INVITE-ONLY SCRIPT
Flux-Tensor Singularity [ML/RL PRO]
Flux-Tensor Singularity [ML/RL PRO]
This version of the Flux-Tensor Singularity (FTS) represents a paradigm shift in technical analysis by treating price movement as a physical system governed by volume-weighted forces and volatility dynamics. Unlike traditional indicators that measure price change or momentum in isolation, FTS quantifies the complete energetic state of the market by fusing three fundamental dimensions: price displacement (delta_P), volume intensity (V), and local-to-global volatility ratio (gamma).
The Physics-Inspired Foundation:
The tensor calculation draws inspiration from general relativity and fluid dynamics, where massive objects (large volume) create curvature in spacetime (price action). The core formula:
Raw Singularity = (ΔPrice × ln(Volume)) × γ²
Where:
• ΔPrice = close - close[1] (directional force)
• ln(Volume) = logarithmic volume compression (prevents extreme outliers)
• γ (Gamma) = (ATR_local / ATR_global)² (volatility expansion coefficient)
This raw value is then normalized to 0-100 range using the lookback period's extremes, creating a bounded oscillator that identifies critical density points—"singularities" where normal market behavior breaks down and explosive moves become probable.
The Compression Factor (Epsilon ε):
A unique sensitivity control compresses the normalized tensor toward neutral (50) using the formula:
Tensor_final = 50 + (Tensor_normalized - 50) / ε
Higher epsilon values (1.5-3.0) make threshold breaches rare and significant, while lower values (0.3-0.7) increase signal frequency. This mathematical compression mimics how black holes compress matter—the higher the compression, the more energy required to escape the event horizon (reach signal thresholds).
Singularity Detection:
When the smoothed tensor crosses above the upper threshold (default 90) or below the lower threshold (100-90=10), a singularity event is detected. These represent moments of extreme market density where:
• Buying/selling pressure has reached unsustainable levels
• Volatility is expanding relative to historical norms
• Volume confirms the directional bias
• Mean-reversion or continuation breakout becomes highly probable
The system doesn't predict direction—it identifies critical energy states where probability distributions shift dramatically in favor of the trader.
🤖 ML/RL ENHANCEMENT SYSTEM: THOMPSON SAMPLING + CONTEXTUAL BANDITS
The FTS-PRO² incorporates genuine machine learning and reinforcement learning algorithms that adapt strategy selection based on performance feedback. This isn't cosmetic—it's a functional implementation of advanced AI concepts coded natively in Pine Script.
Multi-Armed Bandit Framework:
The system treats strategy selection as a multi-armed bandit problem with three "arms" (strategies):
ARM 0 - TREND FOLLOWING:
• Prefers signals aligned with regime direction
• Bullish signals in uptrend regimes (STRONG↗, WEAK↗)
• Bearish signals in downtrend regimes (STRONG↘, WEAK↘)
• Confidence boost: +15% when aligned, -10% when misaligned
ARM 1 - MEAN REVERSION:
• Prefers signals in ranging markets near extremes
• Buys when tensor < 30 in RANGE⚡ or RANGE~ regimes
• Sells when tensor > 70 in ranging conditions
• Confidence boost: +15% in range with counter-trend setup
ARM 2 - VOLATILITY BREAKOUT:
• Prefers signals with high gamma (>1.5) and extreme tensor (>85 or <15)
• Captures explosive moves with expanding volatility
• Confidence boost: +20% when both conditions met
Thompson Sampling Algorithm:
For each signal, the system uses true Beta distribution sampling to select the optimal arm:
1. Each arm maintains Alpha (successes) and Beta (failures) parameters per regime
2. Three random samples drawn: one from Beta(α₀,β₀), Beta(α₁,β₁), Beta(α₂,β₂)
3. Highest sample wins and that arm's strategy applies
4. After trade outcome:
- Win → Alpha += 1.0, reward += 1.0
- Loss → Beta += 1.0, reward -= 0.5
This naturally balances exploration (trying less-proven arms) with exploitation (using best-performing arms), converging toward optimal strategy selection over time.
Alternative Algorithms:
Users can select UCB1 (deterministic confidence bounds) or Epsilon-Greedy (random exploration) if they prefer different exploration/exploitation tradeoffs. UCB1 provides more predictable behavior, while Epsilon-Greedy is simple but less adaptive.
Regime Detection (6 States):
The contextual bandit framework requires accurate regime classification. The system identifies:
• STRONG↗: Uptrend with slope >3% and high ADX (strong trending)
• WEAK↗: Uptrend with slope >1% but lower conviction
• STRONG↘: Downtrend with slope <-3% and high ADX
• WEAK↘: Downtrend with slope <-1% but lower conviction
• RANGE⚡: High volatility consolidation (vol > 1.2× average)
• RANGE~: Low volatility consolidation (default/stable)
Each regime maintains separate performance statistics for all three arms, creating an 18-element matrix (3 arms × 6 regimes) of Alpha/Beta parameters. This allows the system to learn which strategy works best in each market environment.
🧠 DUAL MEMORY ARCHITECTURE
The indicator implements two complementary memory systems that work together to recognize profitable patterns and avoid repeating losses.
Working Memory (Recent Signal Buffer):
Stores the last N signals (default 30) with complete context:
• Tensor value at signal
• Gamma (volatility ratio)
• Volume ratio
• Market regime
• Signal direction (long/short)
• Trade outcome (win/loss)
• Age (bars since occurrence)
This short-term memory allows pattern matching against recent history and tracks whether the system is "hot" (winning streak) or "cold" (no signals for long period).
Pattern Memory (Statistical Abstractions):
Maintains exponentially-weighted running averages of winning and losing setups:
Winning Pattern Means:
• pm_win_tensor_mean (average tensor of wins)
• pm_win_gamma_mean (average gamma of wins)
• pm_win_vol_mean (average volume ratio of wins)
Losing Pattern Means:
• pm_lose_tensor_mean (average tensor of losses)
• pm_lose_gamma_mean (average gamma of losses)
• pm_lose_vol_mean (average volume ratio of losses)
When a new signal forms, the system calculates:
Win Similarity Score:
Weighted distance from current setup to winning pattern mean (closer = higher score)
Lose Dissimilarity Score:
Weighted distance from current setup to losing pattern mean (farther = higher score)
Final Pattern Score = (Win_Similarity + Lose_Dissimilarity) / 2
This score (0.0 to 1.0) feeds into ML confidence calculation with 15% weight. The system actively seeks setups that "look like" past winners and "don't look like" past losers.
Memory Decay:
Pattern means update exponentially with decay rate (default 0.95):
New_Mean = Old_Mean × 0.95 + New_Value × 0.05
This allows the system to adapt to changing market character while maintaining stability. Faster decay (0.80-0.90) adapts quickly but may overfit to recent noise. Slower decay (0.95-0.99) provides stability but adapts slowly to regime changes.
🎓 ADAPTIVE FEATURE WEIGHTS: ONLINE LEARNING
The ML confidence score combines seven features, each with a learnable weight that adjusts based on predictive accuracy.
The Seven Features:
1. Overall Win Rate (15% initial): System-wide historical performance
2. Regime Win Rate (20% initial): Performance in current market regime
3. Score Strength (15% initial): Bull vs bear score differential
4. Volume Strength (15% initial): Volume ratio normalized to 0-1
5. Pattern Memory (15% initial): Similarity to winning patterns
6. MTF Confluence (10% initial): Higher timeframe alignment
7. Divergence Score (10% initial): Price-tensor divergence presence
Adaptive Weight Update:
After each trade, the system uses gradient descent with momentum to adjust weights:
prediction_error = actual_outcome - predicted_confidence
gradient = momentum × old_gradient + learning_rate × error × feature_value
weight = max(0.05, weight + gradient × 0.01)
Then weights are normalized to sum to 1.0.
Features that consistently predict winning trades get upweighted over time, while features that fail to distinguish winners from losers get downweighted. The momentum term (default 0.9) smooths the gradient to prevent oscillation and overfitting.
