Iconic Kybernetic AI

ICONIC KYBERNETIC AI+ — OMNI-NEXUS CYBERNETIC CORE

One chart. Two markets. Four core technologies. Five adaptive learning layers.

Most Expert Advisors trade a single symbol and call themselves intelligent. A handful attempt two symbols at once and call themselves a portfolio system. ICONIC KYBERNETIC AI+ runs a four-technology cognitive architecture that treats Bitcoin and Gold as a unified, causally linked system — continuously reading the information flow between them, distributing capital accordingly, and adapting its own learning behaviour over time.

Drop it on any single chart. From that moment, the OMNI-NEXUS CYBERNETIC CORE takes over. Both BTCUSD and XAUUSD are monitored, analyzed, and traded simultaneously, with two isolated AI brains managing execution and a cross-asset intelligence layer governing everything above them. No second chart, no manual split, no configuration ritual.

THE UNDERLYING TRADING STRATEGY

Strip away the cognitive core and the reinforcement learning, and ICONIC KYBERNETIC AI+ is built on a structural trend-following breakout strategy — the same category of approach a discretionary price-action trader would use, executed mechanically on two markets at once. The AI layers do not invent trade ideas out of nothing; they decide how aggressively, how often, and at what size to act on this underlying structure.

The core logic shared by both engines:

• Trend filter first. No trade is considered against the prevailing trend. BTC uses a 100-period H1 EMA; Gold uses a 50-period H1 EMA combined with a 100-period H4 EMA for higher-timeframe confirmation, so a short-term Gold setup must also agree with the dominant 4-hour structure before it becomes eligible.
• Entry only at structural levels, never mid-air. Pending stop orders are placed at previous-day high/low, current-day high/low, support/resistance zones built from a rolling lookback window, and order-block boundaries — never at an arbitrary price simply because an indicator crossed.
• Breakout confirmation, not breakout assumption. A level must be cleanly broken: the system checks recent price history near the level and candle strength, and on Gold explicitly screens out liquidity-sweep wicks that fake a breakout and reverse, so the engine is not repeatedly caught by stop-hunts.
• Volatility regime awareness. Both engines actively avoid dead or choppy conditions. BTC scores recent range compression and price oscillation and stands down when the market is structurally directionless; Gold detects compressed-range conditions and requires either a strong breakout candle or higher-timeframe alignment before acting in them.
• Minimum distance-to-target filter. A breakout level must sit far enough from current price, measured in ATR, before it is considered tradeable — removing setups where the realistic reward no longer justifies the risk.
• Session-aware. BTC trades across a broad multi-hour window to capture global Bitcoin liquidity; Gold is restricted to the London/New York overlap, the window with the deepest XAUUSD liquidity and tightest realistic spread.
• ATR-scaled stops and targets. Every stop loss and take profit is sized dynamically off the current Average True Range rather than a fixed point value, so risk per trade adapts to current volatility instead of staying static through calm and violent markets alike.

This structural skeleton is what the AI operates on. The reinforcement-learning engine, the adaptive layers introduced in version 1.50, and the OMNI-NEXUS core do not replace this base strategy — they decide, trade by trade, whether to act on a structurally valid setup, how much confidence to assign it, and how to size it, learning from the outcome over time. A structurally weak signal stays a structurally weak signal regardless of how the AI layers are configured; the system is designed to refine execution around a real, pre-existing edge, not to manufacture one where none exists.

VERSION 1.50 — ADAPTIVE LEARNING UPDATE

This version adds five new adaptive learning layers on top of the original OMNI-NEXUS core, plus a live on-chart learning feed. All new layers are individually switchable and default to off or neutral, so existing setups behave exactly as before until a layer is deliberately enabled.

Adaptive Conformal Inference (ACI) — the confidence gate threshold recalibrates itself online so the realised error rate converges toward a target value. The gate becomes stricter after a string of losses and relaxes again once the engine is winning reliably, instead of running on a single fixed number forever.

EWRLS forgetting factor — the reservoir's online readout (see Liquid State Machine, below) can apply an exponential forgetting factor so it stays adaptive to the current market regime rather than slowly freezing as more historical data accumulates.

Welford online normalisation — every learning feature is standardised on the fly using a numerically stable running mean and variance, keeping the reinforcement-learning and meta-labeling layers on a consistent scale as market conditions shift.

Meta-Labeling size control — a second, independent model estimates the probability that a given setup will win and smoothly scales the position size to that probability. It does not invert or override the primary trade direction; very low-probability setups are skipped outright, high-probability setups are sized up.

Experience Replay — closed trades are stored in a ring buffer and replayed in small batches after every new trade closes, so each real trading outcome is learned from multiple times instead of once. This improves data efficiency on instruments that only produce a handful of trades per day.

AI Learning Feed — an optional scrolling text panel next to the dashboard streams real learning events as they happen: trade closes with their realised reward and updated win rate, new pending-order armings with the AI's confidence score, and causal-direction changes between BTC and Gold. It is a live read-out of what the engine is actually doing, not a cosmetic animation.

