YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
MT5 INTEL — Autonomous Cognitive Neural Trading Brain
A fully self‑contained MetaTrader 5 Expert Advisor that operates as a brain‑inspired, hybrid cognitive trading system. Every part of the intelligence — perception, memory, forecasting, reasoning, planning, risk, execution, self‑evaluation and continuous self‑improvement — runs entirely inside MQL5.
No Python · No DLLs · No external APIs · No cloud. 100% MQL5.
The system runs a closed cognitive loop on every cycle:
Observe → Understand → Predict → Plan → Execute → Monitor → Learn → Improve
Table of Contents
- Highlights
- Cognitive Architecture
- Repository Structure
- Requirements
- Installation
- Quick Start
- Trading Modes
- Inputs Reference
- How It Works
- Persistence & Continuous Learning
- Validation & Backtesting
- Verified Results
- Risk Disclaimer
- Roadmap
- License
Highlights
- Brain‑inspired 10‑layer architecture — sensory cortex, pattern cortex, temporal memory, neural forecasting, reasoning, strategic planning, risk intelligence, execution, self‑evaluation and meta‑learning.
- Dual‑mode, manually selectable — high‑frequency Scalping (M1/M5) and Swing/Long, plus an automatic Hybrid that scalps small but holds runners for big moves. Works on any timeframe.
- Full multi‑timeframe awareness — perceives a ladder of 4 timeframes at once so it understands the macro trend and avoids pullbacks / liquidity traps while scalping.
- Ensemble intelligence — 8 independent experts (trend, momentum, mean‑reversion, breakout, volatility, liquidity, neural MLP, adaptive) fused by weighted consensus, Bayesian log‑odds and a Markov regime model. Never relies on a single model.
- Genuine online learning — a trainable neural network, multiplicative‑weights ensemble adaptation, per‑regime Bayesian win priors and a meta‑learner that tunes thresholds, risk appetite and targets from realised outcomes.
- Explainable by design — every decision logs its inputs, sub‑scores, confidence, expected value, macro bias, pullback risk and rationale. No black box.
- Robust risk control — volatility position sizing, drawdown halt, exposure / risk‑budget caps, per‑trade leverage cap, break‑even and adaptive trailing.
- Cross‑session memory — STM/MTM/LTM episodic memory, trade memory and learned weights persist to disk and reload on restart.
- Built‑in validation — Monte‑Carlo, walk‑forward, stress and parameter‑stability tests run automatically on the live trade series.
Cognitive Architecture
| Layer | Module | Responsibility |
|---|---|---|
| 1 | SensoryCortex |
Multi‑timeframe perception → 40‑dimensional Market State Tensor; macro trend; pullback risk; Kalman surprise |
| 2 | PatternCortex |
Detects trend/range/breakout/reversal/accumulation/distribution/liquidity‑sweep/compression; pattern sequence memory |
| 3 | TemporalMemory |
GRU recurrent context encoder + episodic k‑NN recall over long‑term memory |
| 4 | ForecastEngine |
8‑expert ensemble + Markov regimes + Bayesian fusion → probabilistic direction / magnitude / risk / confidence |
| 5 | ReasoningEngine |
Evaluates probability, reward, risk, uncertainty, expected value → explainable decision |
| 6 | StrategicPlanner |
Builds the trade plan: entry, SL, TP, scaling, re‑entry, trailing, break‑even, emergency exit, runners |
| 7 | RiskIntelligence |
Drawdown tracking, exposure & risk‑budget control, volatility + leverage‑capped position sizing |
| 8 | ExecutionEngine |
Order entry/modification, break‑even, trailing, scaling, emergency exit; per‑position registry |
| 9 | SelfEvaluation |
Detects closed trades, computes R‑multiples, accuracy, decision quality and full performance metrics |
| 10 | MetaLearning |
Adapts caution, risk appetite and target multipliers; feature‑importance; prevents de‑risking lock‑up |
Supporting engines (Include/MT5INTEL/Intelligence): trainable MLP + GRU
(NeuralNet), KalmanFilter, BayesianEngine, MarkovModel, Ensemble.
Memory (Memory/MemoryEngine) and Validation (Validation/Backtest,
Validation/Validator). Everything is orchestrated by Brain.mqh.
