OPEN-SOURCE SCRIPT
AetherEdge - STRATA | SMC + ML/RL
🖊️ Overview
AE-STRATA is an overlay toolkit that brings the main Smart Money Concepts (SMC / ICT) elements together and layers a genuine learning engine — online machine learning plus reinforcement learning — on top. Beyond drawing "where" (order blocks, FVGs, liquidity, structure shifts), it learns from that context to show which way price is more likely to go next (ML probability) and what stance to take now (RL action). The SMC drawings are re-implemented from public, generic ICT methods, with an honest learning layer added on top.
🔶 Key Features
Structure analysis — BOS/CHoCH (break/change of character) + swing labels (HH/HL/LH/LL)
Two-layer order blocks — two tiers of zones that fade on mitigation (testing)
FVG (fair value gaps) — tracked from creation to fill
Liquidity — EQH/EQL, sweeps, and traps
Premium / discount — expensive/cheap zones within the range
Deviation oscillator (Pulse) — quantifies stretch from the mean
ML probability — learns P(up over the next H bars) from seven SMC features
RL action — reinforcement learning suggests flat / long / short
A Smart Money Score (0–100) + ML probability + RL action in a gold HUD, dark theme
Learning is based on realized outcomes and gated on bar close — no repaint
🧠 Technical Architecture
SMC engine: order blocks, FVGs, BOS/CHoCH, liquidity, and premium/discount, re-implemented as public, generic ICT concepts.
Preprocessing: features are standardized online via EWMA, keeping a stable scale as the market changes.
ML (online logistic regression): seven SMC features feed an SGD + L2 model that learns P(up over the next H bars). Labels are H-bar-lagged realized returns, so the current prediction uses no future data.
RL (tabular Q-learning, TD(0)): a Q-table over 27 regime-discretized states × (flat / long / short). The reward is ATR-scaled realized return, and updates gate on barstate.isconfirmed; with more data it converges to the stance that pays in each regime.
Honest scope: online logistic regression + tabular Q-learning. Not deep learning (no DQN), and not a guarantee of the future.
⚙️ Recommended Settings & Tuning Guide
Main levers: structure (swing) length, order-block/FVG sensitivity, ML horizon H, learning rate, RL reward scale, Smart Money Score threshold.
Larger structure length → focus on major OBs/structure; smaller → more detections (and noise)
Larger horizon H → slower learning, swing-oriented; smaller → short-term
Higher learning rate → faster adaptation to recent conditions; lower → steadier
Crypto starting points (tune on your chart):
BTC / ETH (15m–4H): defaults are the baseline
SOL / XRP and high-vol alts: slightly longer structure length and tighter OB/FVG sensitivity to suppress wick noise
Scalping (1–15m): shorter horizon, slightly higher learning rate
Swing (4H–daily): longer horizon, ample warmup to fill out learning
The clearest setups are when ML probability, RL action, and the Smart Money Score all point the same way
💡 How to Use in Practice
Trade reactions off zones: price returns to a discount bullish OB or FVG with ML probability rising and RL = long supports a dip-buy (mirror for bearish)
Confirm structure: read BOS/CHoCH for the regime shift and check it agrees with the ML/RL direction
Smart Money Score: a 0–100 composite for the current edge at a glance
Reading the ML probability: high favors upside, low favors downside; near 50% is neutral — stand aside
RL action: a flat/long/short stance suggestion — read alongside the probability and score
Combinations: pairing with a strength oscillator or a higher-timeframe view improves precision
⚠️ Important Notes
Learning period: ML probability and RL action are unsettled until the model spins up (warmup)
Learning reset: changing inputs, symbol, or timeframe re-learns the internal state (weights and Q-table)
SMC is probabilistic: OBs/FVGs/liquidity are levels likely to react, not certainties
Probability, not a guarantee: the ML probability and RL action can be wrong — always use stops and position sizing
🚨 Disclaimer
This indicator is for educational and informational purposes only and is not financial advice or a recommendation to buy or sell. No method guarantees future profits; past performance does not indicate future results, and trading carries the risk of loss. All trading decisions are your own — use proper backtesting and disciplined risk management.
AE-STRATA is an overlay toolkit that brings the main Smart Money Concepts (SMC / ICT) elements together and layers a genuine learning engine — online machine learning plus reinforcement learning — on top. Beyond drawing "where" (order blocks, FVGs, liquidity, structure shifts), it learns from that context to show which way price is more likely to go next (ML probability) and what stance to take now (RL action). The SMC drawings are re-implemented from public, generic ICT methods, with an honest learning layer added on top.
