Hello, Traders,

A new version of the Expert Advisor Plus.mq5 v6.8 has been uploaded.

Hi Popov!

I discovered a small bug regarding MT5 backtester. The grid-generated EA compiles successfully with no errors, but MT5 Strategy Tester Inputs tab is blank. The issue appears when custom enum declarations are placed between input declarations. Moving all enum declarations above the first input fixed the problem and then I could backtest in MT5. It compiled fine in MetaEditor either way though, the editor didn't catch or flag it, but the backtester is more strict about the order I guess.

during the preparation of my new DOE comparing Single Trade vs Grid Fibonacci vs Grid Semi-Martingale, I realized there is one fundamental design point that I would like to clarify.

My objective is not to transform a successful Single Trade EA into a pure Grid system with no original exit logic. My intention is to evaluate the additional value that the Grid can bring to the original strategy by helping recover the 35–40% of trades that would normally end in a loss.

One point is already clear from my previous tests: I will remove the original Stop Loss, because otherwise the Grid may not have enough room to activate and perform its recovery function.

The key question is therefore about the original Indicator Exit and Take Profit.

For example:

* A Single Trade EA enters a position.
* The market moves against the trade and the Grid opens additional positions.
* Before the Grid reaches its Pair Closing or Grid Take Profit target, the original strategy generates an indicator exit signal or reaches its original Take Profit.

What happens in this situation?

1. Does the original indicator exit or TP close the entire Grid basket immediately?
2. Are the original exits ignored once the Grid sequence has started?
3. Is there any internal priority logic that gives control to Pair Closing and Grid management once more than one Grid position exists?

The answer is very important for the DOE design.

The purpose of my experiment is to understand whether Grid can improve a profitable EA by managing the losing trades more efficiently, not to replace the original trading logic with a completely different pure Grid system.

I would appreciate your guidance on what configuration best represents the intended use of EA Studio Plus Grid.

Thank you!
Vincenzo

I wanted to share where I've gotten to and ask the people who've gone further than me for some pointers.

Chapter 1 — The textbook pipeline. Built the full walk-forward: In-Sample to generate, Out-of-Sample to filter, a true held-out Forward as the final judge. Multi-asset, multi-timeframe, multiple brokers' data, Monte-Carlo, all of it. The plumbing's solid — I've stress-tested it.

Chapter 2 — Rich pools don't translate. I can churn out big, healthy-looking pools: nice IS curves, survive OOS, pass the usual robustness gates. But none of that richness carries into Forward. Every metric I'd rank on (expectancy, R²/linearity, stability ... I (think) I've tried them all) shows basically zero correlation with what the strategy actually does forward.

Chapter 3 — ML didn't save it. So I did the obvious next thing: build features from the IS/OOS walk-forward behaviour, label by forward outcome, train a model to pick survivors. Same wall. The IS side (IS/OOS degradation etc) just doesn't carry enough signal about the future. The model can re-rank a useless ranking, but it can't invent signal that isn't there.

Chapter 4 — A genetic-selection wrapper around gen.js. Latest move: I wrapped gen.js in my own GA — breeding strategies generation-to-generation with my own fitness functions, instead of leaning on the built-in search. Lots more control, and some encouraging signs… but I keep bumping into the same questions about what to actually select on, which is why I'm here.

The questions:

1. Any IS/OOS metric that actually correlates with forward, not just looks good in backtest?
2. Shorter / more recent in-sample windows — do they transfer better for you than long ones?
3. Is the real edge in constraining what you generate (skeletons, regime typing, indicator grammar) rather than filtering after?

Happy to share more on the methodology. Where am I being naive?

Cheers.
Ben

I'm very happy that I succeeded in importing Tick Data from MetaTrader 5 Strategy Tester.

This opens the possibility of having real-time, real-ticks, and real-spread testing in our next strategy builder and generator.

I already have a prototype of the data import.

