OPEN-SOURCE SCRIPT
AetherEdge - Inefficiency Refill Model
🖊️ Overview
AE-IRM detects inefficient price moves made on thin volume — "voids" — and learns the probability that they are refilled (mean-revert) along with how long that takes. Many tools merely flag voids; AE-IRM answers "will it refill, and when?" with two learning layers. Surges and drops without participation tend statistically to be retraced, and this quantifies that probability and timing.
🔶 Key Features
A void oscillator that emphasizes low-participation inefficient moves (amplifies below-average-volume moves)
A learned refill probability — online logistic learns P(refill within N bars) from realized outcomes
A survival model — estimates timing (median bars-to-fill, average fill latency) from hazards
Fade (mean-reversion) signals: buy-side void → SHORT, sell-side void → LONG
A gold HUD showing void side, refill probability, median/average fill bars, pending voids, regime
Voids resolve on realized price — no repaint, signals gate on bar close
🧠 Technical Architecture
Void detection: standardized log-return retZ is multiplied by 1 if volume is below average and by a fade factor otherwise, giving osc. When |osc| exceeds a threshold a "void event" fires. The refill target (origin) is the price a few bars before the move departed.
Refill probability (ML): six features — void magnitude, participation shortfall, move extremeness, efficiency ratio (range/trend), volatility regime, and distance to origin (ATR) — are standardized online (EWMA), and an online logistic regression learns P(refill). Each void is trained at resolution with its realized label (filled, or censored at N bars) — no lookahead.
Survival (hazard / Kaplan-Meier-style): each pending void is tracked; at resolution a life table over age buckets (reached/filled) is updated. Per-bucket hazard h = filled/reached builds a survival curve S = Π(1 − h), and the age at which cumulative fill probability reaches 50% is the median bars-to-fill. Censored voids (unfilled at N) are correctly counted in the risk set.
Honest scope: a linear logistic classifier plus a nonparametric survival estimate. Not deep learning, and not a guarantee of refill.
⚙️ Recommended Settings & Tuning Guide
Key parameters: void threshold (thr), fade factor, displacement lookback (vLook), age bucket width × count (= refill horizon N), minimum refill probability (probThr).
Raise thr → only strong inefficiencies (fewer, higher quality); lower → more detections
Lower the fade factor → stricter thin-volume condition (more strongly excludes moves on volume)
Set the refill horizon N (= bucket width × count) to the timeframe you expect reversion over
Crypto starting points (tune on your chart):
BTC / ETH (15m–1H): defaults are the baseline (thr 2.0, fade 0.3, N = 40 bars)
SOL / XRP and high-vol alts: thr 2.5 to filter noisy voids, fade 0.2 to tighten the thin-volume condition
Scalping (1–5m): bucket width ~3 for a shorter N, targeting immediate retraces
Swing (4H–daily): larger bucket width and longer N, raise warmup so the statistics fill out
Raising probThr narrows signals to voids the model finds more likely to refill
💡 How to Use in Practice
Fade (counter-trend): SHORT on a buy-side void (thin-volume surge), LONG on a sell-side void (thin-volume drop); higher refill probability = greater edge
Reading the probability: high refill probability = strong reversion expectation; low = the move may continue (trend continuation)
Timing: median/average fill bars are your take-profit guide — design exits around "fills in ~X bars"
Use the regime: in RANGE, refills work better; in TREND, voids may run without filling — read it alongside the HUD regime
Combinations: pair with AE-VECTOR's target band or AE-QUORUM's directional probability, and use IRM's probability and timing to judge whether to fade and how to size
⚠️ Important Notes
Learning period: no signals until warmup bars; the classifier and survival statistics need time to spin up
Learning reset: changing inputs, symbol, or timeframe re-learns the internal state (weights and hazards)
Probability, not a guarantee: in strong trends voids can stay unfilled (censored) for a long time — avoid fading low-probability voids
Oscillator only: it does not draw void zones on price — operate from the signals and the HUD's probability and timing
🚨 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-IRM detects inefficient price moves made on thin volume — "voids" — and learns the probability that they are refilled (mean-revert) along with how long that takes. Many tools merely flag voids; AE-IRM answers "will it refill, and when?" with two learning layers. Surges and drops without participation tend statistically to be retraced, and this quantifies that probability and timing.
