Over the past ten days we have published, about our own product: that we deleted three features because they cost money, that our backtest changes sign depending on the simulator, that a replica of the system we studied lost 99.99 %, that the broker rejected every one of our close orders, that we spent thirteen hours with open positions unwatched, and that we built an alternative system which also came out mediocre.
That is not the usual product launch sequence. I am closing the series by explaining why it is the one we chose.
A number without its counterweight informs nobody
Our measured configuration gives +$32,272 at a factor of 1.49 over the first half of 2026. It is a real number, measured rigorously, and we publish it.
On its own it means nothing. Because:
- The same configuration loses $6,608 over 2024.
- The same backtest gives −93,797 under the other fill model.
- A 65-configuration sweep produced none that cleared our own validation threshold.
If I show the first and stay quiet about the next three, I have informed nobody: I have selected the window where I look good. That is the operation all trading-system marketing performs, and it is why the sector has the reputation it has.
The counterweight does not weaken the good number. It is what makes it interpretable.
The selfish argument
There is a version of this that does not depend on ethics, and it probably persuades better.
If we sign someone up on false expectations, the problem does not go away: it comes back to us in three months, with an angry client who lost money and is entirely right to be. The cost of that conversation exceeds the revenue from the sale.
And there is a cost most people do not price in: when someone loses money, the marketing material is the first thing reviewed. A page that promised what it could not deliver is not only a reputation problem.
We would rather have fewer clients who understand what they are running. That is not generosity, it is that the other model does not hold.
What honesty does not buy
I want to be precise here, because there is an easy trap: describing your failures with enough eloquence can look like a superior form of selling.
Publishing the bad results does not make the product good. Cerberus is still a high-risk experimental system. It still averages down. Its profit per cycle is still capped while the loss is not. None of these articles changes that, nor is meant to.
What transparency buys is more modest: that you can decide. That you know where the model risk sits, which scenarios the guardian does not cover, and what account size cannot absorb the measured drawdown. With that, you decide rather than us deciding for you.
How it translates into practice
Three rules we apply, and that anyone can copy:
1. No number without its window. A result without the period, the instrument and the fill model is not a result, it is a figure.
2. Measure one thing at a time. Changing three things and watching the total go up hides the one that is costing money.
3. Publish the whole experiment, not the conclusion. Including the runs that refute what you expected. Those are the ones that bound where not to keep looking.
You do not need a financial product for this to apply. It holds for any technical claim you are asking someone to believe.
Full series on building Cerberus at Tuurt Labs. High-risk instrument: trading leveraged CFDs can cost you your entire capital. This is not investment advice.