We removed the edge from our own EA. It still passed 25.6%.
Every prop firm EA advertises a pass rate. Almost none of them tell you what a system with no edge at all would have scored on the same rules. We measured it on ours, and published the gap.
We deleted the edge from our own Expert Advisor and ran it through 7,000 simulated FTMO challenges. It passed 25.6% of them. Not because it was good. Because a two-phase challenge with a drawdown floor is a game you win a quarter of the time by accident.
That number is the reason every pass rate you read online is worth less than it looks, including ours. This page explains the test, publishes the result, and shows what is left once you subtract the luck.
Why a pass rate on its own means nothing
A prop firm challenge has a shape that flatters bad systems. You need to reach a profit target before you touch a daily loss limit or an overall drawdown floor. There is no time pressure worth speaking of, and there is no penalty for taking a long time. If your equity wanders around randomly for long enough, it will eventually touch the target on some paths and the floor on others.
The ratio between those two outcomes is not 50/50. It depends on how far the target sits from the floor, and on the shape of the return distribution. On a standard two-phase account, that geometry alone gets you through a meaningful share of attempts.
So when a vendor advertises a 96.8% pass rate, the honest question is not is that true. It is how much of that is the strategy and how much is the shape of the game. Nobody publishes the second number, because measuring it requires deliberately breaking your own product.
The test: same system, no edge
We took the daily return series of our engine over 1,221 trading days, from 1 January 2022 to 31 August 2026, built on 244 million real ticks. Then we removed the mean.
Every day keeps its size. The volatility is identical. The fat tails are identical. The clustering of good and bad days is identical. The only thing that changes is that the average daily return is now exactly zero: the system has no expectancy, and over an infinite horizon it makes nothing.
Then we ran both versions, the real one and the zero-edge one, through the same first-passage simulation: 7,000 independent paths per configuration, resampled in ten-day blocks so that streaks survive the shuffling.
The result
| Challenge format | Real system | Zero-edge control | Contribution |
|---|---|---|---|
| FTMO, two phases 10% and 5% | 90.8% | 25.6% | +65.3 pts |
| Two phases 8% and 5% | 92.5% | 29.9% | +62.6 pts |
| One phase, 8% target and 8% floor | 93.3% | 43.9% | +49.4 pts |
Read the last row carefully, because it is the least comfortable one for us. On a one-phase challenge, a system with no edge whatsoever passes 43.9% of the time. Our 93.3% looks spectacular next to it, and it is a real result. But the honest way to describe it is that the engine contributes 49 percentage points, not 93.
What this means when you are comparing vendors
Three things follow from the test, and they apply to any EA you are looking at, not only ours.
A pass rate without a control is unfalsifiable
If a vendor tells you 96.8% and nothing else, you cannot distinguish a genuinely strong system from a mediocre one running at high risk on a forgiving format. Both produce a high number. Ask which format the number refers to, and what a random system would have scored on the same format.
One-phase formats inflate everything
Because there is no second barrier to clear, one-phase challenges are easier for everyone, including systems that do not work. If a vendor quotes their best number and it happens to come from a one-phase account, they are quoting the format as much as the strategy.
The gap is the product
The useful number is the difference between the system and the control on the same rules. That is what you are paying for. Everything below that line you would have got by flipping a coin with the right variance.
What survives once the luck is removed
Removing the edge is one control. Four more tests decide whether the remaining edge is real or an artefact of how the strategy was built.
| Test | What it rules out | Result |
|---|---|---|
| Walk-forward, 970 rolling 12-month windows | a good average hiding bad periods | 0 negative windows |
| Out-of-sample split at 31 Dec 2024 | parameters fitted to the data | −8% decay |
| Slippage stress to break-even | an edge that only exists at zero cost | 0.129R |
| Monte Carlo, 8,000 resampled paths | one lucky ordering of trades | 0.00% negative |
| Deflated Sharpe ratio, 40 configurations | the best of many attempts | 99.66% |
The out-of-sample number is the one we would look at first if we were you. Nothing after 31 December 2024 ever influenced a single parameter, and on that untouched period the edge per trade fell from +0.133R to +0.122R. A system fitted to its own history typically loses half its edge out of sample, and often changes sign.
The slippage figure deserves a line too. It takes 0.129R of extra cost per trade, roughly 88 USD on a 100,000 account, to wipe the edge out completely. Slippage measured on live client accounts runs between 0 and 0.02R, so there is a margin of six to ten times before the strategy stops working.
The parts we are less pleased about
A test page that only contains good news is a brochure. Two things the same analysis showed that we would rather it had not.
The historical sequence was luckier than typical. The observed maximum drawdown was 27.6R. Across 8,000 resampled orderings the median was 32.4R and the 95th percentile was 49.7R. A drawdown close to twice what we actually lived through is entirely plausible with nothing broken. We size for that, and you should expect it.
Two markets, not many. The engine trades the Nasdaq 100 and gold. We tested the S&P 500, the Dow, the DAX and the FTSE 100 and it did not work on any of them. That is a limit on where the edge lives, and we do not think it goes away by adding more instruments.
How to run this test on any system
You do not need our data to do this. If you have a daily return series, the control takes ten minutes:
- Take the daily P/L series of the system you are evaluating.
- Subtract the mean from every day. The distribution shape stays, the expectancy goes to zero.
- Resample both series in blocks of about ten days, so streaks survive.
- Run each resampled path against the challenge rules: profit target, daily loss limit, overall floor.
- Compare the two pass rates. The gap is the edge.
If a vendor will not give you a daily series, that in itself is an answer. And if the gap comes out small, the pass rate they advertised was mostly the shape of the game.