How it works in my tests:

1. Run a special Expert Advisor on any symbol, M1, and "Every tick based on real ticks". It took 10 seconds for 5 years of tick data.
2. A tick data file is exported. Drag and drop it onto the new Strategy Builder Professional - 2 seconds for the import.
3. The Tick data are available within the program. Higher time frames are composed from the tick data. The data is stored locally in the browser.

Tick Data Import - 1.8 million barts

Higher Time-frame data composed from Ticks

A future version of the backtester will be able to open positions within the bars without ambiguity

Trade safe!

You can also use your MT5 data in the Express Generator to create and validate your strategies.

• Export data from MT with the “Data Export.mq5”. (You can download the script from EA Studios, Data Import page.) Set the proper count of bars and commission values in the script's Input.

• Place the files in a subdirectory of ExGen's data directory. For example: C:\express-generator\data\MyData

• Use the directory's name as a server name when running the generator.

node .\bin\gen.js --server MyData --symbol EURUSD --period M15 --max-data-bars 100000

This is exactly what I've been doing, since I bought EG. In the past few days I've been using Darwinex data instead of my own because I thought it would be the same as the Darwinex demo account I'm testing on, so I thought I could skip that step. Turns out I might not be able to and get consistent results.

When the walk forward optimization is set in this way :

- No flag o sl/tp
- No flag on indicator
- Flag on use acceptance criteria

What is ES really optimizing then in this step ?

Thx
Vincenzo

I downloaded the price data from my FTMO MT5 to the EAStudio
with spread of 10 points (the typical spread is 2-5 points- EURUSD)
then created strategies - optimized for 2016-2025
the good ones i (bactested on MT5 for the first months of 2026, and then) backtested on MT5 for 2016- 2025
usually number of trades is similar, but the results many times very different

Who can explain? Which of them is right?

Regards,
Avraham

I found something wrong in Grid Expert Advisor Plus. When I run EA that generated from EAS and set as Grid (Average down) on M5 XAUUSDc symbol of cent trading account that show you attached picture below. 1st position was normally opened follow indicator signal. When price chart moves against the 1st position to grid distance level. 2nd position was opened and open with many position shotly interval as I draw a red square. This is the abnormal situation that I found.

But another account and different broker I run EA on Forex symbol with standard account, no found any problem It's normal.

What happen and any one give the instruction for me to solve this problem.

Thanks.

I would like to ask: did anyone check if it is a good advice to throw correlated strategies after backtesting.
Let's say after backtesting we have 3 correlated strategies, A B and C. On backtesting they are similar, but in live B is the best of them. The problem is that B was created after A, therefore it was thrown away by the software, and we lost the best strategy.

I will be glad to hear your opinion.

Avraham Shenkar

1) The starting point
The starting point was simple.
Like everyone, starting in May/June 2024, I started generating EAs like hell.

But very quickly, the first real question came up:
How do you consistently produce, validate, and deploy multiple strategies at scale — without losing control?

2) Generation — creating the raw material
Everything starts with generation in EA Studio Reactor.
•    building strategies
•    defining rules and settings
•    producing a large and diverse pool of EAs

At this stage, the objective is not perfection.
It is: breadth and diversity

3) Incubation — the first real filter
The first obvious step was the installation of multiple demo accounts, which quickly turned into what we now call the incubation phase.
•    running many EAs
•    letting them accumulate trades
•    observing real behavior over time

The goal here is simple: exposure to real data and initial validation
No shortcuts.
No assumptions.
Just letting strategies run.
Over time, this scaled.

Today the environment is:

•    ~30 MT4/MT5 instances
•    ~1,000 EAs running
•    ~86,000 cumulative trades
•    ~9,000 trades per month

Every time I explain this setup, the reaction is usually the same: “You’re crazy.” Maybe.

But without this, I could not have learned what I know today.

At that scale, something important happens: you stop thinking in terms of individual strategies and start thinking in terms of systems

This is what the incubation layer looks like today:

4) The problem that emerged

As the number of strategies increased, a clear problem appeared:

•    too many EAs
•    too much data
•    discretionary top selections were not really working live
•    no consistent way to evaluate them

This is where the need became obvious: control and structure

5) Building the workflow — FxBlue + AI

To bring structure into the process, we needed two elements:

•    a reliable data source
•    a consistent way to process it

Then FxBlue became the connector.

