Thanks very much

I’m not sure if I got your question exactly right, but let me explain how I use WFA — maybe this helps.

I actually do different things depending on the phase:

Reactor (generation phase) →
When I generate strategies, I apply a very light optimization (only TP/SL, no indicator parameters) with 20% OOS.
I also always leave out the last 6–12 months completely.

Validation phase →
When I validate existing collections or single strategies, I backtest from 2016–2024, then filter to pick the top performers and run a full 2025 validation (WFA). Here there’s no need to split into segments, since that data was never used before.

Reactor (new setup) →
I recently launched a new creation batch using Walk-Forward Optimization (WFO and no additional optimization) with 10 segments and 40% OOS. This one will run until the end of the week — then we’ll see the results.

Did that answer your question?

Do you have a standard that you use?

In order to keep our discussion constructive and focused on ideas rather than individuals, please avoid referencing personal qualifications, credentials, or experiences in your responses.

Just to be clear: in none of my posts have I ever shared trading results or performance claims.
What I share here is about developing topics and workflows, not promoting systems or selling anything.

Since I bought EA Studio, I’ve reached out to many users to learn from their experience — including Popov, the creator of EA Studio, with whom I’ve have regular exchanges about ideas and workflow development.
That’s the purpose of being active here: sharing, learning, and improving through open discussion — nothing commercial, nothing hidden.

Regarding your accusation about “manipulating results” or “moving accounts between brokers” — that’s simply false.
I use multiple brokers intentionally for cross-validation and robustness testing, which is a standard and transparent practice in algorithmic research. Nothing is hidden or altered; everything I post can be verified or replicated by anyone who wants to test it.

My Darwinex account is fully public, with a four-year live track record that includes both investor capital and platform allocations. The goal isn’t to “look good” — that would be impossible to fake without being delisted — but to trade consistently and within a defined risk framework.

In live trading, the ability to pause, replace, or add EAs to hedge or manage drawdowns isn’t manipulation; it’s one of the core skills of responsible risk management.

I’m sharing my current journey focused on building a repeatable, data-driven process with EA Studio, while my older profitable accounts were based on third-party EAs, which I’ve always acknowledged openly.
They’re two different paths — one past and still running, one experimental and evolving.

If you want to discuss methodology, validation logic, or testing processes, I’m open to that.
But I won’t engage in personal arguments — they don’t help anyone and only distract from what really matters.

Let’s keep the conversation technical and useful for the community.

All the best
Vincenzo

WFA can really open your eyes. It’s often the moment you realize which strategies are truly robust and which ones just look good on paper. Or, the moment you get unexpected surprises.

But, I also run 20+ demo accounts (the oldest over 500 days, three just opened this week) with 1,000+ EAs running at any time to validate workflow, setup, and criteria changes. The demo stage is another gatekeeper — and then, of course, the live environment puts everything under discussion again.

I treat this as a pipeline with gatekeepers:

Gate 1: Generation + Robustness tests

Gate 2: Walk-Forward Analysis in EA Studio

Gate 3: Demo (≥180 days, 50+ trades)

Gate 4: Live (≥30 trades)

My view is, if you create rubbish at Gate 1, your success rate at Gate 4 becomes unmanageable. With only 20–30 % EAs success rate, it’s almost impossible to maintain a profitable portfolio long-term.

But if you filter properly along the way, you only need around a 60–70 % success rate. In a 10-EA portfolio, that’s usually enough for 6–7 to perform well — and that’s what makes the difference.

The road is long and not easy.
I’m sharing what I do not to teach, but to be constructively challenged and learn from others’ experience too.

I’ve been trading for over 10 years, almost 5 of them working with bought, selected, and “domesticated” EAs. What I’ve learned (also thanks to others sharing) is that data sustain results.
The time you invest in studying and testing is much larger than the time needed to manage portfolios.

If we all share a bit of experience — supported by data — we could all learn faster and make this journey more sustainable. Or we could even work together to validate ideas.

Thanks
Vincenzo

If you want to go deeper, try running the same existing collection through a few different filter combinations to select strategies before the WFA. It doesn’t take long, but you’ll see huge differences — sometimes 20 % vs 100 % success rates.

AI can open a world we couldn’t imagine before — but only if you build the right environment and ask the right questions. That’s when it truly becomes a learning tool, not just a helper.
For example, I’m fine-tuning a workflow to select promising strategies running on demo and promote them to live, as well as a systematic process to maintain my live portfolios on a monthly basis.

I also switched to brokers already integrated into EA Studio to skip manual data downloads.
Right now, I’m testing what works best: creating on Premium Data and validating on BlackBull + Darwinex, or the other way around.

So far, I’ve noticed that strategies built on Darwinex often don’t perform as well on BlackBull or Premium — still collecting data on that.

Did you also eperience it?

Interesting discussion, Thanks.

Also the Walk Forward is great. I have a lot of strategies from Generator that pass through all my criteria but sometimes they just look like crap on the WFA. So I use that Walk Forward as a filter for all top strategies now.

I've also learned a lot about trading and understanding statistics by studying the reports and then asking AI chatbots questions like:
In EA Studio, what does the column SQN represent? What should I look for in the Win/Loss category? What about Return/DD?

Experiment #1 — Can the filters you use to select EAs impact Walk-Forward Analysis (WFA) success?
I took the same collection of 233 EAs, the same dataset, and the same tool.
This collection was built early in my EA Studio journey — nothing special at all.

Then I ran it over and over, each time applying a different combination of filters before validation.
After filtering, the remaining strategies were tested on unseen data (WFA).
In total, I ran 91 combinations, each representing one complete backtest + WFA cycle.

The results: (attachment)

Some combinations achieved 100 % WFA success, others dropped below 20 % — all with identical data and strategies (Example: 50 profitable in WFA out of 50 that passed backtest = 100 %.
20 profitable in WFA out of 100 that passed = 20 %.).

The takeaway:

- Even small variations in settings and filters can drastically change the outcome.
- EA Studio lets you run structured learning experiments — showing with data how your choices affect robustness and performance.
- It’s not the real market — but each improvement that reduces failure from backtest → demo → live moves you closer to a sustainable, winning portfolio.

All the best
Vincenzo