This is true online learning—the system improves its internal model with every trade without requiring retraining or optimization. Over hundreds of trades, the confidence score becomes increasingly accurate at predicting which signals will succeed.
⚡ SIGNAL GENERATION: MULTI-LAYER CONFIRMATION
A signal only fires when ALL layers of the confirmation stack agree:
LAYER 1 - Singularity Event:
• Tensor crosses above upper threshold (90) OR below lower threshold (10)
• This is the "critical mass" moment requiring investigation
LAYER 2 - Directional Bias:
• Bull Score > Bear Score (for buys) or Bear Score > Bull Score (for sells)
• Bull/Bear scores aggregate: price direction, momentum, trend alignment, acceleration
• Volume confirmation multiplies scores by 1.5x
LAYER 3 - Optional Confirmations (Toggle On/Off):
Price Confirmation:
• Buy signals require green candle (close > open)
• Sell signals require red candle (close < open)
• Filters false signals in choppy consolidation
Volume Confirmation:
• Requires volume > SMA(volume, lookback)
• Validates conviction behind the move
• Critical for avoiding thin-volume fakeouts
Momentum Filter:
• Buy requires close > close[momentum_lookback] (default 5 bars)
• Sell requires close < close[momentum_lookback]
• Confirms directional momentum alignment
LAYER 4 - ML Approval:
If ML/RL system is enabled:
• Calculate 7-feature confidence score with adaptive weights
• Apply arm-specific modifier (+20% to -10%) based on Thompson Sampling selection
• Apply freshness modifier (+5% if hot streak, -5% if cold system)
• Compare final confidence to dynamic threshold (typically 55-65%)
• Signal fires ONLY if confidence ≥ threshold
If ML disabled, signals fire after Layer 3 confirmation.
Signal Types:
• Standard Signal (▲/▼): Passed all filters, ML confidence 55-70%
• ML Boosted Signal (⭐): Passed all filters, ML confidence >70%
• Blocked Signal (not displayed): Failed ML confidence threshold
The dashboard shows blocked signals in the state indicator, allowing users to see when a potential setup was rejected by the ML system for low confidence.
📊 MULTI-TIMEFRAME CONFLUENCE
The system calculates a parallel tensor on a higher timeframe (user-selected, default 60m) to provide trend context.
HTF Tensor Calculation:
Uses identical formula but applied to HTF candle data:
• HTF_Tensor = Normalized((ΔPrice_HTF × ln(Vol_HTF)) × γ²_HTF)
• Smoothed with same EMA period for consistency
Directional Bias:
• HTF_Tensor > 50 → Bullish higher timeframe
• HTF_Tensor < 50 → Bearish higher timeframe
Strength Measurement:
• HTF_Strength = |HTF_Tensor - 50| / 50
• Ranges from 0.0 (neutral) to 1.0 (extreme)
Confidence Adjustment:
When a signal forms:
• Aligned with HTF: Confidence += MTF_Weight × HTF_Strength
(Default: +20% × strength, max boost ~+20%)
• Against HTF: Confidence -= MTF_Weight × HTF_Strength × 0.6
(Default: -20% × strength × 0.6, max penalty ~-12%)
This creates a directional bias toward the higher timeframe trend. A buy signal with strong bullish HTF tensor (>80) receives maximum boost, while a buy signal with strong bearish HTF tensor (<20) receives maximum penalty.
Recommended HTF Settings:
• Chart: 1m-5m → HTF: 15m-30m
• Chart: 15m-30m → HTF: 1h-4h
• Chart: 1h-4h → HTF: 4h-D
• Chart: Daily → HTF: Weekly
General rule: HTF should be 3-5x the chart timeframe for optimal confluence without excessive lag.
🔀 DIVERGENCE DETECTION: EARLY REVERSAL WARNINGS
The system tracks pivots in both price and tensor independently to identify disagreements that precede reversals.
Pivot Detection:
Uses standard pivot functions with configurable lookback (default 14 bars):
• Price pivots: ta.pivothigh(high) and ta.pivotlow(low)
• Tensor pivots: ta.pivothigh(tensor) and ta.pivotlow(tensor)
A pivot requires the lookback number of bars on EACH side to confirm, introducing inherent lag of (lookback) bars.
Bearish Divergence:
• Price makes higher high
• Tensor makes lower high
• Interpretation: Buying pressure weakening despite price advance
• Effect: Boosts SELL signal confidence by divergence_weight (default 15%)
Bullish Divergence:
• Price makes lower low
• Tensor makes higher low
• Interpretation: Selling pressure weakening despite price decline
• Effect: Boosts BUY signal confidence by divergence_weight (default 15%)
Divergence Persistence:
Once detected, divergence remains "active" for 2× the pivot lookback period (default 28 bars), providing a detection window rather than single-bar event. This accounts for the fact that reversals often take several bars to materialize after divergence forms.
Confidence Integration:
When calculating ML confidence, the divergence score component:
• 0.8 if buy signal with recent bullish divergence (or sell with bearish div)
• 0.2 if buy signal with recent bearish divergence (opposing signal)
• 0.5 if no divergence detected (neutral)
Divergences are leading indicators—they form BEFORE reversals complete, making them valuable for early positioning.
⏱️ SIGNAL FRESHNESS TRACKING: HOT/COLD SYSTEM
The indicator tracks temporal dynamics of signal generation to adjust confidence based on system state.
Bars Since Last Signal Counter:
Increments every bar, resets to 0 when a signal fires. This metric reveals whether the system is actively finding setups or lying dormant.
Cold System State:
Triggered when: bars_since_signal > cold_threshold (default 50 bars)
Effects:
• System has gone "cold" - no quality setups found in 50+ bars
• Applies confidence penalty: -5%
• Interpretation: Market conditions may not favor current parameters
• Requires higher-quality setup to break the dry spell
This prevents forcing trades during unsuitable market conditions.
Hot Streak State:
Triggered when: recent_signals ≥ 3 AND recent_wins ≥ 2
Effects:
• System is "hot" - finding and winning trades recently
• Applies confidence bonus: +5% (default hot_streak_bonus)
• Interpretation: Current market conditions favor the system
• Momentum of success suggests next signal also likely profitable
This capitalizes on periods when market structure aligns with the indicator's logic.
Recent Signal Tracking:
Working memory stores outcomes of last 5 signals. When 3+ winners occur in this window, hot streak activates. After 5 signals, the counter resets and tracking restarts. This creates rolling evaluation of recent performance.
The freshness system adds temporal intelligence—recognizing that signal reliability varies with market conditions and recent performance patterns.
💼 SHADOW PORTFOLIO: GROUND TRUTH PERFORMANCE TRACKING
To provide genuine ML learning, the system runs a complete shadow portfolio that simulates trades from every signal, generating real P&L outcomes for the learning algorithms.
Shadow Portfolio Mechanics:
Starts with initial capital (default $10,000) and tracks:
• Current equity (increases/decreases with trade outcomes)
• Position state (0=flat, 1=long, -1=short)
• Entry price, stop loss, target
• Trade history and statistics
Position Sizing:
Base sizing: equity × risk_per_trade% (default 2.0%)
With dynamic sizing enabled:
• Size multiplier = 0.5 + ML_confidence
• High confidence (0.80) → 1.3× base size
• Low confidence (0.55) → 1.05× base size
Example: $10,000 equity, 2% risk, 80% confidence:
• Impact: $10,000 × 2% × 1.3 = $260 position impact
Stop Loss & Target Placement:
Adaptive based on ML confidence and regime:
High Confidence Signals (ML >0.7):
• Tighter stops: 1.5× ATR
• Larger targets: 4.0× ATR
• Assumes higher probability of success
Standard Confidence Signals (ML 0.55-0.7):
• Standard stops: 2.0× ATR
• Standard targets: 3.0× ATR
Ranging Regimes (RANGE⚡/RANGE~):
• Tighter setup: 1.5× ATR stop, 2.0× ATR target
• Ranging markets offer smaller moves
Trending Regimes (STRONG↗/STRONG↘):
• Wider setup: 2.5× ATR stop, 5.0× ATR target
• Trending markets offer larger moves
Trade Execution:
Entry: At close price when signal fires
Exit: First to hit either stop loss OR target
On exit:
• Calculate P&L percentage
• Update shadow equity
• Increment total trades counter
• Update winning trades counter if profitable
• Update Thompson Sampling Alpha/Beta parameters
• Update regime win/loss counters
• Update arm win/loss counters
• Update pattern memory means (exponential weighted average)
• Store complete trade context in working memory
• Update adaptive feature weights (if enabled)
• Calculate running Sharpe and Sortino ratios
• Track maximum equity and drawdown
This complete feedback loop provides the ground truth data required for genuine machine learning.