🔶 Key Features
Structure analysis — BOS/CHoCH (break/change of character) + swing labels (HH/HL/LH/LL)
Two-layer order blocks — two tiers of zones that fade on mitigation (testing)
FVG (fair value gaps) — tracked from creation to fill
Liquidity — EQH/EQL, sweeps, and traps
Premium / discount — expensive/cheap zones within the range
Deviation oscillator (Pulse) — quantifies stretch from the mean
ML probability — learns P(up over the next H bars) from seven SMC features
RL action — reinforcement learning suggests flat / long / short
A Smart Money Score (0–100) + ML probability + RL action in a gold HUD, dark theme
Learning is based on realized outcomes and gated on bar close — no repaint
🧠 Technical Architecture
SMC engine: order blocks, FVGs, BOS/CHoCH, liquidity, and premium/discount, re-implemented as public, generic ICT concepts.
Preprocessing: features are standardized online via EWMA, keeping a stable scale as the market changes.
ML (online logistic regression): seven SMC features feed an SGD + L2 model that learns P(up over the next H bars). Labels are H-bar-lagged realized returns, so the current prediction uses no future data.
RL (tabular Q-learning, TD(0)): a Q-table over 27 regime-discretized states × (flat / long / short). The reward is ATR-scaled realized return, and updates gate on barstate.isconfirmed; with more data it converges to the stance that pays in each regime.
Honest scope: online logistic regression + tabular Q-learning. Not deep learning (no DQN), and not a guarantee of the future.
⚙️ Recommended Settings & Tuning Guide
Main levers: structure (swing) length, order-block/FVG sensitivity, ML horizon H, learning rate, RL reward scale, Smart Money Score threshold.
Larger structure length → focus on major OBs/structure; smaller → more detections (and noise)
Larger horizon H → slower learning, swing-oriented; smaller → short-term
Higher learning rate → faster adaptation to recent conditions; lower → steadier
Crypto starting points (tune on your chart):
BTC / ETH (15m–4H): defaults are the baseline
SOL / XRP and high-vol alts: slightly longer structure length and tighter OB/FVG sensitivity to suppress wick noise
Scalping (1–15m): shorter horizon, slightly higher learning rate
Swing (4H–daily): longer horizon, ample warmup to fill out learning
The clearest setups are when ML probability, RL action, and the Smart Money Score all point the same way
💡 How to Use in Practice
Trade reactions off zones: price returns to a discount bullish OB or FVG with ML probability rising and RL = long supports a dip-buy (mirror for bearish)
Confirm structure: read BOS/CHoCH for the regime shift and check it agrees with the ML/RL direction
Smart Money Score: a 0–100 composite for the current edge at a glance
Reading the ML probability: high favors upside, low favors downside; near 50% is neutral — stand aside
RL action: a flat/long/short stance suggestion — read alongside the probability and score
Combinations: pairing with a strength oscillator or a higher-timeframe view improves precision
⚠️ Important Notes
Learning period: ML probability and RL action are unsettled until the model spins up (warmup)
Learning reset: changing inputs, symbol, or timeframe re-learns the internal state (weights and Q-table)
SMC is probabilistic: OBs/FVGs/liquidity are levels likely to react, not certainties
Probability, not a guarantee: the ML probability and RL action can be wrong — always use stops and position sizing
🚨 Disclaimer
This indicator is for educational and informational purposes only and is not financial advice or a recommendation to buy or sell. No method guarantees future profits; past performance does not indicate future results, and trading carries the risk of loss. All trading decisions are your own — use proper backtesting and disciplined risk management.
Open-source Skript
Ganz im Sinne von TradingView hat dieser Autor sein/ihr Script als Open-Source veröffentlicht. Auf diese Weise können nun auch andere Trader das Script rezensieren und die Funktionalität überprüfen. Vielen Dank an den Autor! Sie können das Script kostenlos verwenden, aber eine Wiederveröffentlichung des Codes unterliegt unseren Hausregeln.
Haftungsausschluss
Die Informationen und Veröffentlichungen sind nicht als Finanz-, Anlage-, Handels- oder andere Arten von Ratschlägen oder Empfehlungen gedacht, die von TradingView bereitgestellt oder gebilligt werden, und stellen diese nicht dar. Lesen Sie mehr in den Nutzungsbedingungen.
Open-source Skript
Ganz im Sinne von TradingView hat dieser Autor sein/ihr Script als Open-Source veröffentlicht. Auf diese Weise können nun auch andere Trader das Script rezensieren und die Funktionalität überprüfen. Vielen Dank an den Autor! Sie können das Script kostenlos verwenden, aber eine Wiederveröffentlichung des Codes unterliegt unseren Hausregeln.
Haftungsausschluss
Die Informationen und Veröffentlichungen sind nicht als Finanz-, Anlage-, Handels- oder andere Arten von Ratschlägen oder Empfehlungen gedacht, die von TradingView bereitgestellt oder gebilligt werden, und stellen diese nicht dar. Lesen Sie mehr in den Nutzungsbedingungen.