Repository Structure
MQL5/
├─ Experts/
│ └─ MT5INTEL/
│ └─ MT5_INTEL.mq5 # the Expert Advisor (entry point)
├─ Include/
│ └─ MT5INTEL/
│ ├─ Brain.mqh # cognitive orchestrator (owns all layers)
│ ├─ Core/ # Config, Types, MathUtils, Logger
│ ├─ Intelligence/ # NeuralNet, Kalman, Bayesian, Markov, Ensemble
│ ├─ Memory/ # MemoryEngine (STM/MTM/LTM, persistence)
│ ├─ Cognition/ # the 10 cognitive layers
│ └─ Validation/ # Backtest metrics + Validator battery
└─ Presets/
└─ MT5INTEL_scalp_live.set # ready‑to‑use scalping configuration
docs/
└─ ARCHITECTURE.md # detailed design notes
Requirements
- MetaTrader 5 (build 4000+ recommended; developed/tested on build 5836).
- A symbol with sufficient history on your broker (developed on EURUSD).
- No third‑party libraries, indicators or DLLs.
Installation
- Locate your terminal's data folder: in MetaTrader 5 → File → Open Data Folder.
- Copy the contents of this repository's
MQL5/folder into the terminal'sMQL5/folder, preserving the structure:MQL5/Experts/MT5INTEL/MT5_INTEL.mq5MQL5/Include/MT5INTEL/...MQL5/Presets/MT5INTEL_scalp_live.set
- Open MetaEditor, open
MT5_INTEL.mq5and press Compile (F7). Expected result: 0 errors, 0 warnings.
Headless compile (Windows):
& "C:\Program Files\MetaTrader 5\MetaEditor64.exe" `
/compile:"<DataFolder>\MQL5\Experts\MT5INTEL\MT5_INTEL.mq5" `
/log:"<DataFolder>\MQL5\Experts\MT5INTEL\compile.log"
# Note: the log is UTF-16; check the "Result: N errors, M warnings" line.
Quick Start
- Open a chart for your symbol (for scalping, an M1 chart is ideal).
- Drag MT5_INTEL onto the chart.
- Enable Algo Trading in the toolbar and tick Allow Algo Trading in the dialog.
- (Optional) Load the preset
MT5INTEL_scalp_live.setin the Inputs tab. - The on‑chart HUD shows the live regime, forecast, decision, expert votes, account risk and learning statistics.
Always validate on a demo account and in the Strategy Tester before considering any live deployment. See the Risk Disclaimer.
Trading Modes
Set via the InpTradeMode input:
| Mode | Best for | Behaviour |
|---|---|---|
| Scalping | M1 / M5 | High frequency, tight SL/TP, many small‑pip trades, break‑even fast, strict spread & pullback filters |
| Swing | M15+ | Fewer, larger trades on higher timeframes; may fade pullbacks; wider targets |
| Hybrid | M5 | Scalps small by default but converts strong, macro‑aligned setups into runners (wide TP + trailing) to hold for big moves |
InpBaseTF selects the execution timeframe (or PERIOD_CURRENT = auto by mode). The EA
always builds a 4‑rung timeframe ladder above the base so it keeps full macro context —
for example scalping on M1 still watches M1 / M5 / M15 / H1.
Inputs Reference
Trading Mode
| Input | Default | Description |
|---|---|---|
InpTradeMode |
Scalping | Scalping / Swing / Hybrid |
InpBaseTF |
Current | Execution timeframe (Current = auto by mode) |
Risk Intelligence
| Input | Default | Description |
|---|---|---|
InpRiskPerTrade |
0.01 | Risk per trade (fraction of equity) |
InpMaxRiskBudget |
0.06 | Max aggregate open‑risk fraction |
InpMaxDrawdownStop |
0.30 | Halt new trades beyond this drawdown |
InpMaxPositions |
4 | Max concurrent positions |
InpMaxTradesPerDay |
0 | Daily entry cap (0 = unlimited) |
InpMaxLeverage |
3.0 | Cap nominal exposure to N× equity per trade |
Decision
| Input | Default | Description |
|---|---|---|
InpMinConfidence |
0.52 | Minimum confidence to act |
InpMinEdge |
0.02 | Minimum expected‑value edge (R) |
InpMaxSpreadATR |
0.12 | Scalp: skip when spread/ATR exceeds this |
Behaviour
| Input | Default | Description |
|---|---|---|
InpAtrSLMult / InpAtrTPMult |
1.8 / 2.6 | Swing stop / target (ATR units) |
InpScalpSLATR / InpScalpTPATR |
0.9 / 1.1 | Scalp stop / target (ATR units) |
InpBeTriggerATR |
0.6 | Move SL to break‑even after this profit (ATR) |
InpLetRunners / InpRunnerTPATR |
true / 4.0 | Allow runners and their wide target |
InpAllowTrading |
true | Master execution switch |
InpOnBarOnly |
true | Decide once per exec‑TF bar (false = every tick) |
Learning & System
| Input | Default | Description |
|---|---|---|
InpLearnRate / InpMetaLearnRate |
0.02 / 0.05 | Neural / ensemble learning rates |
InpValidateEvery |
20 | Run the validation battery every N closed trades |
InpMagic |
20260623 | Magic number |
InpPersist |
true | Persist learning across sessions |
InpVerbose |
false | Verbose logging |
InpTimerSeconds |
5 | Management timer interval (seconds) |
How It Works
On each decision cycle the brain:
- Observes the market across four timeframes and builds a normalised Market State Tensor (returns, momentum, trend, volatility, structure, liquidity, range position, plus macro trend, multi‑timeframe alignment and pullback risk).