THE OMNI-NEXUS CYBERNETIC CORE

Four technologies run simultaneously inside MetaTrader 5 RAM, built with native MQL5 matrix and vector types — no DLLs, no external connections, no third-party dependencies. Every trade decision on both instruments passes through this layer before reaching the market.

I. Causal AI — Transfer Entropy Gate. Correlation is symmetric and tells you nothing about direction of influence. Transfer Entropy is asymmetric: it measures how much knowing the past of one asset reduces uncertainty about the future of the other, beyond what that asset already predicts about itself. The core computes directed Transfer Entropy between BTC and Gold log-returns across a configurable lookback window, with a minimum threshold before causation is accepted as real. When one asset is detected as the causal driver, the other engine's correlated position sizing is adjusted accordingly; when no direction clears the threshold, neither engine receives a false signal.

II. Liquid State Machine — Echo-State Reservoir. A compact sparse random reservoir (64 neurons by default, configurable) acts as a temporal memory layer. Market data flows through a leaky integrator network with a fixed spectral radius kept below the critical value of 1.0, which gives the network the echo-state property: it retains temporal memory of incoming signals without diverging into noise amplification. An online ridge-regression readout, trained by recursive least squares, maps the reservoir state to a real-time portfolio stress scalar between 0 and 1, which is consulted before every trade and applies a direct confidence penalty when stress is elevated.

III. PINN Margin Axiom — Physics-Informed Barrier. Before any order is transmitted, the system projects the combined margin of the intended lot pair against current free margin. If the projection exceeds a configurable cap (35% of free margin by default), the order size is reduced or discarded — no confidence score overrides this. The EA's own position sizing cannot be the cause of a margin call; gaps and slippage from external events remain a broker-side risk, as with any EA.

IV. Stochastic Tunneling — Nash Budget Allocation. Each cycle, the system splits the available risk budget between BTC and Gold by solving a game-theoretic allocation problem with simulated annealing and a Stochastic Tunneling transform. The annealer searches for the split where neither engine can improve its expected risk-adjusted return by taking budget from the other, flattening local minima so the solver does not lock onto a suboptimal fixed split the way a simple gradient method would.

TWO ISOLATED AI BRAINS THAT LEARN OVER TIME

Beneath the OMNI-NEXUS layer, two independent execution engines run in parallel, each with its own Q-value table, feature vector, trade history, and persistence file written to disk at the end of every session.

The BTC Engine reads 8 market features per signal cycle: trend direction, trend strength, volatility regime, trading session, daily bias, drawdown fraction, ATR expansion state, and spread quality. It trades across a broad session window to capture global Bitcoin liquidity.

The Gold Engine reads 20 market features, including compressed-market detection, liquidity-sweep identification, higher-timeframe alignment via a dedicated H4 EMA, and candle-quality filters. It trades within the London/New York session overlap, the highest-liquidity window for XAUUSD, with a substantially tighter spread cap than BTC.

Neither brain resets on a terminal restart unless you explicitly enable that option. Every closed trade updates the Q-values, eligibility traces, and (where enabled) the adaptive layers above. The system accumulates trading experience across weeks and months of live operation, not just within a single session.

THE BOLTZMANN REINFORCEMENT LEARNING SYSTEM

Each engine selects from five discrete actions on every signal evaluation: SKIP (confidence below the minimum threshold — no trade), DEFENSIVE (reduced sizing under elevated risk), NORMAL (baseline sizing), CONFIDENT (increased sizing on strong signal convergence), and AGGRESSIVE (maximum sizing on high conviction across all active filters).

Selection uses Boltzmann probability sampling rather than a hard arg-max, so the favoured action is not guaranteed every time — this controlled randomness prevents the system from locking onto a repetitive pattern that stops working when conditions shift. After each trade closes, rewards propagate backward through eligibility traces, crediting state-action pairs that contributed to the outcome several trades earlier. The AI veto is active by default on both engines: when computed confidence falls below the gate, the system skips the trade regardless of what the structural signal says.

RISK MANAGEMENT ARCHITECTURE

There is no grid, no martingale, and no averaging down into a losing position. These are not disabled features — they do not exist in this codebase.

Every trade on both engines opens with a hard stop loss calculated before the order is placed, sized dynamically off ATR. The take profit uses a configurable reward-to-risk ratio with hard maximum caps per symbol, so every trade starts with positive expected structure before the market has moved a single tick. Break-even activation and a stepped trailing stop lock in profit incrementally as a position develops. On the Gold engine, a loss-streak hard brake suspends the engine entirely for a multi-hour cooldown after a configurable number of consecutive losses, rather than continuing to trade through an adverse regime.

PORTFOLIO HARMONIZATION — THREE-TIER PROTECTION

The cross-engine coordination layer monitors combined equity drawdown in real time and enforces an escalating response: Tier 1 (1.5% intraday drawdown) forces both engines into defensive sizing with raised confidence thresholds. Tier 2 (3.0%) blocks new pending orders on both symbols while existing positions continue under full trailing management. Tier 3 (5.0%) closes every open position on both symbols immediately and suspends the EA for a multi-hour cooldown.

