We Real-Tick Tested 800+ Market EAs — How Many Survived?

"This EA shows thousands of percent per year in its backtest — is it real?" We have been asked this so many times that we decided to stop being the one asked and become the one testing. We downloaded 800+ market EAs from the top of the MQL5 Market ranking and backtested every single one on real ticks.

This article is the record of that test. The short version: the number of EAs that survived is far smaller than the sales pages would ever suggest.

Replacing demo numbers with real ticks

Most published results are built on coarse bar data and ideal spreads. We re-ran the same EAs on real ticks — tick-level price reconstruction — with the spread that actually occurs. We tested year by year, resetting to the same starting capital each year, which strips out the inflated appearance that compounding creates.

Just aligning the method changed everything. The higher up the ranking, the more the EAs cluster into gold pairs — and the moment we switched those gold EAs to real ticks, their maximum drawdown exploded and the accounts melted.

How 800+ EAs got filtered out

Step by step. First, only about one in five EAs even ran cleanly and finished positive. The rest never reached a verdict: no trades, too heavy and timing out, or crashing on load.

Next we imposed a safety condition on the survivors: a profit factor above a threshold, a drawdown within a sane range, and a positive net result. Only a few dozen across the whole set passed. Most of the popular products were gone by this point.

The surviving numbers hide a second trap

Here is the real point. The survivors that cleared the safety bar still had a trap left to expose.

We checked the "build date" of each product binary. It turned out that every single high-scoring EA was compiled after our test window. Fit a curve to a past whose answers you already know, and the equity curve can be made as beautiful as you like. That is not proof of edge — it is hindsight.

The awkward part is that you cannot see this build date on the sales page. The "updated" date MQL5 returns is a copy of the original publish date, not the date of the version being shipped today. A product listed in 2017 can perfectly well hand you a binary compiled in 2026. So we stopped reading the product page and started reading the compile timestamp embedded in the binary itself. Across our 1,146 downloads, 70% were 2026 builds.

That kills the idea that an older listing is a safer bet. Authors keep reworking whatever sells, so even a long-established product ships a recent binary. Keep only the EAs whose build date predates the test window — genuinely tested on an unknown period — and the field drops to zero.

How many were left at the end

The build-date gate had already emptied the field, but we checked the third condition anyway: is there any live track record? However beautiful a backtest is, without a record of a real account staying positive over a long period, it is just a number on paper.

The answer was the same — zero. Not one of the EAs that cleared our real-tick safety bar publishes a long-term live record of its own.

We real-tick tested 800+ market EAs, excluded the hindsight cases by build date, and then required a live track record — and not a single one satisfied all three conditions. The products with the flashiest demo results were precisely the ones with no live evidence behind them.

What this test tells us

Demo backtests are useful for rejecting EAs. An EA that melts on real ticks, or whose build date is after the test window, can be dropped on the spot. But the reverse does not hold. No matter how excellent a demo backtest looks, it alone is not proof that the EA will work in live trading.

That is exactly why, for our own EAs, we stopped displaying flashy demo numbers. We show only measured real-tick figures, and we publish all of them — the good numbers and the bad. We guarantee the validity of verification with build dates, and we drop anything that breaks on real ticks from our lineup. Our priority axis is surviving a long period, not looking impressive.

When you choose an EA, use the sales-page numbers as material for rejection. The only things you can truly trust are measured real-tick figures and a live track record built up on a real account. Our measured ranking is public on the EA comparison page. If you want to understand the verification method in more depth, see the real-tick verification report as well.