It provides:

•    unified tracking across accounts
•    consistent trade data
•    a stable base for analysis

On top of that, we built the workflow using AI. Not to predict markets, but to:

•    process large volumes of data
•    apply deterministic rules
•    standardize classification
•    generate repeatable outputs

This allowed us to move from:

•    manual observation

to:

a structured, reproducible workflow

6) FxBlue Workflow — bringing order

With FxBlue as data source and the workflow on top (upstream), strategies are no longer just running.
They are: continuously evaluated and classified based on objective rules

Over time, each strategy moves through defined states:

•    Ongoing Incubation
•    Promotion Watchlist
•    Ready for Live
•    Pruning Box
•    Earth Birds

This transforms incubation from:

•    a collection of EAs

into:

a controlled pipeline

FxBlue governance snapshot — distribution of strategies across the pipeline

From ~1,000 running strategies, only ~40 reach the Top Band and are considered for Masters.

7) Masters — structuring what survived

When strategies reach the Top Band, they are not used directly. They are combined into Masters within Quant Analyzer.

Masters are: structured portfolios of validated EA Studio strategies

Built to balance:

•    size
•    symbols
•    assets
•    equity behavior

The objective is not to find the best combination.

It is:

to verify that validated strategies remain stable once combined into a portfolio structure

This is where individual strategies become a portfolio — return and risk combined

8) Master Governance — keeping the structure clean

Once Masters (Darwinex demo accounts) are running, a second layer (downstream) of control is applied.

This layer focuses on:

•    removing clear failures
•    monitoring degradation
•    tracking inactivity
•    maintaining structure over time

The goal is not to re-evaluate everything again.

It is:

to keep the signal pool clean and controlled

9) Moving from demo to real

Moving EAs directly from demo to real accounts does not work reliably. The fallback risk to poor performance is simply too high. So we moved to a different approach: signal copy trading.

We do not reinstall EAs on real accounts:

•    we open real accounts
•    we select the best Masters of the month
•    we copy trades from Masters into real accounts

The goal is not to overcomplicate the structure.

It is:

to preserve performance by keeping execution in the same environment where it was validated

10) Where we are today

After:
•    generation
•    incubation
•    building the workflow
•    classifying strategies
•    structuring Masters
•    validating signals

we now have: a structured and continuously filtered set of strategies

11) Next step — deployment

Tomorrow we move to the next phase → Launch of 2 new DARWINEX Zero portfolios

These will:

•    select validated Masters and copy their trades
•    understand how Darwinex risk calibration behaves on our structure
•    apply portfolio construction rules
•    target seed allocation (~2 months) and investor capital (~6 months)

Final thought

This is not about a single EA. It is about building a process that:

•    generates strategies
•    accumulates data
•    creates structure
•    filters signals
•    and only then deploys them

We might still be wrong. But one thing became clear:

the focus should not be on the single strategy, but on the system that manages and scales them

Of course, we’ll also share what happens with the new DARWINs.

P.S.: Last but not least, all of this was possible thanks to a team of five traders working together toward a shared goal. Thanks to this forum, two more members will join us next week. I can’t wait to have them on board.

Vincenzo

Here, for good readers only :-) — meaning time and patience required — are some interesting readings in the context of trading strategy selection, statistical relevance, correlations, and a few other related topics.

They might be interesting for anyone who likes to look beyond single backtests and think more deeply about how trading strategies should be evaluated.

• The Adaptive Markets Hypothesis (Andrew Lo).
https://www.researchgate.net/publicatio … erspective

Concept: Explains why trading edges are perishable. It justifies our high-turnover replacement model and the 3-month median lifespan of promoted strategies.