- Understands the state — classifies the regime (softmax posterior + entropy), detects the dominant pattern and updates recurrent + episodic memory.
- Predicts direction, magnitude, volatility, risk and confidence via the ensemble.
- Reasons about probability, reward, risk, uncertainty and expected value, applying the macro‑trend and pullback guards, then emits an explainable decision.
- Plans a complete trade (entry, SL, TP, sizing, scaling, break‑even, trailing, emergency exit; runner vs. quick‑scalp).
- Executes and monitors the position with trailing, break‑even and scaling.
- Learns from every closed trade — trains the network, adapts ensemble weights and Bayesian priors, and the meta‑learner tunes caution, risk appetite and targets.
Every entry is logged with its full rationale, e.g.:
SELL | P(up)=0.34 edge=0.31 conf=0.50 EV=0.51R pWin=0.68 risk=0.45 unc=0.59 |
macro=-0.91 pull=0.18 spr/atr=0.00 | regime=1(0.63) pat=1(str0.69) |
trend=-1.00 mom=-0.23 mr=0.62 neural=-0.15 | mode=SCALP caution=1.00 -> SELL
Persistence & Continuous Learning
When InpPersist is enabled, the brain saves its full state — adaptive normalisation,
GRU/MLP/Markov/Bayesian/ensemble weights, STM/MTM/LTM memory, trade memory and
meta‑learning parameters — to MQL5/Files/MT5INTEL/<symbol>_<magic>.intel and reloads it
on restart, so it keeps improving across sessions.
Validation & Backtesting
A validation battery runs automatically every InpValidateEvery closed trades and logs:
- Monte‑Carlo bootstrap (final‑equity percentiles, max‑drawdown, probability of profit / ruin),
- Walk‑forward consistency across segments,
- Stress test (spread/slippage haircut + loss amplification),
- Parameter‑stability (jackknife), and
- a per‑regime breakdown.
OnTester() returns a robustness‑weighted objective for Strategy‑Tester optimisation.
Verified Results
Built and tested with MetaTrader 5 (MetaEditor build 5836) on EURUSD. Compiles with 0 errors / 0 warnings. Representative Strategy‑Tester runs (1‑minute model, persistence off, generic untuned parameters):
| Scenario | Trades | Notes |
|---|---|---|
| Scalping, M1, $100k, 3 months | 2027 | 0 runtime failures; max lot 1.99 (2× leverage cap); +0.27% |
| Hybrid, M5, 7 months | 97 | 58 runner trades; 0 failures |
| Swing, M15, 12 months | 5 | +0.31%; conservative by design |
The EA was also verified live on a demo account (real‑time ticks): it connected, attached to an M1 chart, perceived live data and executed trades with full, explainable rationale.
Honesty note. With generic, untuned parameters the strategy is roughly break‑even — the goal of this project is the cognitive architecture, dual‑mode high‑frequency engine, multi‑timeframe awareness, risk control and online learning, not a pre‑optimised money printer. Sustained profitability requires per‑symbol/per‑broker tuning via the built‑in walk‑forward tooling and
OnTester()optimisation. Past performance does not guarantee future results.
Risk Disclaimer
This software is provided for research and educational purposes. Trading foreign exchange, CFDs and other leveraged instruments carries a high level of risk and may not be suitable for all investors. You can lose some or all of your capital. Always test on a demo account first. Nothing in this repository constitutes financial advice. You are solely responsible for any use of this software and any resulting trades. The authors accept no liability for any losses incurred.
Roadmap
- Per‑symbol parameter optimisation profiles.
- Portfolio mode (multi‑symbol shared risk budget).
- Optional news / session filters.
- Extended pattern library and additional ensemble experts.
License
Proprietary. See LICENSE. © 2026 root975638‑alt. All rights reserved.