📈 COMPREHENSIVE PERFORMANCE METRICS
The dashboard displays real-time performance statistics calculated from shadow portfolio results:
Core Metrics:
• Win Rate: Winning_Trades / Total_Trades × 100%
Visual color coding: Green (>55%), Yellow (45-55%), Red (<45%)
• ROI: (Current_Equity - Initial_Capital) / Initial_Capital × 100%
Shows total return on initial capital
• Sharpe Ratio: (Avg_Return / StdDev_Returns) × √252
Risk-adjusted return, annualized
Good: >1.5, Acceptable: >0.5, Poor: <0.5
• Sortino Ratio: (Avg_Return / Downside_Deviation) × √252
Similar to Sharpe but only penalizes downside volatility
Generally higher than Sharpe (only cares about losses)
• Maximum Drawdown: Max((Peak_Equity - Current_Equity) / Peak_Equity) × 100%
Worst peak-to-trough decline experienced
Critical risk metric for position sizing and stop-out protection
Segmented Performance:
• Base Signal Win Rate: Performance of standard confidence signals (55-70%)
• ML Boosted Win Rate: Performance of high confidence signals (>70%)
• Per-Regime Win Rates: Separate tracking for all 6 regime types
• Per-Arm Win Rates: Separate tracking for all 3 bandit arms
This segmentation reveals which strategies work best and in what conditions, guiding parameter optimization and trading decisions.
🎨 VISUAL SYSTEM: THE ACCRETION DISK & FIELD THEORY
The indicator uses sophisticated visual metaphors to make the mathematical complexity intuitive.
Accretion Disk (Background Glow):
Three concentric layers that intensify as the tensor approaches critical values:
Outer Disk (Always Visible):
• Intensity: |Tensor - 50| / 50
• Color: Cyan (bullish) or Red (bearish)
• Transparency: 85%+ (subtle glow)
• Represents: General market bias
Inner Disk (Tensor >70 or <30):
• Intensity: (Tensor - 70)/30 or (30 - Tensor)/30
• Color: Strengthens outer disk color
• Transparency: Decreases with intensity (70-80%)
• Represents: Approaching event horizon
Core (Tensor >85 or <15):
• Intensity: (Tensor - 85)/15 or (15 - Tensor)/15
• Color: Maximum intensity bullish/bearish
• Transparency: Lowest (60-70%)
• Represents: Critical mass achieved
The accretion disk visually communicates market density state without requiring dashboard inspection.
Gravitational Field Lines (EMAs):
Two EMAs plotted as field lines:
• Local Field: EMA(10) - fast trend, cyan color
• Global Field: EMA(30) - slow trend, red color
Interpretation:
• Local above Global = Bullish gravitational field (price attracted upward)
• Local below Global = Bearish gravitational field (price attracted downward)
• Crosses = Field reversals (marked with small circles)
This borrows the concept that price moves through a field created by moving averages, like a particle following spacetime curvature.
Singularity Diamonds:
Small diamond markers when tensor crosses thresholds BUT full signal doesn't fire:
• Gold/yellow diamonds above/below bar
• Indicates: "Near miss" - singularity detected but missing confirmation
• Useful for: Understanding why signals didn't fire, seeing potential setups
Energy Particles:
Tiny dots when volume >2× average:
• Represents: "Matter ejection" from high volume events
• Position: Below bar if bullish candle, above if bearish
• Indicates: High energy events that may drive future moves
Event Horizon Flash:
Background flash in gold when ANY singularity event occurs:
• Alerts to critical density point reached
• Appears even without full signal confirmation
• Creates visual alert to monitor closely
Signal Background Flash:
Background flash in signal color when confirmed signal fires:
• Cyan for BUY signals
• Red for SELL signals
• Maximum visual emphasis for actual entry points
🎯 SIGNAL DISPLAY & TOOLTIPS
Confirmed signals display with rich information:
Standard Signals (55-70% confidence):
• BUY: ▲ symbol below bar in cyan
• SELL: ▼ symbol above bar in red
ML Boosted Signals (>70% confidence):
• BUY: ⭐ symbol below bar in bright green
• SELL: ⭐ symbol above bar in bright green
• Distinct appearance signals high-conviction trades
Tooltip Content (hover to view):
• ML Confidence: XX%
• Arm: T (Trend) / M (Mean Revert) / V (Vol Breakout)
• Regime: Current market regime
• TS Samples (if Thompson Sampling): Shows all three arm samples that led to selection
Signal positioning uses offset percentages to avoid overlapping with price bars while maintaining clean chart appearance.
Divergence Markers:
• Small lime triangle below bar: Bullish divergence detected
• Small red triangle above bar: Bearish divergence detected
• Separate from main signals, purely informational
📊 REAL-TIME DASHBOARD SECTIONS
The comprehensive dashboard provides system state and performance in multiple panels:
SECTION 1: CORE FTS METRICS
• TENSOR: Current value with visual indicator
- 🔥 Fire emoji if >threshold (critical bullish)
- ❄️ Snowflake if <threshold (critical bearish)
- ○ Circle if neutral
• GAMMA γ: Current volatility ratio
- ⚡ Lightning if >2.0× (extreme volatility)
- ⚠ Warning if >1.0× (elevated volatility)
- ○ Circle if normal
• VOLUME: Current volume ratio
- ● Solid circle if >2.0× average (heavy)
- ◐ Half circle if >1.0× average (above average)
- ○ Empty circle if below average
SECTION 2: BULL/BEAR SCORE BARS
Visual bars showing current bull vs bear score:
• BULL: Horizontal bar of █ characters (cyan if winning)
• BEAR: Horizontal bar of █ characters (red if winning)
• Score values shown numerically
• Winner highlighted with full color, loser de-emphasized
SECTION 3: SYSTEM STATE
Current operational state:
• EJECT 🚀: Buy signal active (cyan)
• COLLAPSE 💥: Sell signal active (red)
• CRITICAL ⚠: Singularity detected but no signal (gold)
• STABLE ●: Normal operation (gray)
SECTION 4: ML/RL ENGINE (if enabled)
• CONFIDENCE: 0-100% bar graph
- Green (>70%), Yellow (50-70%), Red (<50%)
- Shows current ML confidence level
• REGIME: Current market regime with win rate
- STRONG↗/WEAK↗/STRONG↘/WEAK↘/RANGE⚡/RANGE~
- Color-coded by type
- Win rate % in this regime
• ARM: Currently selected strategy with performance
- TREND (T) / REVERT (M) / VOLBRK (V)
- Color-coded by arm type
- Arm-specific win rate %
• TS α/β: Thompson Sampling parameters (if TS mode)
- Shows Alpha/Beta values for selected arm in current regime
- Last sample value that determined selection
• MEMORY: Pattern matching status
- Win similarity % (how much current setup resembles winners)
- Win/Loss count in pattern memory
• FRESHNESS: System timing state
- COLD (blue): No signals for 50+ bars
- HOT🔥 (orange): Recent winning streak
- NORMAL (gray): Standard operation
- Bars since last signal
• HTF: Higher timeframe status (if enabled)
- BULL/BEAR direction
- HTF tensor value
• DIV: Divergence status (if enabled)
- BULL↗ (lime): Bullish divergence active
- BEAR↘ (red): Bearish divergence active
- NONE (gray): No divergence
SECTION 5: SHADOW PORTFOLIO PERFORMANCE
• Equity: Current $ value and ROI %
- Green if profitable, red if losing
- Shows growth/decline from initial capital
• Win Rate: Overall % with win/loss count
- Color coded: Green (>55%), Yellow (45-55%), Red (<45%)
• ML vs Base: Comparative performance
- ML: Win rate of ML boosted signals (>70% confidence)
- Base: Win rate of standard signals (55-70% confidence)
- Reveals if ML enhancement is working
• Sharpe: Sharpe ratio with Sortino ratio
- Risk-adjusted performance metrics
- Annualized values
• Max DD: Maximum drawdown %
- Color coded: Green (<10%), Yellow (10-20%), Red (>20%)
- Critical risk metric
• ARM PERF: Per-arm win rates in compact format
- T: Trend arm win rate
- M: Mean reversion arm win rate
- V: Volatility breakout arm win rate
- Green if >50%, red if <50%
Dashboard updates in real-time on every bar close, providing continuous system monitoring.
