OPEN-SOURCE SCRIPT
更新済

Ultimate Confluence Engine Synapse v6

1 456
Ultimate Confluence Engine v6

Description:
Ultimate Confluence Engine v6 is a multi-layer signal confluence system that combines seven independent analytical engines into a single, unified buy/sell framework. A signal fires only when all active filters agree — dramatically reducing noise and false positives compared to single-indicator approaches.

How It Works
The indicator stacks seven confirmation layers. Every active layer must pass before a signal label appears on the chart.
① Evasive SuperTrend A volatility-adaptive SuperTrend that adjusts its multiplier dynamically using ATR, volume weight, and trend correlation strength. A noise threshold prevents premature flips during sideways chop; an expansion alpha gently repositions the band during consolidation rather than snapping aggressively.
② KNN Machine Learning Filter A K-Nearest Neighbors classifier trained on four normalized features: RSI momentum, price deviation from its moving average, RSI signal divergence, and a Choppiness Index component. The model evaluates the current bar against up to 500 historical samples, weights neighbors by Gaussian distance, and outputs a directional probability. Signals pass only when the bullish or bearish probability exceeds a configurable threshold (default 0.65).
③ Cumulative Volume Delta (CVD) Intrabar buying and selling pressure is estimated from the high-low-close relationship and accumulated into a running CVD. A moving average of CVD acts as a trend filter — buy signals require CVD above its MA, sell signals require the opposite.
④ VWAP Bands Session VWAP is computed from cumulative volume-weighted prices. Standard deviation bands (±1σ, ±2σ) are plotted. The filter confirms directional bias: long signals require price above VWAP, short signals require price below.
⑤ Candlestick Pattern Recognition Five bullish reversal patterns are scored per bar: Bullish Engulfing, Hammer, Morning Star, Bullish Harami, and Three Rising Soldiers. A minimum pattern count threshold (default 1) gates the signal. The score and matched pattern names appear in the signal tooltip.
⑥ Volume Spike Confirmation A volume multiplier threshold (default 1.3×) relative to a rolling average filters out low-conviction bars. Volume ratio also dynamically adjusts ribbon opacity — higher volume produces a more vivid ribbon.
⑦ Multi-Timeframe ADX Trend Alignment Optional MTF filter queries ADX trend direction across 1h, 4h, and daily timeframes. A minimum count of aligned bullish or bearish timeframes (configurable) must be met. Disabled by default for performance; enable when trend alignment context matters.

Quality Score
Each signal receives a quality rating (WEAK / GOOD / STRONG / EXCELLENT) based on volume spike magnitude and ATR expansion. The score is displayed in the dashboard and in the signal tooltip.

Trend Duration Prediction
After each SuperTrend flip, the indicator samples historical durations of the same trend direction, computes an exponentially-weighted average (EWA) and median, and projects a predicted trend length as a fading box sequence. Box opacity decreases over time to represent declining probability. Probability percentages (85% / 50% / 20% / 5%) are marked along the projection. Prediction length is adjusted by trading style multiplier, asset type multiplier, and current ATR context.

Trading Style Presets
Auto-detection maps the current timeframe to Scalping, Day Trading, Swing, or Position. Each preset adjusts SuperTrend ATR length, multiplier, and prediction weighting automatically. A Custom mode exposes all parameters manually.

Visuals
15-layer gradient ribbon between close and SuperTrend band, color-coded bull/bear, opacity modulated by volume
SuperTrend line rendered as solid, dotted (noisy zone), or circle (flip bar)
VWAP with ±1σ and ±2σ bands
KNN confidence expressed as per-bar color gradient
Buy/Sell labels placed at flip bars only — not on every bar
Bottom-right dashboard showing all filter states, quality score, and trend bar count

How to Use
Add to any chart. Style auto-detects from your timeframe.
A Buy label marks confirmed bullish SuperTrend flips where all active filters agree. A Sell label marks confirmed bearish flips with the same requirement.
Hover the label tooltip for full signal breakdown: ML probability, CVD direction, VWAP position, volume ratio, MTF alignment, pattern matches, and quality score.
Use the prediction box to estimate approximate trend duration — treat it as a probabilistic guide, not a target.
Disable any filter layer independently via the input groups if your setup or timeframe does not require it.
Set alerts using the four built-in alert conditions: Ultimate BUY, Ultimate SELL, ST → Bull, ST → Bear.

Settings Overview
Group
Key Inputs
Trading Style
Auto / Scalping / Day Trading / Swing / Position / Custom
Evasive SuperTrend
ATR Length, Multiplier, Noise Threshold, Adaptive toggle
KNN ML
K-Neighbors, Sample Window, Stride, Probability Threshold
CVD
Enable toggle, MA Length
VWAP
Enable toggle, Show toggle, Band multipliers
Patterns
Minimum pattern count (0 = disabled)
Volume + MTF
Volume multiplier, MTF toggle, Min bullish TFs
Prediction
Mode (Simple / Standard / Advanced), Asset Type, EWA Decay


Notes
Signals appear on confirmed bars only (barstate.isconfirmed) — no repainting on close.
KNN recalculates every 3 bars to preserve chart performance.
MTF filter uses lookahead_on on historical data; disable if strict no-lookahead compliance is required for your workflow.
Past signal accuracy does not guarantee future results. Use in conjunction with your own risk management framework.
リリースノート
Removed unnecessary unrelated chart objects. Added "ML" to the indicator title
リリースノート
Hey! Happy Friday!

v6's bars (~148b, ~153b, ~81b, ~84b) come from a simple median×0.6 + avg×0.4 scaled by fixed style/asset/volatility multipliers — and its "85%/50%/20%/5%" labels are just static text at fixed positions, not actually computed probabilities.

v7's bars (~46b–77b) come from a real Weibull survival model, outlier-trimmed statistics, and five compounding multipliers (Structure 1×, Type 1.35×, Flip 1.18×, Learn 0.93×, Regime 1×) — its percentage labels are genuinely computed survival probabilities, and the Error Learning multiplier means it adjusts itself based on past misses.

v7 has a built-in Prediction Record that v6 never had — Success Rate 8/10 (80%) and Avg Accuracy 35% over the last 10 completed trends. v6 has no equivalent self-check; you can only eyeball its boxes against what later happened.

免責事項

これらの情報および投稿は、TradingViewが提供または承認する金融、投資、取引、またはその他の種類の助言もしくは推奨であることを意図したものではなく、またこれらに該当するものでもありません。詳細は利用規約をご覧ください。