Riyadh Almalki

Architecture

Expert Advisor 27,600 lines · indicator 18,400 · shared libraries 3,300. Fifteen strategies run in parallel, each isolated by its own magic number, entry logic, exit logic and risk parameters over a shared execution and reporting layer.

Gold (XAUUSD) · Bitcoin (BTCUSD) · German DAX (DE30) · US Nasdaq (USTEC) · Silver (XAGUSD). Nine terminals across a VPS estate and a local rig, synchronised over WireGuard and Syncthing with per-binary deployment and versioned rollback.

Machine learning

ONNX models loaded and evaluated inside MQL5 at signal time, applied as a probability-threshold filter on candidate entries.

k-NN behaviour-recurrence engines match the live 24-feature market state against historical analogues — raw price deliberately excluded to avoid level dependence.

A feature-based direction-prediction model trained on historical M15 data, with a live per-candle prediction-versus-outcome verification log and hit-rate reporting by confidence bucket.

Instrumentation

Forty-three fields recorded automatically per trade — entry, every stop and target level, maximum favourable and adverse excursion, close reason, realised P&L, balance before and after.

A class of repaint and non-determinism defects was identified and eliminated, so the same bar produces the same signal on every terminal — the precondition for any honest backtest.

Measured progression

Win rate across the recorded ledger window rose from 50.0% in the first week to 80.2% and 71.1% in the final full weeks. First thousand closed trades 47.1%; last thousand 66.6%.

Full window: 68.6% of 5,682 recorded closed trades finished in profit across fifteen strategies and six instruments. The strongest automated strategy ran at 74.3% over 2,400 trades.

The output of an iteration loop rather than one release — build V2.0.536, a documented change report per version, every cycle measured against the ledger and strategies revised or retired on that evidence.