A few people asked the right questions under my last post, so let’s put the numbers back into context.

Yes, this is an MT5 demo account.

But the market data itself is not synthetic. The engine processes a real-time CQG feed from the CME futures market. Real prices, real volatility, real order flow, real regime changes.

The fills are simulated.
The market being observed is not.

Now, the obvious limitation:

This track record covers only around three days.

That is nowhere near enough to claim long-term robustness across months, seasons and multiple volatility regimes. Anyone pretending otherwise would be selling smoke in a lab coat.

But there is another dimension to the sample.

The system generated 648 executed trade events during the original statement, through an aggressive scale-in / scale-out architecture.

That is a very large execution sample, even if it remains a very small time and regime sample.

Those are not the same thing.

A system may trade only ten times over three months. Another may produce hundreds of executions in three days.

The first has more calendar history.
The second has more repeated execution events.

Neither dimension is sufficient alone.

The real question is what the sample actually allows us to study.

In this case, it gives us a dense first look at:

execution behavior, fill frequency, scale management, short-term recurrence, exposure time and the way the engine behaves inside detected RANGE regimes.

And exposure time matters far more than most people admit.

From the completed position cycles:

Average holding time: 10 minutes 10 seconds
Median holding time: 1 minute 2 seconds
79% of positions closed within five minutes
Some lasted only a few seconds.
Only a handful remained open for hours.

This raises a genuine risk-management question.

Imagine two systems both producing approximately 1% in one trading day.

System A takes two or three trades, but remains exposed for several hours, potentially through economic releases, liquidity shocks, regime changes and unexpected market events.

System B takes one hundred very short trades, closes most positions within seconds or minutes, and repeatedly returns to a flat, liquid state.

Which system is actually carrying more risk?

The obvious answer is not necessarily the correct one.

A high-frequency system introduces other risks:

more turnover, more fees, more execution dependency, more opportunities for correlated errors and potentially violent inventory accumulation during a failed regime classification.

But a slower system carries prolonged market exposure.

Every additional minute inside a position is another minute during which the market can mutate.

So risk is not only:

How much can I lose?

It is also:

How long am I exposed?
How quickly can I become flat?
How much inventory exists when the regime changes?
How dependent is the result on one continuous market assumption?

This is the philosophical divide.

Would you rather produce the same daily return through:

100 trades lasting seconds or minutes

or

2–3 trades lasting several hours?

One concentrates risk in execution frequency.

The other concentrates risk in time exposure.

And neither is automatically safer.

The real answer probably lives in the interaction between:

frequency, duration, position size, turnover, regime detection and tail-loss control.

That is precisely what the next stage of the research must measure.

The first results are extremely encouraging.

But the purpose of forward testing is not to celebrate the first equity curve.

It is to discover where the machine breaks before the market does it for us.

So what is your risk philosophy?

More trades, shorter exposure, faster return to flat?

Or fewer trades, longer exposure, and more time for each thesis to unfold?

Same target. Completely different risk architecture.

The debate is open. ⚡

https://metaquantuniverse.com/nexus