Iconic HullX AI
- 지표
- 버전: 2.0
- 업데이트됨: 26 6월 2026
- 활성화: 10
ICONIC HULLX AI v2.0 | NEUROCORE EDITION ICONIC HULLX AI v2.0 — NEUROCORE 2.0 EDITION
Hull Suite × Volatility-Band Squeeze × 5-Layer AI Intelligence Stack
ICONIC HULLX AI v2.0 is a professional-grade signal indicator for MetaTrader 5, built around the Hull Moving Average suite and extended by five independent, mathematically rigorous AI filter gates. Every signal that reaches the chart has passed through a multi-layer intelligence stack that learns in real-time from your actual trade outcomes — no static rules, no fixed thresholds that go stale.
The system is designed around one principle: a signal that cannot be confirmed by multiple independent statistical layers does not appear. Each gate targets a different source of false positives. Together they form a cascading noise filter that adapts to changing market regimes without external configuration or manual intervention.
THE INTELLIGENCE STACK — 5 INDEPENDENT FILTER GATES
All five gates sit before the NEUROCORE ensemble decision and are individually toggleable. This means you can run them in HUD-only mode first, observe which gates fire most frequently on your symbol, and then activate vetoing selectively. Each gate logs its decisions when Debug_SignalLog is enabled, so you always know exactly why a signal was suppressed.
1. Diagonal S6 Mamba State-Space Model (SSM Gate)
A linear state-space model with 16 hidden channels whose system matrix is strictly diagonal and stable. Because the matrix is diagonal, the continuous-to-discrete Zero-Order Hold transform reduces to a pure element-wise exp() — no matrix exponential, no matrix inversion, no NaN paths over any lookback horizon. Each hidden channel is an independent first-order IIR pole spread across multiple time constants, giving the system a multi-horizon view of the feature stream in O(N) linear time.
The hidden state h[t] is persistent across ticks. A new bar is a single 16-dimensional update — never a history rescan. Online-learned cluster centroids (exponential fading memory) define the geometry of historically winning market contexts. When the current context vector lies outside every learned winning cluster by more than the tolerance radius, the gate blocks the signal.
The gate is permanently permissive until it has accumulated enough real winning trades to learn from. It never vetoes on an empty model.
2. RLS Adaptive Feature Weighting
A Recursive Least Squares filter with a bounded forgetting factor (λ clamped to [0.97, 0.999]) that learns, purely in RAM, which of the six market features carry predictive signal in the current regime and which are noise.
The covariance matrix P is trace-bounded on every update to prevent the classic RLS windup divergence. Weights are applied mean-normalised — the total signal energy into the SSM is preserved, but the distribution shifts: a predictive feature is amplified (up to 2.5×), a noise feature is attenuated (down to 0.2×). In a trending regime the Efficiency Ratio carries weight; in chop the RSI is suppressed. The rebalancing happens automatically within a handful of resolved trades.
A minimum sample gate (default: 25 resolved trades) holds the flat default weighting until the filter is statistically trusted. Day-one behaviour is identical to a standard indicator.
3. Koopman-Style Nonlinear Feature Lifting
The six raw market features are blended with a set of fixed nonlinear observables — interaction products and squared terms — that expose market states that look identical in the linear feature space. For example:
- Efficiency Ratio × Hull Momentum — trend conviction (strong trend with rising hull)
- ADX Slope × Volatility Expansion — breakout energy
- RSI × Hull Momentum — overbought/oversold alignment with trend direction
- Volatility Expansion − Body Size — range expansion without candle body (noise flag)
The lifting is dimension-preserving (still 6 inputs into the SSM), controlled by a single KOOP_Mix parameter (default 0.35 = light blend), and carries zero SVD, zero per-tick matrix operations. O(N) cost and NaN-safety are fully preserved.
4. Momentum-Exhaustion Band Filter
A targeted mean-reversion veto at the volatility band edges. When a signal fires with the entry already pressed deep into the outer band and the Hull momentum slope has decayed below a floor, the move is statistically more likely to reverse than to extend. These are precisely the signals that the Hull suite generates at the tail end of a trend and that produce the highest ratio of late entries.
The filter checks two conditions simultaneously: band penetration fraction (≥85% of the way to the outer band) and normalised momentum slope (below configurable floor). A genuine momentum breakout, where the slope is still rising, passes untouched.
5. Cross-Timeframe Wilson Hit-Rate Gate
Each indicator instance writes its live TP1 win/loss tally to a shared slot in the MetaTrader Common folder, keyed by symbol and timeframe. At every signal, the instance reads all peer slots for the same symbol and compares hit rates — not raw percentages, but Wilson score lower bounds at 95% confidence.
The Wilson lower bound is the statistically correct pessimistic estimate of the true win rate given a finite sample. Its critical property: a timeframe with only 4 trades and a raw rate of 75% receives a Wilson lower bound of just 30%, which cannot beat a timeframe with 100 trades and a raw rate of 54% (lower bound 44%). Small samples can never trigger a veto.
A signal is suppressed only when all three conditions are met simultaneously:
- Both the current TF and the best peer TF have ≥ TFG_MinSamples resolved trades
- The peer's Wilson lower bound exceeds the current TF's lower bound by ≥ TFG_MarginPP (default: 10 percentage points)
- The peer's data was written within the last TFG_StaleHours (default: 72 hours)
If any condition is not met — insufficient data, margin too small, stale slot — the signal passes through. The gate is permanently permissive when running on a single chart.
