Astra Neural Engine

Neural Gold Trading. Two Time Horizons. One Engine.

Astra Neural Gold brings a complete neural trading workflow directly into MetaTrader 5: local model training, adaptive market analysis and controlled automated execution — with no Python installation, no DLLs and no external AI subscription.

Built specifically for Gold (XAUUSD), Astra combines two Temporal Convolutional Network branches with a causal Transformer to analyze both short-term and wider historical market patterns. The result is a directional evaluation expressed in net R, designed to connect the model’s analysis to the actual risk and reward structure of each trade.

Key features:

• Dual TCN architecture: analyzes recent market behavior and longer historical context on the same timeframe
• Causal Transformer: combines both temporal perspectives without accessing future bars during inference
• Net R decision output: estimates LONG and SHORT outcomes relative to the initial stop distance
• Fully local training: train and run the model inside MetaTrader 5
• Automated execution: entries, exits, risk controls and position sizing managed by the EA
• No external dependencies: no Python, no DLL, no cloud AI subscription

How the Neural Engine Works

Astra does not rely on a single model or a single time horizon. It uses two parallel TCN branches to read the same chart timeframe from different temporal perspectives.

• Local TCN: focuses on shorter sequences, identifying recent price behavior, volatility conditions and momentum patterns
• Wide TCN: examines a broader historical window, adding longer-term context to the analysis
• Causal Transformer: processes the combined representation and helps the model evaluate relationships across the available sequence

The output layer produces separate LONG and SHORT estimates in net R. The complete architecture — both TCN branches and the Transformer — is trained jointly.

From Neural Analysis to Trading Decisions

Astra is not only a model trainer. It is a complete trading system inside MetaTrader 5.

After training, the EA compares the directional estimates with its configured entry threshold and applies execution and risk checks. It can:

• Enter LONG
• Enter SHORT
• Remain out of the market

The EA does not need to open a trade on every bar. Its goal is to act only when the model output and the configured trading conditions align.

Because the output is expressed in net R, the model’s evaluation is directly related to the trade’s stop distance, rather than being a generic probability score.

Before You Start: Disable ExecutionCheck

To use Astra for neural trading or training, set ExecutionCheck = false in the EA inputs before starting.

When enabled, ExecutionCheck runs a separate execution diagnostic, which may open and close a minimum-volume test trade. Neural trading remains paused during this diagnostic session.

After completing the check, stop the EA or tester, set ExecutionCheck = false, and restart with TRAIN or TRADE selected for your intended workflow.

Train Locally. Keep Full Control.

Use the TRAIN mode in the Strategy Tester to prepare features, targets and train the neural network directly inside MetaTrader 5.

• Select the historical interval using the Strategy Tester dates
• No separate CSV export is required for the standard workflow
• Historical bars and ticks must be available from your broker
• Choose a chronological validation percentage, or train on all eligible samples with validation disabled
• A star beside NEW BEST indicates an improvement in the selected loss metric

TRADE mode uses a saved model with fixed weights. Training and trading are separate operations, so the model does not silently change while the EA is running.

Built-In Execution and Risk Controls

Astra includes practical execution controls designed for a more disciplined automated workflow:

• Volatility-based stop loss and take profit
• Position sizing based on configured risk
• Maximum holding period
• Order volume, margin and exposure checks
• Daily loss and drawdown controls that can pause new entries

These controls are designed to support disciplined execution, but they cannot guarantee a maximum realized loss in the event of price gaps, low liquidity or execution differences.

Commission and slippage assumptions are part of model preparation. The reference configuration models 0.0025% commission per side and two points of slippage per side. Actual broker spread, commissions, swaps and execution conditions can affect live results.

Running backtest Gold Reference Configuration

The documented Gold reference configuration is:

• Symbol: XAUUSD
• Timeframe: M30
• ExecutionCheck = false
  
• Training and historical replay period: 1 January 2025 – 7 October 2026
• Architecture: two TCN branches followed by two Transformer blocks
• Historical data quality: 100% real-tick history

The supplied MetaTrader 5 report covers this period and reports:

• 1,043 trades
• Profit factor: 4.92
• Maximum relative equity drawdown: 1.51%
• Initial balance: EUR 10,000
• Reported net profit: EUR 180,441.84

This is a historical simulation on the training data. It demonstrates the fitted model’s behavior on that specific period and is not an independent test, live track record or future performance forecast. Results on unseen data, different brokers and different market conditions can differ substantially.

Launch Pricing

Launch price: USD 97 for the first 20 paid licenses.

Secure the launch price before the next tier begins. The price will increase as the following sales milestones are reached. Every tier includes the same full product functionality; only the purchase price changes.

License numbers refer to cumulative paid license sales. Price increases apply only to future purchases. Existing license holders do not need to upgrade tiers or pay the difference when the price increases.

Prices apply to future purchases at each threshold. Existing licenses do not require a tier upgrade or an additional payment when the price increases.