Enrique Enguix / Profil

I refuse to believe that traders cannot see the value of AntiOverfit PRO.

In our latest experiment, we tested 10,000 strategies across different symbols and random date ranges:

EURUSD, GBPUSD, USDJPY, USDCAD, AUDUSD and XAUUSD.

The result was not subtle.

The higher the AntiOverfit PRO score, the higher the probability that a strategy achieved a better Forward Profit Factor.

Not every high-score strategy wins.
Not every low-score strategy fails.

That is not the point.

The point is that AntiOverfit PRO does what most traders desperately need but almost never measure properly:

It helps estimate, probabilistically, whether an Expert Advisor has a real chance of surviving outside the backtest.

A beautiful backtest is easy.
A curve-fitted backtest is easy.
Finding a strategy with a better probability of working in the future is the hard part.

And that is exactly where AntiOverfit PRO has value.

If someone cannot see the importance of a tool that helps separate random historical luck from more robust trading candidates, then the problem is not the tool.

The problem is how they understand trading.

AntiOverfit PRO is not a magic button.
It does not promise certainty.
It does not guarantee future profits.

But it gives traders something far more valuable than another optimized equity curve:

A probabilistic edge in strategy selection.

And in systematic trading, that difference matters.

This may be one of my last public posts about AntiOverfit PRO.

But I will say it clearly:

If you are serious about Expert Advisors, robustness testing should not be optional.

It should be part of the process.

https://www.mql5.com/en/market/product/168279

After analyzing more than 10,000 strategies across different symbols and random historical periods, we observed a clear tendency:

AntiOverfit PRO scores appear to reflect forward robustness reasonably well.

Strategies with low AOP scores were more frequently associated with weak forward results, often with a Profit Factor below 1.0. As the AOP score increased, forward performance tended to improve progressively in periods where the strategies had not been optimized.

This does not mean the score is perfect.

Some high-score strategies still failed in forward conditions. Likewise, some low-score strategies went on to perform well. Trading remains inherently uncertain, and no robustness metric can eliminate that uncertainty completely.

However, within this sample, those situations appeared to be the exception rather than the rule.

The overall pattern was straightforward:

- Low AOP score → higher probability of forward deterioration.
- High AOP score → higher probability of forward survival.

AntiOverfit PRO is not designed to predict the future. Its purpose is to evaluate how dependent a strategy may be on a single historical path.

This experiment suggests that robustness information can provide meaningful insight before exposing a strategy to unseen market conditions.

A backtest shows one path.

Robustness measures what may survive when the path changes.

🔗 Product:
https://www.mql5.com/en/market/product/168279

From score to action.
A robustness score is only useful if it helps you make a better decision.
That’s why I added a simple practical guide for interpreting AntiOverfit PRO results:
Grade A — High Robustness
Grade B — Solid Robustness
Grade C — Limited Robustness
Grade D — Insufficient Robustness
The product still gives a 0–100 robustness score.
These grades are just a practical way to act on that score.
The goal is simple:
Don’t just admire a backtest.
Check if the EA can survive when the market path changes.
AntiOverfit PRO for MT5:
https://www.mql5.com/es/market/product/168279

Why AntiOverfit PRO uses 100 worlds and around 5–6 years of history

In AntiOverfit PRO, we do not believe that “more history is always better”.

That sounds intuitive, but in trading it is often wrong.

A market from 10, 12 or 15 years ago may contain information that is no longer useful today.

Different liquidity.
Different volatility structure.
Different execution conditions.
Different macro environment.
Different participants.
Different market microstructure.

So the question is not:

“How much history can I use?”

The real question is:

“How much relevant history should I use?”

For most robustness analysis, a 5–6 year window gives a better balance.

It is long enough to include different market regimes:

- trending phases
- ranging phases
- high-volatility periods
- low-volatility periods
- drawdown periods
- recovery periods
- structural changes inside the same market

But it is not so long that the test becomes contaminated by very old market behavior that may have little relationship with current conditions.

Using too little history has the opposite problem.

If you test only a short window, the EA may look robust simply because it was tested inside one favorable regime.

That is not robustness.

That is regime dependency.

This is why AntiOverfit PRO normally works best with:

100 synthetic worlds
around 5–6 years of historical data

The 100 worlds are not there to create “random noise”.

They are there to observe the distribution of outcomes when the same EA is exposed to many plausible alternative versions of the same market.

With too few worlds, the robustness score becomes unstable.

With too much historical data, the test may start measuring behavior from a market that no longer matters.

The goal is not to make the backtest longer.

The goal is to make the robustness test more meaningful.

AntiOverfit PRO does not ask:

“Did this EA win on one perfect historical path?”

It asks:

“Does this EA still make sense when the market path changes within a plausible range?”

That is the difference.

