Section 3. Entry Decision Technology

The neural network architecture generates exclusively probabilities. The final trading decision is made based on a strict mathematical formula.

3.1. Analysis via Model Ensemble

For each identified market model, the system calculates several hundred numerical features: from the proportions of local price movements to cluster affiliation and the context of adjacent models.

An ensemble of specialized AI models (Deep Learning and Gradient Boosting) performs parallel processing of these features. Each model in the ensemble solves a highly specialized task:

1. «Will a trend reversal occur from the resistance level?»
2. «Will the price reach the first target level?»
3. «Can the price reach the second, deeper level?»
4. «Are there signs that the model's potential is exhausted and the trade should be closed immediately?»

The result of each model's work is the probability of a specific event occurring (a number from 0 to 1).

Attestation "passports" of models: Each ML model is equipped with an attestation passport containing evidence-based metrics of its accuracy on historical data. This ensures transparency of probabilistic assessments and eliminates the "black box" effect.

3.2. Expected Value (EV) Filter

The raw probability is integrated into the Expected Value (EV) mathematical filter:

EV = P(win) × Profit − P(loss) × Risk

Where:

• P(win) — the probability of success, verified by the neural network based on the historical accuracy of the model (from the passport) in a similar confidence range.
• Profit — the distance from the entry price to the target level (in capital).
• Risk — the distance from the entry price to the scenario cancellation level (Stop Loss).

Imperative rule: A trading signal is retransmitted to the broker's terminal exclusively under the condition EV > 0.

3.3. Algorithmic Risk Management

In addition to probabilistic assessment, the system implements strict algorithmic capital control. When forming each order, the trade volume is calculated mathematically so that the potential loss upon reaching the scenario cancellation level (Stop Loss) strictly does not exceed the value allowed by the given strategy (e.g., a fixed percentage of the available balance). Risk management is embedded in the entry calculation formula itself and prevents deposit "overload".

---