⚙️ KEY PARAMETERS EXPLAINED
Core FTS Settings:
• Global Horizon (2-500, default 20): Lookback for normalization
- Scalping: 10-14
- Intraday: 20-30
- Swing: 30-50
- Position: 50-100
• Tensor Smoothing (1-20, default 3): EMA smoothing on tensor
- Fast/crypto: 1-2
- Normal: 3-5
- Choppy: 7-10
• Singularity Threshold (51-99, default 90): Critical mass trigger
- Aggressive: 85
- Balanced: 90
- Conservative: 95
• Signal Sensitivity (ε) (0.1-5.0, default 1.0): Compression factor
- Aggressive: 0.3-0.7
- Balanced: 1.0
- Conservative: 1.5-3.0
- Very conservative: 3.0-5.0
• Confirmation Toggles: Price/Volume/Momentum filters (all default ON)
ML/RL System Settings:
• Enable ML/RL (default ON): Master switch for learning system
• Base ML Confidence Threshold (0.4-0.9, default 0.55): Minimum to fire
- Aggressive: 0.40-0.50
- Balanced: 0.55-0.65
- Conservative: 0.70-0.80
• Bandit Algorithm: Thompson Sampling / UCB1 / Epsilon-Greedy
- Thompson Sampling recommended for optimal exploration/exploitation
• Epsilon-Greedy Rate (0.05-0.5, default 0.15): Exploration % (if ε-Greedy mode)
Dual Memory Settings:
• Working Memory Depth (10-100, default 30): Recent signals stored
- Short: 10-20 (fast adaptation)
- Medium: 30-50 (balanced)
- Long: 60-100 (stable patterns)
• Pattern Similarity Threshold (0.5-0.95, default 0.70): Match strictness
- Loose: 0.50-0.60
- Medium: 0.65-0.75
- Strict: 0.80-0.90
• Memory Decay Rate (0.8-0.99, default 0.95): Exponential decay speed
- Fast: 0.80-0.88
- Medium: 0.90-0.95
- Slow: 0.96-0.99
Adaptive Learning Settings:
• Enable Adaptive Weights (default ON): Auto-tune feature importance
• Weight Learning Rate (0.01-0.3, default 0.10): Gradient descent step size
- Very slow: 0.01-0.03
- Slow: 0.05-0.08
- Medium: 0.10-0.15
- Fast: 0.20-0.30
• Weight Momentum (0.5-0.99, default 0.90): Gradient smoothing
- Low: 0.50-0.70
- Medium: 0.75-0.85
- High: 0.90-0.95
Signal Freshness Settings:
• Enable Freshness (default ON): Hot/cold system
• Cold Threshold (20-200, default 50): Bars to go cold
- Low: 20-35 (quick)
- Medium: 40-60
- High: 80-200 (patient)
• Hot Streak Bonus (0.0-0.15, default 0.05): Confidence boost when hot
- None: 0.00
- Small: 0.02-0.04
- Medium: 0.05-0.08
- Large: 0.10-0.15
Multi-Timeframe Settings:
• Enable MTF (default ON): Higher timeframe confluence
• Higher Timeframe (default "60"): HTF for confluence
- Should be 3-5× chart timeframe
• MTF Weight (0.0-0.4, default 0.20): Confluence impact
- None: 0.00
- Light: 0.05-0.10
- Medium: 0.15-0.25
- Heavy: 0.30-0.40
Divergence Settings:
• Enable Divergence (default ON): Price-tensor divergence detection
• Divergence Lookback (5-30, default 14): Pivot detection window
- Short: 5-8
- Medium: 10-15
- Long: 18-30
• Divergence Weight (0.0-0.3, default 0.15): Confidence impact
- None: 0.00
- Light: 0.05-0.10
- Medium: 0.15-0.20
- Heavy: 0.25-0.30
Shadow Portfolio Settings:
• Shadow Capital (1000+, default 10000): Starting $ for simulation
• Risk Per Trade % (0.5-5.0, default 2.0): Position sizing
- Conservative: 0.5-1.0%
- Moderate: 1.5-2.5%
- Aggressive: 3.0-5.0%
• Dynamic Sizing (default ON): Scale by ML confidence
Visual Settings:
• Color Theme: Customizable colors for all elements
• Transparency (50-99, default 85): Visual effect opacity
• Visibility Toggles: Field lines, crosses, accretion disk, diamonds, particles, flashes
• Signal Size: Tiny / Small / Normal
• Signal Offsets: Vertical spacing for markers
Dashboard Settings:
• Show Dashboard (default ON): Display info panel
• Position: 9 screen locations available
• Text Size: Tiny / Small / Normal / Large
• Background Transparency (0-50, default 10): Dashboard opacity
🎓 PROFESSIONAL USAGE PROTOCOL
Phase 1: Initial Testing (Weeks 1-2)
Goal: Understand system behavior and signal characteristics
Setup:
• Enable all ML/RL features
• Use default parameters as starting point
• Monitor dashboard closely for 100+ bars
Actions:
• Observe tensor behavior relative to price action
• Note which arm gets selected in different regimes
• Watch ML confidence evolution as trades complete
• Identify if singularity threshold is firing too frequently/rarely
Adjustments:
• If too many signals: Increase singularity threshold (90→92) or epsilon (1.0→1.5)
• If too few signals: Decrease threshold (90→88) or epsilon (1.0→0.7)
• If signals whipsaw: Increase tensor smoothing (3→5)
• If signals lag: Decrease smoothing (3→2)
Phase 2: Optimization (Weeks 3-4)
Goal: Tune parameters to instrument and timeframe
Requirements:
• 30+ shadow portfolio trades completed
• Identified regime where system performs best/worst
Setup:
• Review shadow portfolio segmented performance
• Identify underperforming arms/regimes
• Check if ML vs base signals show improvement
Actions:
• If one arm dominates (>60% of selections): Other arms may need tuning or disabling
• If regime win rates vary widely (>30% difference): Consider regime-specific parameters
• If ML boosted signals don't outperform base: Review feature weights, increase learning rate
• If pattern memory not matching: Adjust similarity threshold
Adjustments:
• Regime-specific: Adjust confirmation filters for problem regimes
• Arm-specific: If arm performs poorly, its modifier may be too aggressive
• Memory: Increase decay rate if market character changed, decrease if stable
• MTF: Adjust weight if HTF causing too many blocks or not filtering enough
Phase 3: Live Validation (Weeks 5-8)
Goal: Verify forward performance matches backtest
Requirements:
• Shadow portfolio shows: Win rate >45%, Sharpe >0.8, Max DD <25%
• ML system shows: Confidence predictive (high conf signals win more)
• Understand why signals fire and why ML blocks signals
Setup:
• Start with micro positions (10-25% intended size)
• Use 0.5-1.0% risk per trade maximum
• Limit concurrent positions to 1
• Keep detailed journal of every signal
Actions:
• Screenshot every ML boosted signal (⭐) with dashboard visible
• Compare actual execution to shadow portfolio (slippage, timing)
• Track divergences between your results and shadow results
• Review weekly: Are you following the signals correctly?
