My live Gold account currently shows 41 trades.
40 profitable.
1 losing.
Of course I like seeing that.
I spent a lot of time building this system, rewriting parts of it, testing assumptions, and occasionally discovering that an idea I liked yesterday needed more work today.
But every time I look at those 41 trades, one thought matters more than the result:
41 trades are still only 41 trades.
That is not proof.
It is simply a good start.
Good results can be dangerous too
When a system performs badly, developers investigate.
When it performs unusually well, there is a temptation to stop asking questions.
That is where I become cautious.
After more than 20 years in software development, I have learned that when a complex system looks perfect, the useful question is not:
“Why is this so good?”
It is:
“What haven’t I tested yet?”
Trading systems make this even harder.
Normal software has users.
An EA has the market.
And the market does not send bug reports.
The backtest cannot answer everything
Before going live, I tested the combined system over 2,366 trades across nine strategy modules.
Profit Factor: 4.16.
Sharpe Ratio: 8.49.
Recovery Factor: 17.75.
Profitable trades: 82.21%.
Relative Equity Drawdown: 24.21%.
The test used Every Tick based on real ticks, processing more than 120 million ticks.
That sample is far more meaningful statistically than 41 live trades.
But it still cannot tell me exactly what happens tomorrow.
What if execution gets worse?
What if spread expands during a violent Gold move?
What if several strategies independently take the same directional risk?
What if the next market regime looks different from the development period?
And most importantly:
What happens when one of my assumptions is wrong?
That is the question I care about most.
I actually want to see a bad week
Not because I enjoy losses.
I don’t.
But one losing trade tells me almost nothing.
I want to see how the system behaves during an uncomfortable period.
Several losses close together.
Bad execution.
A market where the usual setups stop behaving normally.
A day when the EA does nothing while the chart looks full of opportunities.
Those situations reveal much more about a system than another easy winner.
Recently I tested the EA with spread multiplied by two, three and five.
It survived.
For a moment, that felt reassuring.
Then I realized the obvious problem:
real markets rarely make only one thing worse.
Spread can increase while liquidity deteriorates, execution becomes less predictable and price accelerates.
That is the kind of testing I want to focus on next.
Nine strategies can still make one mistake
The EA contains nine separate strategy modules.
Different entry conditions.
Different timing.
Different trade management.
Technically, nine strategies.
But if three of them go LONG on XAUUSD at the same time, the account does not care how independent the code is.
It only sees exposure.
So now I am looking more closely at questions like:
How often do strategies overlap?
Do they lose together?
Are several different algorithms actually depending on the same market behaviour?
I do not have every answer yet.
That is fine.
For me, unanswered questions are more useful than pretending the system is already fully understood.
I am not planning to disappear after launch
This part matters to me.
I do not want to publish an EA, talk about it for a few weeks and move on.
That is not how I like building software.
People use a product.
They find strange behaviour.
They ask uncomfortable questions.
Sometimes everything works exactly as designed.
Sometimes the user finds something the developer missed.
And sometimes that question becomes the next improvement.
That is the relationship I want to build here too.
Keep testing.
Keep improving.
Publish what I learn.
Fix real problems when they appear.
And make the next version better than the current one.
I see this as a long-term project.
So I am not celebrating yet
The live result looks encouraging.
The backtest looks encouraging too.
I am happy with both.
But I am not willing to turn either into the sentence:
“This proves the EA works.”
Markets do not issue certificates.
They only give us more data.
So I want more trades.
Different regimes.
Real slippage.
Drawdowns.
Bad weeks.
Uncomfortable conditions.
And if something eventually breaks, I would rather understand why than explain it away.
That is also why I keep the live history public.
A live trade cannot be optimized away afterwards.
It happened.
I like that.
It keeps the developer honest.
One question
If you test trading robots yourself, especially Gold EAs, what usually destroys a promising system in your experience?
Execution?
News?
Broker differences?
A regime change?
Strategy correlation?
Something else?
I am collecting ideas for the next tests, and some of them will probably become future posts.
If you want to look at the actual system behind these experiments, it is called Golden Royal. The demo and public live history are available from my MQL5 profile.
Do not judge it by 41 trades.
Test it yourself.
Try uncomfortable settings.
Watch the live account longer.
Ask me a difficult question.
And if you find a weakness I have not found yet, send it to me.
The project is still young.
That may be exactly what makes it interesting to watch.
