How We Test
Expert Advisors

Full Reviews And Early Reviews

Some Expert Advisors are searched for long before they have two years of real-money trading behind them. Rather than say nothing until then, we review them early, with a lower ceiling.

The Thresholds, In Full

These are the objective criteria an Expert Advisor must clear. Failing any one of them means no Recommended verdict, regardless of headline returns. They are published here in full, on the page, because a standard you cannot read is not a standard.

Evidence integrity

Before any figure is assessed, we establish that the evidence describes the product being sold.

• ❌ The track record must belong to the product under review. Where a developer markets an older product's live signal alongside a newer release, that record evidences the older product. The newer one is treated as unverified, however long the signal has run.
• ⚠️ To be reviewed at all, an Expert Advisor needs a real-money live record of at least 12 months and 300 closed trades, and every criterion on this page must be assessable from it. Under 24 months it is an Early Review and cannot be Recommended. The record does not have to be third-party verified for us to review it. It does have to be verified for a Recommended verdict.
• ❌ A growth figure distorted by deposits and withdrawals is not published. Where an account's cash flows are large relative to its balance, a percentage return measures the funding pattern rather than the strategy, and we say so instead of reproducing the number.
• ❌ The same account shown on two platforms is not independent corroboration. It is one record displayed twice.

Risk

Drawdown is the figure we lead on, because it is the one that describes what happens to your money.

• ❌ A martingale, grid, averaging or recovery strategy is Not Recommended, disclosed or not. Where the trade history shows positions added to a losing trade, or sizes raised to win back a loss, the verdict is Not Recommended whatever the profit, because a sustained move against the position can take the account with it. Where no trade history can be read, the developer's own description of such a mode as the default decides the verdict for the product as sold.
• ❌ A drawdown materially larger than the marketing implies is an automatic fail. There is no universal acceptable number — but a gap between what a buyer is told and what the record shows is a failure of disclosure, not of performance. This applies to live-verified and backtested evidence alike, and it extends to affiliate and reseller marketing the seller benefits from, not only to the seller's own pages.
• ❌ A wipeout or near-wipeout is disqualifying. Poor performance during a stress period is not automatically fatal; losing substantially all of an account is.
• ❌ A floating loss still open and unrecovered 90 days after it began rules out a Recommended verdict, even if it later recovers, because anyone running the Expert Advisor would have had to sit through it. An average holding time can hide a few positions held open in loss for months, so every review publishes the longest time a trade or basket stayed open and the longest open drawdown. Where a record cannot show this, a Recommended verdict cannot rest on it.
• ⚠️ Both drawdown measures are published wherever a platform reports them — by balance and by equity, each labelled. The gap between the two is where a grid or averaging system's real exposure sits, and quoting only one is always a choice about which to show you.
• ⚠️ Where a seller publishes several accounts at different risk settings, we publish the range across their real accounts rather than the best one.

Whether the edge is real

• ❌ A high win rate with a poor payoff ratio is treated as a warning, not a strength. Very high win rates frequently conceal tail risk that the win rate itself cannot show. We have reviewed a real account that won 57 of its 58 trades and still lost 96% of its capital. We therefore never publish a win rate without the profit factor or the average-win-to-average-loss ratio beside it, and we never rank Expert Advisors by win rate.
• ❌ Performance concentrated in a handful of trades fails. If removing a few exceptional trades collapses profitability, the strategy is not robust.
• ❌ A profit factor at or near 1.0 across a large sample indicates an edge that has decayed to breakeven, and will not earn a recommendation however long the record.
• ⚠️ Recovery factor is used as a discriminator; the Sharpe ratio is recorded where a platform reports it but is never a pass or fail criterion and never appears in our reasoning. Across differing monitoring periods and trade counts, platform-reported Sharpe is not comparable, and at two decimal places the differences are noise.
• ⚠️ A suspiciously smooth equity curve with no visible drawdown is treated as a likely sign of curve-fitting or an undisclosed martingale or grid mechanism, both of which can look flawless for a long stretch before one catastrophic loss.