For a long time, I thought the value of synthetic worlds lay in creating versions that were "similar" to the original historical data.

Now I think that wasn't the right question.

The more interesting question is this: What happens if synthetic worlds don't try to resemble the past, but rather represent the type of deformation that historically appears when a training period progresses into its forward stage?

That difference significantly changes the meaning of the test.

In the latest AntiOverfit PRO tests, I've worked on this idea: rules generated in a training period, evaluated against calibrated synthetic scenarios, and then compared with their behavior in the real forward stage.

We've tested it on several sets of 10 symbols and in various time windows.

It's not a definitive conclusion yet.

But the initial results are starting to show something worth noting: the rules that survive better in this new type of synthetic scenario tend, in several cases, to survive better in the forward stage.

This doesn't mean the future can be predicted.

It doesn't mean a high score guarantees anything.

But it does suggest that perhaps the key isn't creating synthetic markets that are "similar" to historical data.

Perhaps the key lies in creating synthetic markets that better represent the transition between training and forward trading.

And that, for AntiOverfit PRO, could be a much more important line of research than the original version of the concept.

#AlgorithmicTrading #MetaTrader5 #QuantResearch #Backtesting #TradingSystems #Robustness #AntiOverfit

I am starting a new AntiOverfit Public Lab sprint.

For the next month I will test several MT5 Expert Advisors under the same question:

Does the EA only look good on one historical path, or does it survive statistically similar market paths?

No accusations.
No investment advice.
No promotional ranking.

Just reproducible robustness records.

The focus will be XAUUSD systems:

Grid EAs.
Multi-strategy EAs.
Breakout EAs.
Scalpers.
FVG systems.
ATR trend systems.
Set files.
Public Lab records.

First public result next week.

The backtest is the sales pitch.
Robustness is the reality check.

More information:
https://www.antioverfit.com

We analyzed several of the top-rated Expert Advisors from the MQL5 Market.

The goal was simple:

not to check whether the backtest looked good,
but to see how robust each EA remained when tested against synthetic market worlds generated from the original symbol.

These synthetic worlds preserve key statistical properties of the original market, but slightly alter the historical path.

The result was surprising.

Some products that look very strong on the original backtest collapsed under small, plausible market-path variations.

Others produced weak or insufficient robustness scores.

And in one case, even a very high score came with a warning signal that makes the result suspicious.

After reviewing the audits, I would probably trust only one of the results.

You can see the full AntiOverfit Audit certificate list here:

https://www.antioverfit.com/certificates

External robustness audit. Not a promise of future results.

AntiOverfit PRO update — new warning flag added

We have added a new diagnostic label designed to detect EAs that may behave in a hardcoded or pre-fixed way.

This new warning is called:

NEAR-CONSTANT TRADES

It joins the previous warning flag already available in AntiOverfit PRO:

FLAT DISTRIBUTION

The new label appears when the number of trades across synthetic worlds is too similar to the original backtest.

Why does this matter?

When an EA produces almost the same trade count across many statistically different synthetic markets, it may indicate that the operations are not truly adapting to market conditions. In some cases, this can suggest that trades are heavily pre-fixed, hardcoded, or driven by rigid date/time/order patterns rather than by a genuine trading logic.

Important clarification:

These warnings must be interpreted with caution.

They are not accusations.

They are diagnostic alerts.

AntiOverfit PRO does not claim that an EA is fake, manipulated, or intentionally hardcoded. It simply highlights statistical behavior that deserves deeper review before trusting the robustness of the strategy.

The goal remains the same:

Do not trust a strategy only because it looks good on one historical path. Test how it behaves across many possible markets.

Stop buying backtests.

This is the single biggest mistake I see in the MQL5 ecosystem.

A smooth equity curve means nothing if it only works on one historical path.

That’s the uncomfortable truth: Most EAs don’t fail because they are “bad”. They fail because they are fragile.

They are fitted to one specific sequence of market events. Change the path — and everything breaks.

This is exactly what AntiOverfit PRO exposes.

Not by adding noise.
Not by randomizing blindly.

But by generating multiple plausible market paths from the same data and asking a much harder question:

What happens when the market doesn’t behave exactly like your backtest?

Same EA.
Different paths.
Very different outcomes.

At that point, you stop looking at profit…
and start looking at robustness.

Because in real trading, you don’t get to choose the path.

—

Stop buying backtests.
Start buying robustness.

—

Robustness validation. Not a promise of future results.

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279

Clear results. Better decisions.

AntiOverfit PRO turns multiple synthetic-market tests into a single Robustness Score.

It evaluates:

• Original Path Bias
• Consistency
• Downside
• Viability

It’s not about trusting the best result.
It’s about knowing whether the result survives.

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279

AntiOverfit PRO doesn’t measure how much you make.

It measures how robust the result is.

Your EA is compared across many plausible market paths.