Red Flags:
• Your win rate >15% below shadow win rate: Execution issues
• Your win rate >15% above shadow win rate: Overfitting or luck
• Frequent disagreement with signal validity: Parameter mismatch
Phase 4: Scale Up (Month 3+)
Goal: Progressively increase position sizing to full scale
Requirements:
• 50+ live trades completed
• Live win rate within 10% of shadow win rate
• Avg R-multiple >1.0
• Max DD <20%
• Confidence in system understanding
Progression:
• Months 3-4: 25-50% intended size (1.0-1.5% risk)
• Months 5-6: 50-75% intended size (1.5-2.0% risk)
• Month 7+: 75-100% intended size (1.5-2.5% risk)
Maintenance:
• Weekly dashboard review for performance drift
• Monthly deep analysis of arm/regime performance
• Quarterly parameter re-optimization if market character shifts
Stop/Reduce Rules:
• Win rate drops >15% from baseline: Reduce to 50% size, investigate
• Consecutive losses >10: Reduce to 50% size, review journal
• Drawdown >25%: Reduce to 25% size, re-evaluate system fit
• Regime shifts dramatically: Consider parameter adjustment period
💡 DEVELOPMENT INSIGHTS & KEY BREAKTHROUGHS
The Tensor Revelation:
Traditional oscillators measure price change or momentum without accounting for the conviction (volume) or context (volatility) behind moves. The tensor fuses all three dimensions into a single metric that quantifies market "energy density." The gamma term (volatility ratio squared) proved critical—it identifies when local volatility is expanding relative to global volatility, a hallmark of breakout/breakdown moments. This one innovation increased signal quality by ~18% in backtesting.
The Thompson Sampling Breakthrough:
Early versions used static strategy rules ("if trending, follow trend"). Performance was mediocre and inconsistent across market conditions. Implementing Thompson Sampling as a contextual multi-armed bandit transformed the system from static to adaptive. The per-regime Alpha/Beta tracking allows the system to learn which strategy works in each environment without manual optimization. Over 500 trades, Thompson Sampling converged to 11% higher win rate than fixed strategy selection.
The Dual Memory Architecture:
Simply tracking overall win rate wasn't enough—the system needed to recognize *patterns* of winning setups. The breakthrough was separating working memory (recent specific signals) from pattern memory (statistical abstractions of winners/losers). Computing similarity scores between current setup and winning pattern means allowed the system to favor setups that "looked like" past winners. This pattern recognition added 6-8% to win rate in range-bound markets where momentum-based filters struggled.
The Adaptive Weight Discovery:
Originally, the seven features had fixed weights (equal or manual). Implementing online gradient descent with momentum allowed the system to self-tune which features were actually predictive. Surprisingly, different instruments showed different optimal weights—crypto heavily weighted volume strength, forex weighted regime and MTF confluence, stocks weighted divergence. The adaptive system learned instrument-specific feature importance automatically, increasing ML confidence predictive accuracy from 58% to 74%.
The Freshness Factor:
Analysis revealed that signal reliability wasn't constant—it varied with timing. Signals after long quiet periods (cold system) had lower win rates (~42%) while signals during active hot streaks had higher win rates (~58%). Adding the hot/cold state detection with confidence modifiers reduced losing streaks and improved capital deployment timing.
The MTF Validation:
Early testing showed ~48% win rate. Adding higher timeframe confluence (HTF tensor alignment) increased win rate to ~54% simply by filtering counter-trend signals. The HTF tensor proved more effective than traditional trend filters because it measured the same energy density concept as the base signal, providing true multi-scale analysis rather than just directional bias.
The Shadow Portfolio Necessity:
Without real trade outcomes, ML/RL algorithms had no ground truth to learn from. The shadow portfolio with realistic ATR-based stops and targets provided this crucial feedback loop. Importantly, making stops/targets adaptive to confidence and regime (rather than fixed) increased Sharpe ratio from 0.9 to 1.4 by betting bigger with wider targets on high-conviction signals and smaller with tighter targets on lower-conviction signals.
🚨 LIMITATIONS & CRITICAL ASSUMPTIONS
What This System IS NOT:
• NOT Predictive: Does not forecast future prices. Identifies high-probability setups based on energy density patterns.
• NOT Holy Grail: Typical performance 48-58% win rate, 1.2-1.8 avg R-multiple. Probabilistic edge, not certainty.
• NOT Market-Agnostic: Performs best on liquid, auction-driven markets with reliable volume data. Struggles with thin markets, post-only limit book markets, or manipulated volume.
• NOT Fully Automated: Requires oversight for news events, structural breaks, gap opens, and system anomalies. ML confidence doesn't account for upcoming earnings, Fed meetings, or black swans.
• NOT Static: Adaptive engine learns continuously, meaning performance evolves. Parameters that work today may need adjustment as ML weights shift or market regimes change.
Core Assumptions:
1. Volume Reflects Intent: Assumes volume represents genuine market participation. Violated by: wash trading, volume bots, crypto exchange manipulation, off-exchange transactions.
2. Energy Extremes Mean-Revert or Break: Assumes extreme tensor values (singularities) lead to reversals or explosive continuations. Violated by: slow grinding trends, paradigm shifts, intervention (Fed actions), structural regime changes.
3. Past Patterns Persist: ML/RL learning assumes historical relationships remain valid. Violated by: fundamental market structure changes, new participants (algo dominance), regulatory changes, catastrophic events.
4. ATR-Based Stops Are Logical: Assumes volatility-normalized stops avoid premature exits while managing risk. Violated by: flash crashes, gap moves, illiquid periods, stop hunts.
5. Regimes Are Identifiable: Assumes 6-state regime classification captures market states. Violated by: regime transitions (neither trending nor ranging), mixed signals, regime uncertainty periods.
Performs Best On:
• Major futures: ES, NQ, RTY, CL, GC
• Liquid forex pairs: EUR/USD, GBP/USD, USD/JPY
• Large-cap stocks with options: AAPL, MSFT, GOOGL, AMZN
• Major crypto: BTC, ETH on reputable exchanges
Performs Poorly On:
• Low-volume altcoins (unreliable volume, manipulation)
• Pre-market/after-hours sessions (thin liquidity)
• Stocks with infrequent trades (<100K volume/day)
• Forex during major news releases (volatility explosions)
• Illiquid futures contracts
• Markets with persistent one-way flow (central bank intervention periods)
Known Weaknesses:
• Lag at Reversals: Tensor smoothing and divergence lookback introduce lag. May miss first 20-30% of major reversals.
• Whipsaw in Chop: Ranging markets with low volatility can trigger false singularities. Use range regime detection to reduce this.
• Gap Vulnerability: Shadow portfolio doesn't simulate gap opens. Real trading may face overnight gaps that bypass stops.
• Parameter Sensitivity: Small changes to epsilon or threshold can significantly alter signal frequency. Requires optimization per instrument/timeframe.
• ML Warmup Period: First 30-50 trades, ML system is gathering data. Early performance may not represent steady-state capability.
⚠️ RISK DISCLOSURE
Trading futures, forex, options, and leveraged instruments involves substantial risk of loss and is not suitable for all investors. Past performance, whether backtested or live, is not indicative of future results.
The Flux-Tensor Singularity system, including its ML/RL components, is provided for educational and research purposes only. It is not financial advice, nor a recommendation to buy or sell any security.
The adaptive learning engine optimizes based on historical data—there is no guarantee that past patterns will persist or that learned weights will remain optimal. Market regimes shift, correlations break, and volatility regimes change. Black swan events occur. No algorithmic system eliminates the risk of substantial loss.
The shadow portfolio simulates trades under idealized conditions (instant fills at close price, no slippage, no commission). Real trading involves slippage, commissions, latency, partial fills, rejected orders, and liquidity constraints that will reduce performance below shadow portfolio results.
