Know which algo you can trust before you pay for it.
Evidence-based rankings for MQL5 Expert Advisors — MT5 and MT4 — and the brokers behind them. Every score is built from what can actually be verified — and shows you the proof.
Every product whose track record we could verify
Each card is a real MQL5 signal tied to that product, either by an explicit link on the signal page or by the vendor's own name matching the product's. The line is that account's actual equity, week by week — not a backtest.
300 products, graded
One tile per indexed Expert Advisor, using the vendor's own image. The border is its grade. Hover for the name, click to open its profile.
Two questions, kept apart
A product can be well evidenced and still capable of destroying an account. One number cannot say both, so we publish two — and the second one caps the grade.
Evidence score 0–100
Half of it — 50 points — comes from the forensics of a real MQL5 account tied to the product. The other half is what can be checked without one: the seller's record, what the listing discloses, and the public thread. A product with no live account is marked UNPROVEN rather than failed, and is scored out of the 50 that remain, so it can never reach a letter grade. A demo account counts for nothing.
Risk of ruin LOW → EXTREME
How badly it can fail, scored independently of how well it has done. Martingale, grid, missing stop losses, margin load and leverage count here whether or not the vendor disclosed them — honesty is credited elsewhere; it does not make the mechanic safer. A high band lowers the letter grade and no amount of clean history lifts it back.
What the score is made of
Fifty points can only be answered by a live account. The other fifty are readable from the listing itself, which is why a product with no account still gets a number and an order — it simply cannot climb past one that showed its record.
Capital at risk 10
How much real money the vendor put behind the claim, and at what leverage. An account opened with $100 proves the code runs; it does not show that its author trusts it.
Capital history 14
Whether money was paid in after the start. Later deposits silently rebase every growth and drawdown figure the account publishes, which is the documented way a bad run is made to look survivable. This is the single most useful line on the page.
Money taken out 8
Withdrawing realised profit is the most expensive claim a provider can make and the hardest to stage — the profit has to have existed. An account that only ever takes money in scores nothing here.
Longevity & independence 9
How long it has traded, how often, and whether anyone else's money follows it. A signal with no subscribers behind a product with hundreds of reviews exists to satisfy a checkbox.
Plausibility 9
Sharpe, profit factor, and two tells: equity drawdown far above realised balance drawdown means losses are being carried rather than taken; and a grid system that has never closed a month down is not showing an absence of losses, only losses not yet realised.
The seller 20
Account age, how many products they carry, how many of those anyone ever rated, and whether they publish live signals at all. The largest term that needs no account, and the one hardest to fabricate: an MQL5 account cannot be back-dated.
Disclosure 12
Whether the description states the mechanics in the open — grid, martingale, whether a stop loss exists, what the drawdown has been. Saying so costs nothing here; leaving a reader to find out is what costs.
The listing 10
Screenshots, documented inputs, a published version and update date, a demo to try. Cheap to satisfy, which is exactly why it is worth only ten — but a listing that cannot manage even this is telling you something.
Community 8
The shape of the rating and what the public thread actually says. Deliberately the smallest term in the model: five-star reviews here are documented as purchasable with a free copy, so they are read for anomalies more than for praise.