• ...and the Cross-Section of Expected Returns (Harvey, Liu, & Zhu).
https://www.researchgate.net/publicatio … ed_Returns

Concept: Addresses the "Data Mining" problem. It supports our 6-KPI "Filter Stack" as a necessary barrier against the thousands of "lucky" backtests generated by automated tools.

• The Deflated Sharpe Ratio (Marcos López de Prado)
https://www.researchgate.net/publicatio … -Normality

Concept: Corrects for "Selection Bias." It validates our rigorous incubation process before capital allocation.

• Does Academic Research Destroy Predictability? (McLean & Pontiff)
https://www.semanticscholar.org/paper/2 … d0a2582415

Concept: Proves that alpha decays post-discovery. This supports our "Strict Thresholds" policy—treating mature strategies with the same skepticism as new ones.

• Portfolio Selection (Harry Markowitz)
https://www.researchgate.net/publicatio … _Selection

Concept: The mathematical root of our 0.80 Correlation Cap. It proves that portfolio risk is driven more by the link between strategies than by individual performance.

• Honey, I Shrunk the Covariance Matrix (Ledoit & Wolf)
https://www.researchgate.net/publicatio … _Selection

Concept: Addresses the instability of live correlations. It is the theoretical backbone for our dynamic monitoring of "Regime Lock" in the Master accounts.

• Stepwise Multiple Testing (Romano & Wolf)
https://ideas.repec.org/p/bge/wpaper/17.html

Concept: (Stable Institutional Mirror) Supports our Hysteresis Loops (PwL/OgI) by proving that consistent performance across multiple stages is the only reliable signal of a lasting edge.

• Sequential Analysis (Abraham Wald)
https://projecteuclid.org/journals/anna … 30491.full

Concept: (Project Euclid Stable Link) Validates our rule by providing a framework for making continuous "Keep/Prune" decisions as new data arrives.

• The 7 Sins of Quantitative Investing (López de Prado)
https://portfoliooptimizationbook.com/b … -sins.html

Concept: A roadmap for avoiding common pitfalls like survivorship and look-ahead bias, supporting our "Lifetime String" data fusion.

• Crowded Trades and Tail Risk (Pedersen & Stein)
https://www.researchgate.net/publicatio … _Tail_Risk

Concept: Discusses how crowded strategies lead to liquidity holes. It justifies our 10% Global Kill-Switch as a safeguard against systemic events.

Enjoy
Vincenzo

I noticed recently that the number of strategies added to my collection seems to vary over time and I just observed it in real-time this morning.

For example, about an hour ago the collection counts across 5 Chrome instances were:
1, 5, 5, 3, 5

and now an hour later they have updated to:
1, 5, 5, 3, 4

I'm aware that in a given Chrome instance, the number of Collection strategies will be lower than the Ascended count due to the correlation filtering. This is all fine.

My concern is that I don't understand how a collection can grow and then shrink like this?

Is this a bug?
Happy to send any details you need.
Thank you

I would like to optimize for other things like the default ones in the optimizers. Is it doable to ajust the code for that?
An example would be to optimize for a specific win rate percentage ond only that. Or add other custom optimize goals.

is the Generate Best For and Optimize best for Goal accessable for a advanced user to code custom requirement?

how would one have to proceed to generate lots  of eas with WinLoss 30% or 50% hit rate. without having to look in the last pages of the collection after generated them?

I’d like to open a practical discussion about AI in the context of EA Studio.

Not the usual idea that AI will somehow generate profitable strategies automatically. I think most people here already know that this is not the real point, YET!

What interests me much more is something simpler and more practical:

Where can AI actually improve the workflow of people developing EAs with EA Studio?

From my experience, the real value is not in replacing EA Studio, but in helping us manage scale, improve decisions, reduce noise, and better understand what is really happening inside a large EA workflow.

Here are some use cases that I think are genuinely useful. Some of them we are already using in practice.