The layer also applies correlation-aware sizing when BTC and Gold trend in the same direction simultaneously, and enforces a global risk budget across both engines as a percentage of equity — set automatically by the AI Mode preset (Aggressive, Moderate, Safe) or manually in Custom mode. A global loss-streak counter across both engines triggers a portfolio-level cooldown after a configurable number of combined losses, regardless of which instrument they came from.

STRUCTURAL ENTRY METHODOLOGY

ICONIC KYBERNETIC AI+ does not enter on indicator crossings. It reads market structure and places pending orders at levels that matter:

• Previous-day high/low levels recalculated each session
• Support and resistance zones from a configurable lookback window, independently tuned per symbol
• Order-block detection with pending entries at block boundaries
• Breakout confirmation filters that reject false breakouts and require clean closes beyond key levels
• Liquidity-sweep detection on Gold that identifies stop-hunt wicks and repositions entries accordingly
• Minimum ATR-distance filters preventing entries too close to current market for meaningful risk-to-reward
• Higher-timeframe alignment on Gold via H4 EMA, ensuring short-term entries agree with the dominant structure

Pending orders are refreshed on independent schedules per symbol, with stale orders deleted and replaced once they age past a configurable limit.

INTEGRATED NEWS FILTER

The EA reads economic calendar data directly from MetaTrader 5's native calendar engine — no third-party feed, no API key. High-impact USD events are filtered by default, with trading paused before and after each event on a configurable window. Both engines share one unified news gate, so a single news event blocks BTC and Gold together without separate configuration.

LIVE DASHBOARD AND AI LEARNING FEED

The on-chart dashboard updates in real time and shows the current AI action and confidence for both engines independently, active drawdown-tier status, the next scheduled news event with countdown, open-position metrics for both symbols, OMNI-NEXUS core state including the current causal direction between BTC and Gold, and today's accumulated closed profit across both engines. The optional AI Learning Feed streams the underlying learning events as text, for anyone who wants to see exactly what the engine is doing rather than just a summary number.

RECOMMENDED VALIDATION METHOD — TWO BACKTEST PASSES

Several of the adaptive layers in version 1.50 only reach their effective operating state after a meaningful amount of trade history: feature normalisation settles within roughly the first 5–30 trades, the experience-replay buffer needs to fill before it contributes fully, meta-labeling stays neutral until its warmup-trade count is reached, and the conformal gate threshold recalibrates gradually over time. A single backtest run starting from a freshly reset brain therefore mixes the system's cold-start behaviour with its trained, steady-state behaviour into one result.

To see the two separated, we recommend running the same period twice:

• Pass 1 — cold start. Run the backtest with the brain-reset input enabled, so both engines start from zero learned experience. This is the baseline you would see on an account opened on day one.
• Pass 2 — trained continuation. Run the identical period and settings again, this time with the brain-reset input disabled, so the engines continue from the brain file saved at the end of Pass 1. This reflects how the system performs once it has already accumulated trading experience.

Comparing the two results isolates the learning effect from the underlying strategy edge. A backtest is still a finite, bounded environment and cannot fully replicate months of live forward learning, but the two-pass comparison gives a direct, repeatable read on whether the adaptive layers are measurably changing outcomes rather than sitting idle.

RECOMMENDED BROKER AND ACCOUNT SETUP

This is not a soft suggestion — it directly determines whether the EA can function at its intended parameters.

• Minimum account balance: 500 USD, ideally 1,000 USD or above, so the risk-based lot sizing has meaningful room across both engines simultaneously.
• Leverage: 1:500 recommended. Lower leverage restricts entry frequency and may cause valid setups to be skipped by the PINN margin barrier.
• Spread: as low as possible. Gold has a tight maximum spread cap; BTC's cap is wider but still enforced. Wide spread conditions during off-hours will cause the EA to skip entries rather than fill at poor prices.
• Commission: zero preferred; if charged, keep the round-turn cost low relative to per-trade risk, since commission compounds across multiple trades per day per engine.
• Required symbols: BTCUSD and XAUUSD, both active in Market Watch before attaching the EA. Both engines initialise at startup or neither does.
• Platform: MetaTrader 5 only — not compatible with MetaTrader 4.
• VPS required for continuous operation. Trailing-stop, break-even, and pending-order refresh cycles need the terminal online without interruption.

The EA attaches to any chart symbol and manages both BTCUSD and XAUUSD regardless of which chart it sits on. Before going live, set the Market Validation Mode input to FALSE — the default TRUE value is required only for the MQL5 Marketplace submission process.

RISK NOTICE

Trading Bitcoin, Gold, and other financial instruments involves substantial risk of financial loss. Past performance, whether in backtesting or on live accounts, does not guarantee future results. No algorithmic system can eliminate the inherent uncertainty of financial markets or protect against extraordinary events including broker insolvency, extreme slippage, connectivity failure, or systemic market disruption.

ICONIC KYBERNETIC AI+ contains no grid logic, no martingale mechanism, and no position averaging. Every trade opens with a defined stop loss. These design decisions reduce certain categories of catastrophic risk but do not eliminate trading risk. You are solely responsible for the capital you deploy and the parameters you configure. Only trade with funds you can afford to lose entirely.