Claims and sourcing

• ❌ No claim without a retrievable source. Every assertion we make about a third party carries a link to where we read it, with the date. Where a source has since disappeared, we say so and place no weight on its specific claims.
• ⚠️ A critic's interests are disclosed as readily as a seller's. Criticism published by someone selling a competing product is cited for transparency, never used as support for a verdict.
• ⚠️ Every figure is a dated snapshot carrying its source and measurement period. We never silently refresh figures to track current values — doing so would let a verdict ride on numbers that were never the ones assessed.

How We Test EAs Against Prop Firm Rules

A conformance test asks one question: would this configuration, exactly as supplied, have passed one prop firm’s published challenge rules? The answer is a Pass, Fail or Inconclusive result, never one of the three verdicts above. Every result so far is on Best EA for a Prop Firm Challenge.

The test environment

Every test runs the same way, so the results can be compared with each other.

• ✅ One firm, tested on its own data. We test against FTMO’s rules in the MetaTrader 5 Strategy Tester logged into an FTMO Free Trial account, so the price history, spreads and commission are FTMO’s own, not a stand-in broker’s. Anyone with a free FTMO trial can repeat a test.
• ✅ Real ticks only: every tick based on real ticks at 99% history quality or above, a £10,000 GBP account at FTMO’s leverage cap of 1:100, and one common window for every test, currently 1 January 2025 to 1 October 2026.
• ✅ Commission is always charged. Every report must show FTMO’s commission. If a server ever charges none in the tester, we add it afterwards from the trade list at FTMO’s rate.
• ⚠️ Orders fill without delay in the tester. Slippage is not modelled, which flatters every result slightly. We say so rather than add a figure we cannot measure.

Which settings we test

• ✅ Only where the developer makes a prop firm claim. We run a conformance test only on an Expert Advisor whose developer says it can pass a prop firm challenge, or supplies prop firm settings. No claim, no test.
• ✅ The developer’s own prop firm settings file, supplied by the developer or published by them, including in their MQL5 product comments, tested unmodified. Altering it before establishing a baseline would destroy the only claim worth making.
• ✅ If none exists, the developer’s default settings, with any prop firm feature built into the EA switched on at the developer’s own value. We say that no prop firm settings were available. The only other change we make is to symbol names, to match the firm’s own. Where an EA ships with a prop firm feature switched off, such as a daily drawdown limiter, we switch it on, because the developer presents it for prop firm trading. We never set the level ourselves, and we always publish the run as installed alongside it. A built-in feature that has no setting, such as a trade randomiser, runs as built. (Rule added 31 October 2026.)
• ❌ Never a third party’s settings file, and never settings we built ourselves. Neither can be traced to the developer, so a result on them proves nothing about the product.
• ✅ The Expert Advisor is the free MQL5 Market demo wherever the product is listed there. Nothing is bought for a test.

How a result is decided

• ❌ A breached limit is a Fail, and so is a profit target never reached over the full test window. Every limit is reported separately with its margin; a near miss is reported as a near miss, with the figure.
• ⚠️ A challenge phase ends when the profit target is reached with all trades closed, so we judge the limits from the start of the test, or of the out-of-sample part, up to the end of that day. Anything later is reported for information only. Where a developer re-optimised inside the test window, we judge the trades from that date as a rebased comparison of the same run: measured from that day’s balance against the starting account’s target and limits, and only when trade sizes did not grow with the balance. (Rule added 1 November 2026.)
• ⚠️ The daily loss limit is measured on equity, including losses on open trades, and the tester reports closed results only. A closed loss above the limit is a breach. Otherwise we publish the worst closed day and record the daily limit as not established, never as passed.
• ⚠️ We separate what a developer could have tuned from what they could not. Every test records the version tested, the settings file’s date and the date the developer last re-optimised the strategy, from their own release notes; bug fixes do not count. The window is split at that date: before it is in-sample, after it is out-of-sample. If the last re-optimisation predates the window, all of it is out-of-sample.