$649 Ox7F: 10-Year Real-Tick Test | Behind the 95% Win Rate

Testing the $649 Ox7F on 10 Years of Real Ticks — A Real Edge, but One That Cannot Be Reproduced

In our testing series so far, we have reported that many popular EAs advertising high profit factors see their results vanish on real ticks or a different feed (our mass test of 43 gold EAs). This time we look at the opposite case.

We backtested the gold EA "Ox7F" ($649), sold on the MQL5 Market, on 100% Dukascopy real ticks over the 10 years from 2016 to 2025. The result: profitable in all 10 years, with a profit factor of at least 1.74 every year. It holds up when the feed changes and it holds up in stress years. That puts it near the top for robustness among all the EAs we have tested.

This article then goes one step further. Using public information and the actual fills from the live signal, we broke down where this performance comes from. To give the conclusion up front: although we were able to work out almost the entire mechanism, the one thing we could not recover from price data at all was the decision that drives the results — when to enter and in which direction.

Test Conditions (Full Disclosure)

Thin-TP scalpers tend to be overstated under the 1-minute OHLC model (Model=1), so this test used real ticks only from the start (we explain the difference in Real Ticks vs. OHLC).

Results: Profitable 10 Years Running, PF 1.74 or Higher Every Year

It did not break down even in stress years such as the 2020 COVID shock (PF 6.45) or the 2022 interest-rate upheaval (PF 3.75). We also confirmed that it is not the type that looks profitable by deferring floating losses through grids or averaging down. Tracing the trade history at tick precision, the maximum floating loss on winning trades had a median of $2.52 and a maximum of just $8.92; not a single position was held past the $9 stop loss. The stop loss genuinely works.

Absolute net profit becomes enormous because of compounding lot sizes, but that is a product of money management, so we do not use it for comparison and evaluate on PF alone.

Breaking Down the Mechanism — What We Could Work Out

By matching the 133 actual fills from the live signal against the actual fills of the binary itself, we were able to determine the operating structure almost completely.

With a risk-reward of 0.4, the break-even win rate is 71.4%. The product runs at a win rate far above that, and that is the true nature of its edge. There is no mechanism for riding out floating losses and recovering them; it wins purely on the precision of taking $3.60 before touching $9.

Note that the public live signal uses a conservative setting restricted to London hours with a fixed 0.01 lot, which is different from the binary's default of 24-hour operation with compounding. That the same product can behave very differently depending on its settings is something to keep in mind when considering a purchase.

Testing: Can That Win Rate Be Reproduced With "Indicators"?

This is the core of the article. For testing purposes, we built a reproduction EA that implements the identified mechanism as is (take profit $3.60 / stop loss $9.00 / 3 simultaneous positions / orders at the same times), and compared it against the original on exactly the same Dukascopy real ticks. All figures below are measurements from this reproduction implementation built by our site (they are not the product's own results).

Here are the results of swapping in direction rules exhaustively and testing on 2024 real ticks.

Whatever rule we used, the win rate lined up at around 67%, falling short of the 71.4% break-even point, and every one lost money. Even after exhaustively adding filters based on ATR, momentum and time of day, the best we got was 68.5% (PF 0.88). With the same take profit/stop loss, the same feed and the same order timing, the original achieves 95% while our reproduction achieves 67%.

We also eliminated the non-direction candidates one by one.

• It is not timing: About 60% of the original's first positions are placed exactly on the hour (:00:00), and our reproduction does the same. Splitting win rate by timing, the original shows 95.4% and 94.7% — no difference.
• It is not a feed difference: On the same Dukascopy data, the original binary scores 90–100% and the reproduction 67%.
• It is not order flow: We computed buy/sell imbalance from 21.26 million ticks in the first half of 2024 and compared it with the original's position direction: a match rate of 42–50% (essentially random).
• Machine learning doesn't get there either: Even multi-feature models reached only 62.9% directional accuracy on out-of-sample periods.
• "When to enter" is also unpredictable: The original enters on only 3.4% per year of its hourly order opportunities. Trying to discriminate that selection from features gave a precision of 2.9% — worse than random.

What This Test Tells Us

Ox7F's edge is real and robust over 10 years. That is a fact confirmed on real ticks. On the other hand, the decision that produces its win rate — when to enter and in which direction — was not a function of any price data we have access to (including all timeframes and tick-level order flow).

For prospective buyers, this cuts both ways.

The good news: The results are not "backtest curve fitting" or an "OHLC model illusion." Many EAs fail right there.

Points to be careful about:

• This is extremely thin scalping with a $3.60 take profit and $9.00 stop loss. It is highly sensitive to spread and execution quality, and results may change if the account environment changes. Even in our reproduction, the results disappeared on a feed with wider spreads.
• The internals are a black box that cannot be verified or inferred from the outside. Because the buyer cannot understand why it wins, there is nothing to go on when performance starts to deteriorate to judge whether it is a "temporary slump" or "the edge disappearing."
• With a 0.4 risk-reward design, profit and loss flip the moment the win rate drops a few points below 71.4%. You should understand that it is a design with a thin win-rate margin.
• The public live signal settings (London only, fixed lot) and the binary default settings (24 hours, compounding) behave differently. You need to distinguish which numbers you are looking at.

Limitations of This Test

• These are measurements over the 10 years from 2016 to 2025, on a single feed (Dukascopy) with default parameters. Results may differ in other brokers' environments.
• The mechanism breakdown is an inference based on 133 closed positions from the live signal and the binary's actual fills; we did not examine the actual source code.
• The figures from the reproduction (67% win rate, etc.) are the results of an implementation our site built for testing purposes, not the results of the product itself. Please do not confuse the two.

Summary

• Ox7F ($649) was profitable every year with PF 1.74 or higher every year over 10 years of real ticks. Its robustness ranks near the top among the EAs we have tested.
• The mechanism is high-win-rate scalping with a $3.60 take profit, $9.00 stop loss and 3 simultaneous positions, making a 0.4 risk-reward work with a 95% win rate.
• The source of that win rate cannot be recovered from price data. Both direction and entry-day selection came out essentially random with every method we tried. The edge is real, but it sits inside a black box that cannot be verified from the outside.
• We also lay out a framework for evaluating high-win-rate EAs in The Pitfalls of 99% Win Rate EAs.

Our own EAs are held to the same standard: published logic, long-term real-tick testing and disclosure of margin-based maximum drawdown. See our Testing Methodology for the test method and the EA Ranking for the list of measured values.

Notes on Naming and Rights

• The EA names and seller names in this article are trademarks or product names of their respective owners. This article uses them descriptively for the purpose of comparison and critique, and we have no affiliation with any seller.
• All figures in this article are our own measured data. No official images or official performance records are reproduced. The mechanism breakdown and figures such as the 67% win rate are measurements from a reproduction implementation our site built for testing purposes, not the results of the product itself.
• If you find any factual error, please let us know via the contact page. We will verify it and correct it promptly. The original data (reports, trade histories, settings) has been preserved.