🔶 Key Features
A void oscillator that emphasizes low-participation inefficient moves (amplifies below-average-volume moves)
A learned refill probability — online logistic learns P(refill within N bars) from realized outcomes
A survival model — estimates timing (median bars-to-fill, average fill latency) from hazards
Fade (mean-reversion) signals: buy-side void → SHORT, sell-side void → LONG
A gold HUD showing void side, refill probability, median/average fill bars, pending voids, regime
Voids resolve on realized price — no repaint, signals gate on bar close
🧠 Technical Architecture
Void detection: standardized log-return retZ is multiplied by 1 if volume is below average and by a fade factor otherwise, giving osc. When |osc| exceeds a threshold a "void event" fires. The refill target (origin) is the price a few bars before the move departed.
Refill probability (ML): six features — void magnitude, participation shortfall, move extremeness, efficiency ratio (range/trend), volatility regime, and distance to origin (ATR) — are standardized online (EWMA), and an online logistic regression learns P(refill). Each void is trained at resolution with its realized label (filled, or censored at N bars) — no lookahead.
Survival (hazard / Kaplan-Meier-style): each pending void is tracked; at resolution a life table over age buckets (reached/filled) is updated. Per-bucket hazard h = filled/reached builds a survival curve S = Π(1 − h), and the age at which cumulative fill probability reaches 50% is the median bars-to-fill. Censored voids (unfilled at N) are correctly counted in the risk set.
Honest scope: a linear logistic classifier plus a nonparametric survival estimate. Not deep learning, and not a guarantee of refill.
⚙️ Recommended Settings & Tuning Guide
Key parameters: void threshold (thr), fade factor, displacement lookback (vLook), age bucket width × count (= refill horizon N), minimum refill probability (probThr).
Raise thr → only strong inefficiencies (fewer, higher quality); lower → more detections
Lower the fade factor → stricter thin-volume condition (more strongly excludes moves on volume)
Set the refill horizon N (= bucket width × count) to the timeframe you expect reversion over
Crypto starting points (tune on your chart):
BTC / ETH (15m–1H): defaults are the baseline (thr 2.0, fade 0.3, N = 40 bars)
SOL / XRP and high-vol alts: thr 2.5 to filter noisy voids, fade 0.2 to tighten the thin-volume condition
Scalping (1–5m): bucket width ~3 for a shorter N, targeting immediate retraces
Swing (4H–daily): larger bucket width and longer N, raise warmup so the statistics fill out
Raising probThr narrows signals to voids the model finds more likely to refill
💡 How to Use in Practice
Fade (counter-trend): SHORT on a buy-side void (thin-volume surge), LONG on a sell-side void (thin-volume drop); higher refill probability = greater edge
Reading the probability: high refill probability = strong reversion expectation; low = the move may continue (trend continuation)
Timing: median/average fill bars are your take-profit guide — design exits around "fills in ~X bars"
Use the regime: in RANGE, refills work better; in TREND, voids may run without filling — read it alongside the HUD regime
Combinations: pair with AE-VECTOR's target band or AE-QUORUM's directional probability, and use IRM's probability and timing to judge whether to fade and how to size
⚠️ Important Notes
Learning period: no signals until warmup bars; the classifier and survival statistics need time to spin up
Learning reset: changing inputs, symbol, or timeframe re-learns the internal state (weights and hazards)
Probability, not a guarantee: in strong trends voids can stay unfilled (censored) for a long time — avoid fading low-probability voids
Oscillator only: it does not draw void zones on price — operate from the signals and the HUD's probability and timing
🚨 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.
開源腳本
秉持TradingView一貫精神，這個腳本的創作者將其設為開源，以便交易者檢視並驗證其功能。向作者致敬！您可以免費使用此腳本，但請注意，重新發佈代碼需遵守我們的社群規範。
免責聲明
這些資訊和出版物並非旨在提供，也不構成TradingView提供或認可的任何形式的財務、投資、交易或其他類型的建議或推薦。請閱讀使用條款以了解更多資訊。
開源腳本
秉持TradingView一貫精神，這個腳本的創作者將其設為開源，以便交易者檢視並驗證其功能。向作者致敬！您可以免費使用此腳本，但請注意，重新發佈代碼需遵守我們的社群規範。
免責聲明
這些資訊和出版物並非旨在提供，也不構成TradingView提供或認可的任何形式的財務、投資、交易或其他類型的建議或推薦。請閱讀使用條款以了解更多資訊。