1. Data source and flow
Market data comes from MetaTrader 5. Among other inputs, the system analyses open, high, low and close prices, volume, the current market price, price dynamics, selected technical indicators and signals generated by machine-learning models. The analysis is conducted on the M30 timeframe, in which each candle represents 30 minutes.
Data from MetaTrader 5 is passed to a local JupyterLab environment, where the models and decision strategies run. Their output is returned to the Expert Advisor in MT5, which compares neural-network estimates with technical-indicator signals. When the required conditions are met, the EA decides whether to place or reject a trade and publishes the current status to the web panel.
Process: MetaTrader 5 → market data → AI models and strategies → filters → BUY, SELL or NO TRADE → execution or rejection → publication on the portal.
2. Artificial-intelligence models
The models do not interpret the market as a human would. They process numerical descriptions of price action and indicator behaviour, then return a forecast or quality assessment. Each output is a probabilistic estimate, not a certain account of what will happen next.
- Direction_CNN1D_Net — estimates whether price is more likely to move higher or lower.
- Range_MLP_Net — estimates the possible favourable move, potential drawdown and prospective reward-to-risk ratio, shown as RR.
- Filter_CNN1D_Net — assesses whether current market conditions meet the quality threshold required before a position can be considered.
Later versions may also use models designed to detect changes in direction, measure agreement between signals or block higher-risk set-ups. Not every model visible in the research environment necessarily influences a live trade.
3. Training data
Datasets covering 2000 to 2026 were prepared for the EUR/USD project. Each row describes one M30 candle and includes OHLCV data, technical indicators and features drawn from the six preceding 30-minute periods. Depending on the dataset version, this produces nearly 300 variables and close to one million training rows.
A large dataset does not remove the risk of error. A model may learn relationships that are specific to the historical sample and fail to recur in live markets. Backtests and historical results therefore offer no guarantee of live performance.
4. Decision strategies
A single neural-network output is not necessarily enough to open a position. A strategy may require agreement across several independent elements: direction, an acceptable Range RR reading, a passed quality filter, confirmation from technical indicators and no exclusion signal.
- ANALOG — direction derived from the agreed set of technical indicators running on the MT5 chart.
- DIRECTION — directional estimate from a network trained on price data and technical features.
- RANGE — estimate of the possible price range, take-profit level, drawdown and reward-to-risk ratio.
- FILTER — assessment of whether current market conditions exceed the required quality threshold.
A strategy can return BUY, SELL or NO TRADE. NO TRADE is a substantive result: the system did not find enough confirmation to open or reverse a position.
5. Trade execution by the EA
After receiving an approved signal, the Expert Advisor compares it with the active technical logic and runs a series of safety checks. These include whether a system position already exists, whether the signal is current, whether the analytical module is connected, whether the required filters have passed and whether user settings block the trade.
The software also includes safeguards designed to reduce accidental re-entry after start-up, manual closure or a change in platform state. Its final behaviour depends on the EA version and the settings in force during the relevant observation period.
6. Position size, TP, SL and trailing functions
Position size may be determined automatically from account equity, leverage and software limits. During some test periods, the system may commit a very high proportion of available margin. This is an exceptionally high-risk experimental setting on a specific account, not a recommended approach to capital management.
The take-profit level may be drawn from the Range MLP forecast and updated if a later estimate points to a wider potential move. The stop-loss level may be set numerically or left unset, depending on the configuration under test. Additional trailing functions may respond to sharp price moves, protect an accrued result or, under specified conditions, close and reverse a position. Protective parameters may therefore differ from one trade to another and may change after entry.
7. What the LIVE panel shows
The live panel presents the current status of the EUR/USD system. It may show the position and direction, entry time and price, current market price, result in points, available stop-loss and take-profit levels, and the time of the latest valid reading. The chart marks the opening or closing point of the position.
When there is no active position, the panel may show the most recently closed trade or the message “No open position”. This distinguishes a flat market status from a loading error.
8. History and charts
Once a position closes, the system writes a record to a dedicated EUR/USD history table. The record may include entry and exit times, direction, prices, result in points, maximum drawdown, cumulative result, closing reason and opening/closing charts. The line chart starts at zero and tracks the cumulative result; the cylinder chart shows each trade separately.
Images supplement rather than replace the numerical record. Because charts and data travel through separate technical channels, they may be saved at slightly different times.
9. Data updates
The panel refreshes automatically, normally every two seconds. The displayed time marks the latest valid reading. If the platform, local computer, server or analytical module loses its connection, the portal may temporarily show the last available status, a waiting message or no data.