The real question:

Is your backtest result normal…
or is it an outlier?

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279

Synthetic data is not random noise.

AntiOverfit PRO builds synthetic worlds that remain statistically coherent with the original market:

• Return distribution
• Temporal structure
• Market-path coherence

It must not only look similar.
It must behave similarly underneath.

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279

From one market → to many plausible markets.

1. Real historical data
2. Synthetic worlds
3. Same EA on all worlds
4. Robustness analysis

We don’t look for the best result.
We look for consistency across different market realities.

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279

A good backtest can fool you.

One clean equity curve ≠ robustness.
It’s just one path the market happened to take.

What happens if the path changes?

Because it will.

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279

One month ago, we launched Anti-Overfit Pro on the Market in version 1.0.

One month later, we have reached version 1.32.

During this time, we have published 12 updates on the Market. In other words, we have not even published every internal build we have created. We have worked continuously to solve every issue that came up and, above all, to refine many practical details related to compatibility, panel behavior, different screen types, and overall operational stability.

Today we released another update focused on improving internal efficiency, world-writing speed, and validation speed.
With this version, we consider the launch stage of the product to be essentially complete.

That also means that Anti-Overfit Pro will soon move to its final price, the one we believe this tool truly deserves.

Because, as we explain in the product description, this is not really about 89 dollars. It is about a tool that can help you save a great deal of time and money.

Right now, the product is priced at $89.
Compare that with:

- a typical EA on the Market priced anywhere from $399 to $1999
- and with the $10,000 you may have in your account, which is ultimately what is really at stake

So what does $89 really mean if it gives you information that, when used properly, can be worth far more than that?

For that reason, in the next 24 hours, we will adjust the price of the product.

I also want to thank everyone who genuinely got involved. The people who, instead of leaving a bad review because of a temporary issue, chose to contact me, share what they were seeing, work with me, and help improve the tool. Thanks to that, we have been able to move very quickly and build something that, right now, does not exist in the Market unless you use Anti-Overfit Pro.

And one last important clarification: I did not invent this methodology. I do not want to claim credit for an idea that is not mine. This way of working comes from highly skilled researchers and traders, including people who have even won algorithmic trading championships. They worked in other languages and in other ecosystems. What I wanted to do was bring that methodology into MetaTrader 5 and turn it into a practical, usable, and accessible tool inside MT5.

Thank you to everyone who trusted the project from the beginning.

https://www.mql5.com/en/market/product/168279

AntiOverfit PRO generates synthetic market variations to test if your EA is truly robust or just lucky on one historical sequence. In this 5-minute tutorial, see exactly how it works and why it matters before buying an expensive EA.
https://www.youtube.com/watch?v=yFZW74KX4Og

AntiOverfit PRO - Quick example ---> https://youtu.be/o2qfsYuxdUU

I don’t want to keep repeating myself with this product or turn this channel into constant promotion, so I’m going to tone that down quite a bit here with AntiOverfit PRO.

But I do want to leave one idea very clear.
For years, I’ve developed and sold Expert Advisors with one very specific obsession: to stay as far away as possible from easy marketing, and to build on robustness, logic, and something I would genuinely be willing to use myself. That has always been my way of doing things.
And that is exactly why AntiOverfit PRO feels so important to me.

Because I do not see it as just another MetaTrader 5 product. I see it as a tool that, honestly, I consider close to essential for anyone who takes this field seriously.
At the end of the day, when someone buys AntiOverfit PRO, they are not just buying software. They are buying a way of approaching this business. They are buying judgment instead of illusion, validation instead of blind faith, and a much more real chance to tell the difference between an EA that only looks good in a backtest and one that may actually stand on solid ground.

That, to me, is where the real value is.

Not in promising easy money, but in offering a tool that helps you spend your money more wisely, filter what you buy more carefully, and get much closer, with real honesty, to the only question that truly matters: whether an EA from the Market could actually work, or only happened to look good.

So yes, I’ll keep mentioning it, but much less. The product is already there for anyone who wants to take a serious look at it. I just wanted to say this once, clearly: very few things I’ve released have felt this necessary to me.

https://www.mql5.com/en/market/product/168279

Monte Carlo Is Not Enough: The Question Almost Nobody Asks When Validating an EA

https://www.mql5.com/en/blogs/post/768419

We’re currently working on improving the visual side of AntiOverfit PRO. Some users pointed out that, on laptops or smaller screens, certain elements could appear too small, making the experience less comfortable than it should be.

I want to correct that properly and release a solid update that fixes this issue in the right way. My goal is for AntiOverfit PRO to be not only a powerful tool, but also clear, comfortable, and enjoyable to use on any setup.

I’m listening closely to user feedback and refining every detail so the final experience feels as polished as possible.

AntiOverfit PRO:
https://www.mql5.com/en/market/product/168279