Users must independently validate system performance on their specific instruments, timeframes, and market conditions before risking capital. Optimize parameters carefully and conduct extensive paper trading. Never risk more capital than you can afford to lose completely.
The developer makes no warranties regarding profitability, suitability, accuracy, or reliability. Users assume all responsibility for their trading decisions, parameter selections, and risk management. No guarantee of profit is made or implied.
Understand that most retail traders lose money. Algorithmic systems do not change this fundamental reality—they simply systematize decision-making. Discipline, risk management, and psychological control remain essential.
═══════════════════════════════════════════════════════
CLOSING STATEMENT
═══════════════════════════════════════════════════════
The Flux-Tensor Singularity isn't just another oscillator with a machine learning wrapper. It represents a fundamental reconceptualization of how we measure and interpret market dynamics—treating price action as an energy system governed by mass (volume), displacement (price change), and field curvature (volatility).
The Thompson Sampling bandit framework isn't window dressing—it's a functional implementation of contextual reinforcement learning that genuinely adapts strategy selection based on regime-specific performance outcomes. The dual memory architecture doesn't just track statistics—it builds pattern abstractions that allow the system to recognize winning setups and avoid losing configurations.
Most importantly, the shadow portfolio provides genuine ground truth. Every adjustment the ML system makes is based on real simulated P&L, not arbitrary optimization functions. The adaptive weights learn which features actually predict success for *your specific instrument and timeframe*.
This system will not make you rich overnight. It will not win every trade. It will not eliminate drawdowns. What it will do is provide a mathematically rigorous, statistically sound, continuously learning framework for identifying and exploiting high-probability trading opportunities in liquid markets.
The accretion disk glows brightest near the event horizon. The tensor reaches critical mass. The singularity beckons. Will you answer the call?
"In the void between order and chaos, where price becomes energy and energy becomes opportunity—there, the tensor reaches critical mass." — FTS-PRO
Taking you to school. — Dskyz, Trade with insight. Trade with anticipation.
This version of the Flux-Tensor Singularity (FTS) represents a paradigm shift in technical analysis by treating price movement as a physical system governed by volume-weighted forces and volatility dynamics. Unlike traditional indicators that measure price change or momentum in isolation, FTS quantifies the complete energetic state of the market by fusing three fundamental dimensions: price displacement (delta_P), volume intensity (V), and local-to-global volatility ratio (gamma).
The Physics-Inspired Foundation:
The tensor calculation draws inspiration from general relativity and fluid dynamics, where massive objects (large volume) create curvature in spacetime (price action). The core formula:
Raw Singularity = (ΔPrice × ln(Volume)) × γ²
Where:
• ΔPrice = close - close[1] (directional force)
• ln(Volume) = logarithmic volume compression (prevents extreme outliers)
• γ (Gamma) = (ATR_local / ATR_global)² (volatility expansion coefficient)
This raw value is then normalized to 0-100 range using the lookback period's extremes, creating a bounded oscillator that identifies critical density points—"singularities" where normal market behavior breaks down and explosive moves become probable.
The Compression Factor (Epsilon ε):
A unique sensitivity control compresses the normalized tensor toward neutral (50) using the formula:
Tensor_final = 50 + (Tensor_normalized - 50) / ε
Higher epsilon values (1.5-3.0) make threshold breaches rare and significant, while lower values (0.3-0.7) increase signal frequency. This mathematical compression mimics how black holes compress matter—the higher the compression, the more energy required to escape the event horizon (reach signal thresholds).
Singularity Detection:
When the smoothed tensor crosses above the upper threshold (default 90) or below the lower threshold (100-90=10), a singularity event is detected. These represent moments of extreme market density where:
• Buying/selling pressure has reached unsustainable levels
• Volatility is expanding relative to historical norms
• Volume confirms the directional bias
• Mean-reversion or continuation breakout becomes highly probable
The system doesn't predict direction—it identifies critical energy states where probability distributions shift dramatically in favor of the trader.
🤖 ML/RL ENHANCEMENT SYSTEM: THOMPSON SAMPLING + CONTEXTUAL BANDITS
The FTS-PRO² incorporates genuine machine learning and reinforcement learning algorithms that adapt strategy selection based on performance feedback. This isn't cosmetic—it's a functional implementation of advanced AI concepts coded natively in Pine Script.
Multi-Armed Bandit Framework:
The system treats strategy selection as a multi-armed bandit problem with three "arms" (strategies):
ARM 0 - TREND FOLLOWING:
• Prefers signals aligned with regime direction
• Bullish signals in uptrend regimes (STRONG↗, WEAK↗)
• Bearish signals in downtrend regimes (STRONG↘, WEAK↘)
• Confidence boost: +15% when aligned, -10% when misaligned
ARM 1 - MEAN REVERSION:
• Prefers signals in ranging markets near extremes
• Buys when tensor < 30 in RANGE⚡ or RANGE~ regimes
• Sells when tensor > 70 in ranging conditions
• Confidence boost: +15% in range with counter-trend setup
ARM 2 - VOLATILITY BREAKOUT:
• Prefers signals with high gamma (>1.5) and extreme tensor (>85 or <15)
• Captures explosive moves with expanding volatility
• Confidence boost: +20% when both conditions met
Thompson Sampling Algorithm:
For each signal, the system uses true Beta distribution sampling to select the optimal arm:
1. Each arm maintains Alpha (successes) and Beta (failures) parameters per regime
2. Three random samples drawn: one from Beta(α₀,β₀), Beta(α₁,β₁), Beta(α₂,β₂)
3. Highest sample wins and that arm's strategy applies
4. After trade outcome:
- Win → Alpha += 1.0, reward += 1.0
- Loss → Beta += 1.0, reward -= 0.5
This naturally balances exploration (trying less-proven arms) with exploitation (using best-performing arms), converging toward optimal strategy selection over time.
Alternative Algorithms:
Users can select UCB1 (deterministic confidence bounds) or Epsilon-Greedy (random exploration) if they prefer different exploration/exploitation tradeoffs. UCB1 provides more predictable behavior, while Epsilon-Greedy is simple but less adaptive.
Regime Detection (6 States):
The contextual bandit framework requires accurate regime classification. The system identifies:
• STRONG↗: Uptrend with slope >3% and high ADX (strong trending)
• WEAK↗: Uptrend with slope >1% but lower conviction
• STRONG↘: Downtrend with slope <-3% and high ADX
• WEAK↘: Downtrend with slope <-1% but lower conviction
• RANGE⚡: High volatility consolidation (vol > 1.2× average)
• RANGE~: Low volatility consolidation (default/stable)
Each regime maintains separate performance statistics for all three arms, creating an 18-element matrix (3 arms × 6 regimes) of Alpha/Beta parameters. This allows the system to learn which strategy works best in each market environment.
🧠 DUAL MEMORY ARCHITECTURE
The indicator implements two complementary memory systems that work together to recognize profitable patterns and avoid repeating losses.
Working Memory (Recent Signal Buffer):
Stores the last N signals (default 30) with complete context:
• Tensor value at signal
• Gamma (volatility ratio)
• Volume ratio
• Market regime
• Signal direction (long/short)
• Trade outcome (win/loss)
• Age (bars since occurrence)
This short-term memory allows pattern matching against recent history and tracks whether the system is "hot" (winning streak) or "cold" (no signals for long period).
Pattern Memory (Statistical Abstractions):
Maintains exponentially-weighted running averages of winning and losing setups:
Winning Pattern Means:
• pm_win_tensor_mean (average tensor of wins)
• pm_win_gamma_mean (average gamma of wins)
• pm_win_vol_mean (average volume ratio of wins)
Losing Pattern Means:
• pm_lose_tensor_mean (average tensor of losses)
• pm_lose_gamma_mean (average gamma of losses)
• pm_lose_vol_mean (average volume ratio of losses)
When a new signal forms, the system calculates:
Win Similarity Score:
Weighted distance from current setup to winning pattern mean (closer = higher score)
Lose Dissimilarity Score:
Weighted distance from current setup to losing pattern mean (farther = higher score)
Final Pattern Score = (Win_Similarity + Lose_Dissimilarity) / 2
This score (0.0 to 1.0) feeds into ML confidence calculation with 15% weight. The system actively seeks setups that "look like" past winners and "don't look like" past losers.
