Thibauld Charles Ghislain Robin / Profil

This is how every Theory of Edge strategy is built.
This is one strategy from the gold databank, generated in the Theory of Edge Studio.
🔵 Blue = in-sample. The only data the Studio was allowed to learn from.🟠 Orange = out-of-sample. Data the strategy had never seen during its construction, from late 2024 to today.
The search never looks at the orange part. A strategy only reaches the databank if it keeps working on data it has never seen. That is the whole point: not a curve fitted to the past, but a behaviour that holds up after it.
ToE Gold is built from strategies selected exactly this way.
🎁 Enjoy 6 months of the Theory of Edge Studio with any purchase of either algo, and build your own strategies with the same tool. Message me after your purchase to get your access (NQ being updated).

MERIDIAN: BIG UPDATE COMING TOMORROW 🚀

I’ve decided to make a major update to Meridian after discovering something that I honestly didn’t expect.

I found that the structure, with the same logic and the same execution parameters and stop orders, produces strong results when trained on custom bars.

But the important part is the generalization.

In the example below, the dataset covers June 23 → August 14, 2026, with 9,132 custom bars. The model is trained on a little more than half of the data, while the remaining data is kept completely separate as a control / out-of-sample period. **Everything outside the training period, including all data before it, has never been seen by the model.

And the important thing is that the performance doesn't simply look good on the training data, the same model continues to produce a strong equity curve on the completely unseen control/OOS period, even though the training set is very small (+/- 2 months).

That level of generalization is extremely rare.

The log shows the same model running on the custom-bar data, with the TRAIN period clearly separated from the OOS/control period.

After seeing this, I decided that this deserves a much bigger update than I originally planned.
I’ve put the full process in the cloud overnight so I can run everything much faster and properly validate the results across the different configurations.

Get ready for an update tomorrow.

Same proprietary structure as Meridian, but built around a custom made Renko engine with direct market execution.

Higher trade frequency (3K trades a month!!!). It seems the model had no intention of taking it easy.

They say markets are non stationary. The model doesn't seem to care.

Training: 03/31/2026 → 08/14/2026
Backtest: 01/01/2026 → 04/01/2026
Modelling: Every tick based on real ticks

NEAT (NeuroEvolution of Augmenting Topologies), completely rewritten in MQL5 with extreme complexity penalty. Coming soon.

Hat den Code Confluence Index Stoch+RSI+MACD veröffentlicht

Hat den Code Manual Scalping With Keyboard veröffentlicht

Hello everyone,

Following recent adjustments, I want to let you know that I’m currently reverting Epsilon back to its original version, focused exclusively on USD pairs. After attempting to expand the strategy, it became clear that the initial structure was far more coherent and effective. This expansion was a mistake, and I apologize for any inconvenience it may have caused.

Epsilon will be updated and back online within the next few days, fully restored to its original and optimized version.

As for the monitoring, the applied risk was 2% per position, so the current -3% drawdown remains within acceptable and controlled limits.

🙏 Thank you for your patience and understanding. The original strategy will be back very soon.

Have a great day