Strategy & EA understanding
    •    Clustering similar EAs*
Group strategies that are basically variations of the same idea, so we do not think we are diversified when in reality we are not.
    •    Winners vs losers analysis*
Compare profitable vs unprofitable EAs to understand what really separates them: logic type, trade frequency, exit structure, SL/TP profile, market regime fit, and so on.
    •    Feature extraction from strategies
Detect recurring patterns like trend-following, mean reversion, breakout behavior, volatility sensitivity, session dependency, and other structural characteristics.
    •    MQL code analysis*
Review and compare EA logic directly from the code. This can be very useful for debugging, understanding third-party EAs, or checking whether two bots that look different are actually doing something very similar.

Trade-level analysis
    •    Trade distribution analysis*
Study how trades are distributed across time, duration, sessions, weekdays, symbols, and setups.
    •    Winners vs losers at trade level*
Analyze what losing trades look like compared with winning trades: duration, volatility context, time of day, adverse excursion, favorable excursion, exit behavior, etc.
    •    Entry and exit behavior analysis*
Understand whether the edge is really in the entry, in the exit, or in the trade management.
    •    Floating drawdown and recovery analysis*
Look at how trades go into negative territory, how deep they go, how often they recover, and what kind of floating pressure an EA creates before closing.
    •    Trade sequence analysis*
Evaluate losing streaks, recovery sequences, and whether deterioration starts appearing first at trade level before it becomes obvious at EA level.

Incubation & live monitoring
    •    Incubator monitoring*
Detect which EAs are improving, stagnating, or deteriorating over time.
    •    Automatic labeling / classification*
Tag EAs into practical buckets like promising, watchlist, pruning, or ready for promotion based on how performance evolves.
    •    Early warning signals
Spot when an EA starts behaving differently from expectations before the damage becomes too large.
    •    Performance drift detection
Identify when live or demo behavior starts drifting away from the original profile.

Portfolio construction
    •    Diversification support*
Help build portfolios with lower correlation across symbols, logic types, and timeframes.
    •    Role classification*
Identify which EAs behave mainly as profit engines, drawdown stabilizers, or hybrids / bridge strategies.
    •    Exposure mapping
Detect hidden concentration, for example several different EAs all leaning on the same currency or market behavior.

Workflow validation
    •    Process validation at scale*
Check whether the generation + filtering workflow is actually producing better candidates over time, not just more output.
    •    Monte Carlo / WFA interpretation
Summarize robustness results across many strategies when the volume becomes too high for manual review.
    •    Success rate tracking*
Measure how many selected EAs actually survive incubation and become usable.

Operations & scaling
    •    Documentation and tagging*
Keep structure and memory across many EAs, tests, and incubators.
    •    Experiment design support*
Help organize structured tests, for example grid vs no-grid, different parameter families, or broker comparisons.
    •    Reporting and dashboards*
Produce clear summaries of what is happening across the whole workflow.
    •    Log analysis
Detect technical issues, broker execution differences, VPS instability, or unusual behavior in platform logs.

Curious to hear from others:
    •    Are you already using AI in your EA workflow?
    •    Where does it help the most?
    •    Have you found use cases that really improve results, and not only save time?

My personal view: AI is not yet the edge. The edge is still the workflow. But AI can make a good workflow significantly stronger.

* The starred use cases are things we are already actively using in our workflow.

Vincenzo

I’m trying to define a Longevity Index for EAs generated with EA Studio, based on how long strategies remain in a “top band” according to your selection criteria, filters, or KPIs once they are promoted to live trading.

The idea is not to change the selection rules, but to better understand persistence and durability after promotion.

Has anyone here worked on something similar?

A few practical questions:
    1.    Do you track how long an EA keeps meeting your “good” criteria after going live?
    2.    Do you see a real difference in longevity between stronger entries and more borderline ones?
    3.    Do you evaluate this with fixed horizons like 1 month, 3 months, 6 months, etc.?
    4.    How do you treat EAs that drop below the threshold and later recover?
    5.    If you already measure this, what is your typical expectation in terms of months?

I’d be interested in practical experience, constructive exchange, proven best practices…not only theory.

Happy Easter everyone — and thanks in advance for any insights.

Vincenzo