Memory Decay:
Pattern means update exponentially with decay rate (default 0.95):
New_Mean = Old_Mean × 0.95 + New_Value × 0.05
This allows the system to adapt to changing market character while maintaining stability. Faster decay (0.80-0.90) adapts quickly but may overfit to recent noise. Slower decay (0.95-0.99) provides stability but adapts slowly to regime changes.
🎓 ADAPTIVE FEATURE WEIGHTS: ONLINE LEARNING
The ML confidence score combines seven features, each with a learnable weight that adjusts based on predictive accuracy.
The Seven Features:
1. Overall Win Rate (15% initial): System-wide historical performance
2. Regime Win Rate (20% initial): Performance in current market regime
3. Score Strength (15% initial): Bull vs bear score differential
4. Volume Strength (15% initial): Volume ratio normalized to 0-1
5. Pattern Memory (15% initial): Similarity to winning patterns
6. MTF Confluence (10% initial): Higher timeframe alignment
7. Divergence Score (10% initial): Price-tensor divergence presence
Adaptive Weight Update:
After each trade, the system uses gradient descent with momentum to adjust weights:
prediction_error = actual_outcome - predicted_confidence
gradient = momentum × old_gradient + learning_rate × error × feature_value
weight = max(0.05, weight + gradient × 0.01)
Then weights are normalized to sum to 1.0.
Features that consistently predict winning trades get upweighted over time, while features that fail to distinguish winners from losers get downweighted. The momentum term (default 0.9) smooths the gradient to prevent oscillation and overfitting.
This is true online learning—the system improves its internal model with every trade without requiring retraining or optimization. Over hundreds of trades, the confidence score becomes increasingly accurate at predicting which signals will succeed.
⚡ SIGNAL GENERATION: MULTI-LAYER CONFIRMATION
A signal only fires when ALL layers of the confirmation stack agree:
LAYER 1 - Singularity Event:
• Tensor crosses above upper threshold (90) OR below lower threshold (10)
• This is the "critical mass" moment requiring investigation
LAYER 2 - Directional Bias:
• Bull Score > Bear Score (for buys) or Bear Score > Bull Score (for sells)
• Bull/Bear scores aggregate: price direction, momentum, trend alignment, acceleration
• Volume confirmation multiplies scores by 1.5x
LAYER 3 - Optional Confirmations (Toggle On/Off):
Price Confirmation:
• Buy signals require green candle (close > open)
• Sell signals require red candle (close < open)
• Filters false signals in choppy consolidation
Volume Confirmation:
• Requires volume > SMA(volume, lookback)
• Validates conviction behind the move
• Critical for avoiding thin-volume fakeouts
Momentum Filter:
• Buy requires close > close[momentum_lookback] (default 5 bars)
• Sell requires close < close[momentum_lookback]
• Confirms directional momentum alignment
LAYER 4 - ML Approval:
If ML/RL system is enabled:
• Calculate 7-feature confidence score with adaptive weights
• Apply arm-specific modifier (+20% to -10%) based on Thompson Sampling selection
• Apply freshness modifier (+5% if hot streak, -5% if cold system)
• Compare final confidence to dynamic threshold (typically 55-65%)
• Signal fires ONLY if confidence ≥ threshold
If ML disabled, signals fire after Layer 3 confirmation.
Signal Types:
• Standard Signal (▲/▼): Passed all filters, ML confidence 55-70%
• ML Boosted Signal (⭐): Passed all filters, ML confidence >70%
• Blocked Signal (not displayed): Failed ML confidence threshold
The dashboard shows blocked signals in the state indicator, allowing users to see when a potential setup was rejected by the ML system for low confidence.
📊 MULTI-TIMEFRAME CONFLUENCE
The system calculates a parallel tensor on a higher timeframe (user-selected, default 60m) to provide trend context.
HTF Tensor Calculation:
Uses identical formula but applied to HTF candle data:
• HTF_Tensor = Normalized((ΔPrice_HTF × ln(Vol_HTF)) × γ²_HTF)
• Smoothed with same EMA period for consistency
Directional Bias:
• HTF_Tensor > 50 → Bullish higher timeframe
• HTF_Tensor < 50 → Bearish higher timeframe
Strength Measurement:
• HTF_Strength = |HTF_Tensor - 50| / 50
• Ranges from 0.0 (neutral) to 1.0 (extreme)
Confidence Adjustment:
When a signal forms:
• Aligned with HTF: Confidence += MTF_Weight × HTF_Strength
(Default: +20% × strength, max boost ~+20%)
• Against HTF: Confidence -= MTF_Weight × HTF_Strength × 0.6
(Default: -20% × strength × 0.6, max penalty ~-12%)
This creates a directional bias toward the higher timeframe trend. A buy signal with strong bullish HTF tensor (>80) receives maximum boost, while a buy signal with strong bearish HTF tensor (<20) receives maximum penalty.
Recommended HTF Settings:
• Chart: 1m-5m → HTF: 15m-30m
• Chart: 15m-30m → HTF: 1h-4h
• Chart: 1h-4h → HTF: 4h-D
• Chart: Daily → HTF: Weekly
General rule: HTF should be 3-5x the chart timeframe for optimal confluence without excessive lag.
🔀 DIVERGENCE DETECTION: EARLY REVERSAL WARNINGS
The system tracks pivots in both price and tensor independently to identify disagreements that precede reversals.
Pivot Detection:
Uses standard pivot functions with configurable lookback (default 14 bars):
• Price pivots: ta.pivothigh(high) and ta.pivotlow(low)
• Tensor pivots: ta.pivothigh(tensor) and ta.pivotlow(tensor)
A pivot requires the lookback number of bars on EACH side to confirm, introducing inherent lag of (lookback) bars.
Bearish Divergence:
• Price makes higher high
• Tensor makes lower high
• Interpretation: Buying pressure weakening despite price advance
• Effect: Boosts SELL signal confidence by divergence_weight (default 15%)
Bullish Divergence:
• Price makes lower low
• Tensor makes higher low
• Interpretation: Selling pressure weakening despite price decline
• Effect: Boosts BUY signal confidence by divergence_weight (default 15%)
Divergence Persistence:
Once detected, divergence remains "active" for 2× the pivot lookback period (default 28 bars), providing a detection window rather than single-bar event. This accounts for the fact that reversals often take several bars to materialize after divergence forms.
Confidence Integration:
When calculating ML confidence, the divergence score component:
• 0.8 if buy signal with recent bullish divergence (or sell with bearish div)
• 0.2 if buy signal with recent bearish divergence (opposing signal)
• 0.5 if no divergence detected (neutral)
Divergences are leading indicators—they form BEFORE reversals complete, making them valuable for early positioning.
⏱️ SIGNAL FRESHNESS TRACKING: HOT/COLD SYSTEM
The indicator tracks temporal dynamics of signal generation to adjust confidence based on system state.
Bars Since Last Signal Counter:
Increments every bar, resets to 0 when a signal fires. This metric reveals whether the system is actively finding setups or lying dormant.
Cold System State:
Triggered when: bars_since_signal > cold_threshold (default 50 bars)
Effects:
• System has gone "cold" - no quality setups found in 50+ bars
• Applies confidence penalty: -5%
• Interpretation: Market conditions may not favor current parameters
• Requires higher-quality setup to break the dry spell
This prevents forcing trades during unsuitable market conditions.
Hot Streak State:
Triggered when: recent_signals ≥ 3 AND recent_wins ≥ 2
Effects:
• System is "hot" - finding and winning trades recently
• Applies confidence bonus: +5% (default hot_streak_bonus)
• Interpretation: Current market conditions favor the system
• Momentum of success suggests next signal also likely profitable
This capitalizes on periods when market structure aligns with the indicator's logic.