CROSS-TIMEFRAME SETUP GUIDE
To activate the Cross-Timeframe Hit-Rate Gate, attach ICONIC HULLX AI v2.0 to four charts of the same symbol simultaneously:
| Chart | Timeframe | Role |
|---|---|---|
| Chart 1 | M5 | High-frequency signal source, accumulates short-term statistics |
| Chart 2 | M15 | Intermediate confirmation layer |
| Chart 3 | M30 | Intermediate confirmation layer |
| Chart 4 | H1 | Structural context layer, typically highest hit rate |
Each instance operates fully independently and shares statistics automatically. No manual data transfer, no external tools. All inter-instance communication happens via a single binary file in the MetaTrader Common folder (ICONIC_HULLX_TFSTATS_v1.bin by default). The gate activates as soon as each timeframe has accumulated the minimum required number of resolved trades — until then, every instance runs permissively without suppressing any signal.
Single-chart users: The gate is permanently permissive with no peer data. All other four gates remain fully active. No setup required.
NEUROCORE 2.0 ENSEMBLE ENGINE
The five precursor gates feed into the existing NEUROCORE ensemble, which has not been altered:
- TD(λ) Q-Table with 5-action Boltzmann-Softmax policy and heuristic priors (no cold-start random behaviour)
- Online Logistic Regression with per-regime weight separation (Trend / Chop / Volatile)
- Pattern kNN with cosine similarity matching against the historical signal ledger
- PAVA Isotonic Calibration — monotone score-to-TP1-rate mapping built from realised outcomes
- ATR-Adaptive SL / TP — all exit levels scale with current volatility, never fixed points
- AI-Adaptive Entry Nudge — entry offset adjusted by ensemble conviction score
The raw ensemble score gates signals (not the calibrated score), preserving the full spread needed for selective filtering.
FULL FEATURE LIST
- Hull Moving Average suite: HMA, EHMA, THMA variants with adaptive length (ER-driven, EMA-smoothed, step-clamped)
- Keltner Channel + Bollinger Band squeeze detection with squeeze counter
- Adaptive volatility bands (ATR × ER-weighted multiplier, inner + outer)
- Buy / Sell arrow signals with configurable bar-close or intrabar trigger
- Full signal ledger with 7 / 30 / 90-day TP1 and TP2 win-rate statistics
- Per-symbol / per-timeframe brain files (AI state, RLS weights, SSM clusters, TF stats)
- Native MetaTrader 5 economic calendar news filter (USD, EUR; configurable)
- Push notifications with Symbol / Direction / Entry / SL / TP1 / TP2
- Cyberpunk HUD with sine-wave animation, live AI score, gate status
- Market profile presets: Forex Majors, Gold / Metals, Crypto, Indices
- All AI learning and state persistence fully in RAM during live session — disk writes are time-throttled (default: every 4 hours) and only occur on trade close or terminal shutdown
- Compatible with MetaTrader 5 Strategy Tester (including visual mode)
INPUT PARAMETERS OVERVIEW
| Hull Suite | |
|---|---|
| Hull_Variant | HMA / EHMA / THMA |
| Hull_Length | Base period (default: 55) |
| Adaptive_Hull | Enable ER-driven adaptive length |
| Intelligence Stack Gates | |
| SSM_Enable / SSM_Gates_Signals | Mamba SSM gate on/off, veto or HUD-only |
| RLS_Enable / RLS_Lambda | Adaptive feature weighting, forgetting factor |
| KOOP_Enable / KOOP_Mix | Nonlinear lifting blend (0.0–1.0) |
| EXH_Enable / EXH_PenetrationFrac | Exhaustion filter, band penetration threshold |
| TFG_Enable / TFG_MinSamples / TFG_MarginPP | Cross-TF gate, min samples, margin in pp |
| NEUROCORE AI Engine | |
| AI_Enable | Master AI switch |
| AI_Signal_MinScore | Minimum raw ensemble score to pass a signal |
| AI_AdaptiveExits | AI-computed SL / TP1 / TP2 |
| AI_UseEnsemble | Enable full multi-model ensemble |
| Signal Filtering | |
| ADX_MinLevel | Minimum ADX for trend confirmation |
| RSI_LongMin / RSI_ShortMax | RSI alignment filter |
| UseNewsFilter | Suppress signals near high-impact news events |
| Alerts | |
| Push_Enable / Push_MinAIScore | Push notification gate and minimum score |
| Signal_On_BarClose | Bar-close only or intrabar signals |
SYSTEM REQUIREMENTS
- Platform: MetaTrader 5 (build 2361 or later recommended)
- Chart type: Any symbol — Forex, Indices, Crypto, Commodities
- Recommended timeframes: M5, M15, M30, H1 (attach all four per symbol for Cross-TF gate)
- No external DLLs required. All computation runs natively inside the MetaTrader 5 sandbox
- Common folder write access required for state persistence and cross-TF communication
IMPORTANT NOTES
No indicator guarantees profitability. All AI components learn from your own live trade outcomes on your specific symbol and timeframe. Performance depends on market conditions, broker, spread, and execution. Always validate on a demo account before deploying on a funded account. Walk-forward testing is strongly recommended before using this indicator in a prop firm challenge or live environment.
The Cross-Timeframe Hit-Rate Gate requires a minimum of TFG_MinSamples resolved trades per timeframe before it becomes active. During the warm-up period all instances run permissively.
The RLS feature weighting requires a minimum of RLS_MinSamples resolved trades before deviating from the flat default. Until this threshold is reached the indicator behaves identically to a standard Hull suite with the NEUROCORE ensemble.
ICONIC HULLX AI v2.0 is developed and maintained by ICONIC FX / NEUROCORE AI. For support, questions, and updates visit the MQL5 profile page.
