5% threshold is 582.40 USD, roughly 1.7 times the actual.
Guideline: Use this “order-sequence-dependent” drawdown when calculating the capital and lot size to recommend on the sales page.
Check 9: List how many variations you tried and chose from
Count, to rough accuracy, how many settings you tried through optimization or AI interaction. As noted in Part 1, the more you try, the more likely a top result is due to chance.
Guideline: If you tried dozens or more, read Checks 1–5 more strictly. For example, don’t consider the average width acceptable until the lower bound clearly exceeds zero. Keep that count in your notes.
Pre-listing checklist (summary)
Print it or copy it to use. If even one item fails, do not list yet.
| No. | Check | Pass criterion |
|---|---|---|
| 1 | Number of trades | At least 30 trades (preferably 100+) |
| 2 | Median per trade | If negative, must verify item 3 |
| 3 | Total excluding big wins | Still positive after removing top 5 |
| 4 | Mean 95% CI | Lower bound positive |
| 5 | First half and second half | Both positive; second half average at least half of first half |
| 6 | Buy and sell | Not earning on only one side. If earning, justify |
| 7 | Actual costs | Buyer’s account costs less than half of the break-even cost |
| 8 | Drawdown | Capital calculated by order-sequence-dependent drawdown |
| 9 | Number of trials | Counted. If many, read items 1–5 strictly |
The sample EA passes items 1 and 6 but fails 3, 4, and 5, and others require caution. The conclusion: “Do not list yet.”
What to write on the sales page
After passing the checklist, document these premises on the sales page so buyers can determine if they can achieve the same results on their own accounts. This also reduces inquiries.
- Backtest period, currency pairs, time frame, number of trades
- Optimization period and non-optimized period (if separated)
- Assumptions for fees and spreads used in testing
- Recommended capital and the drawdown used as the basis
Presenting only “win rate” and “PF” is less credible than providing these four items, which helps buyers read the numbers wisely.
Summary
- Many EAs that collapse after listing can be identified before listing using on-hand reports and spreadsheets
- What buyers do is always “continue the second half.” Always split and review the first half and second half
- If a premise passed, write it on the sales page as-is
The method to write an EA product description that passes this checklist without AI is described in Part 4“Writing EA product descriptions without AI”here.
Bonus: How to carry out this checklist all at once
The numbers in this article come from a self-made verification tool“EA Robustness Check”When you provide MT5 backtest reports, it summarizes Checks 1–8, and returns judgments and points to watch in writing. The judgment criteria are fixed and cannot be softened after seeing the results.
For those who may find it hard to doubt themselves, we also offer a verification service.A third party checks EA backtest robustness14,800 yen (Skijan). We will accept up to three MT5 reports, summarize findings in a PDF, and return within 3 business days. The judgment criteria are fixed and cannot be altered if the result is not favorable.
Note: This article presents the author’s approach to verification and does not recommend specific instruments, methods, or EA trading. Past verification results do not guarantee future performance.