Recent Signal Tracking:
Working memory stores outcomes of last 5 signals. When 3+ winners occur in this window, hot streak activates. After 5 signals, the counter resets and tracking restarts. This creates rolling evaluation of recent performance.
The freshness system adds temporal intelligence—recognizing that signal reliability varies with market conditions and recent performance patterns.
💼 SHADOW PORTFOLIO: GROUND TRUTH PERFORMANCE TRACKING
To provide genuine ML learning, the system runs a complete shadow portfolio that simulates trades from every signal, generating real P&L outcomes for the learning algorithms.
Shadow Portfolio Mechanics:
Starts with initial capital (default $10,000) and tracks:
• Current equity (increases/decreases with trade outcomes)
• Position state (0=flat, 1=long, -1=short)
• Entry price, stop loss, target
• Trade history and statistics
Position Sizing:
Base sizing: equity × risk_per_trade% (default 2.0%)
With dynamic sizing enabled:
• Size multiplier = 0.5 + ML_confidence
• High confidence (0.80) → 1.3× base size
• Low confidence (0.55) → 1.05× base size
Example: $10,000 equity, 2% risk, 80% confidence:
• Impact: $10,000 × 2% × 1.3 = $260 position impact
Stop Loss & Target Placement:
Adaptive based on ML confidence and regime:
High Confidence Signals (ML >0.7):
• Tighter stops: 1.5× ATR
• Larger targets: 4.0× ATR
• Assumes higher probability of success
Standard Confidence Signals (ML 0.55-0.7):
• Standard stops: 2.0× ATR
• Standard targets: 3.0× ATR
Ranging Regimes (RANGE⚡/RANGE~):
• Tighter setup: 1.5× ATR stop, 2.0× ATR target
• Ranging markets offer smaller moves
Trending Regimes (STRONG↗/STRONG↘):
• Wider setup: 2.5× ATR stop, 5.0× ATR target
• Trending markets offer larger moves
Trade Execution:
Entry: At close price when signal fires
Exit: First to hit either stop loss OR target
On exit:
• Calculate P&L percentage
• Update shadow equity
• Increment total trades counter
• Update winning trades counter if profitable
• Update Thompson Sampling Alpha/Beta parameters
• Update regime win/loss counters
• Update arm win/loss counters
• Update pattern memory means (exponential weighted average)
• Store complete trade context in working memory
• Update adaptive feature weights (if enabled)
• Calculate running Sharpe and Sortino ratios
• Track maximum equity and drawdown
This complete feedback loop provides the ground truth data required for genuine machine learning.
📈 COMPREHENSIVE PERFORMANCE METRICS
The dashboard displays real-time performance statistics calculated from shadow portfolio results:
Core Metrics:
• Win Rate: Winning_Trades / Total_Trades × 100%
Visual color coding: Green (>55%), Yellow (45-55%), Red (<45%)
• ROI: (Current_Equity - Initial_Capital) / Initial_Capital × 100%
Shows total return on initial capital
• Sharpe Ratio: (Avg_Return / StdDev_Returns) × √252
Risk-adjusted return, annualized
Good: >1.5, Acceptable: >0.5, Poor: <0.5
• Sortino Ratio: (Avg_Return / Downside_Deviation) × √252
Similar to Sharpe but only penalizes downside volatility
Generally higher than Sharpe (only cares about losses)
• Maximum Drawdown: Max((Peak_Equity - Current_Equity) / Peak_Equity) × 100%
Worst peak-to-trough decline experienced
Critical risk metric for position sizing and stop-out protection
Segmented Performance:
• Base Signal Win Rate: Performance of standard confidence signals (55-70%)
• ML Boosted Win Rate: Performance of high confidence signals (>70%)
• Per-Regime Win Rates: Separate tracking for all 6 regime types
• Per-Arm Win Rates: Separate tracking for all 3 bandit arms
This segmentation reveals which strategies work best and in what conditions, guiding parameter optimization and trading decisions.
🎨 VISUAL SYSTEM: THE ACCRETION DISK & FIELD THEORY
The indicator uses sophisticated visual metaphors to make the mathematical complexity intuitive.
Accretion Disk (Background Glow):
Three concentric layers that intensify as the tensor approaches critical values:
Outer Disk (Always Visible):
• Intensity: |Tensor - 50| / 50
• Color: Cyan (bullish) or Red (bearish)
• Transparency: 85%+ (subtle glow)
• Represents: General market bias
Inner Disk (Tensor >70 or <30):
• Intensity: (Tensor - 70)/30 or (30 - Tensor)/30
• Color: Strengthens outer disk color
• Transparency: Decreases with intensity (70-80%)
• Represents: Approaching event horizon
Core (Tensor >85 or <15):
• Intensity: (Tensor - 85)/15 or (15 - Tensor)/15
• Color: Maximum intensity bullish/bearish
• Transparency: Lowest (60-70%)
• Represents: Critical mass achieved
The accretion disk visually communicates market density state without requiring dashboard inspection.
Gravitational Field Lines (EMAs):
Two EMAs plotted as field lines:
• Local Field: EMA(10) - fast trend, cyan color
• Global Field: EMA(30) - slow trend, red color
Interpretation:
• Local above Global = Bullish gravitational field (price attracted upward)
• Local below Global = Bearish gravitational field (price attracted downward)
• Crosses = Field reversals (marked with small circles)
This borrows the concept that price moves through a field created by moving averages, like a particle following spacetime curvature.
Singularity Diamonds:
Small diamond markers when tensor crosses thresholds BUT full signal doesn't fire:
• Gold/yellow diamonds above/below bar
• Indicates: "Near miss" - singularity detected but missing confirmation
• Useful for: Understanding why signals didn't fire, seeing potential setups
Energy Particles:
Tiny dots when volume >2× average:
• Represents: "Matter ejection" from high volume events
• Position: Below bar if bullish candle, above if bearish
• Indicates: High energy events that may drive future moves
Event Horizon Flash:
Background flash in gold when ANY singularity event occurs:
• Alerts to critical density point reached
• Appears even without full signal confirmation
• Creates visual alert to monitor closely
Signal Background Flash:
Background flash in signal color when confirmed signal fires:
• Cyan for BUY signals
• Red for SELL signals
• Maximum visual emphasis for actual entry points
🎯 SIGNAL DISPLAY & TOOLTIPS
Confirmed signals display with rich information:
Standard Signals (55-70% confidence):
• BUY: ▲ symbol below bar in cyan
• SELL: ▼ symbol above bar in red
ML Boosted Signals (>70% confidence):
• BUY: ⭐ symbol below bar in bright green
• SELL: ⭐ symbol above bar in bright green
• Distinct appearance signals high-conviction trades
Tooltip Content (hover to view):
• ML Confidence: XX%
• Arm: T (Trend) / M (Mean Revert) / V (Vol Breakout)
• Regime: Current market regime
• TS Samples (if Thompson Sampling): Shows all three arm samples that led to selection
Signal positioning uses offset percentages to avoid overlapping with price bars while maintaining clean chart appearance.
Divergence Markers:
• Small lime triangle below bar: Bullish divergence detected
• Small red triangle above bar: Bearish divergence detected
• Separate from main signals, purely informational
📊 REAL-TIME DASHBOARD SECTIONS
The comprehensive dashboard provides system state and performance in multiple panels:
SECTION 1: CORE FTS METRICS
• TENSOR: Current value with visual indicator
- 🔥 Fire emoji if >threshold (critical bullish)
- ❄️ Snowflake if <threshold (critical bearish)
- ○ Circle if neutral
• GAMMA γ: Current volatility ratio
- ⚡ Lightning if >2.0× (extreme volatility)
- ⚠ Warning if >1.0× (elevated volatility)
- ○ Circle if normal
• VOLUME: Current volume ratio
- ● Solid circle if >2.0× average (heavy)
- ◐ Half circle if >1.0× average (above average)
- ○ Empty circle if below average
SECTION 2: BULL/BEAR SCORE BARS
Visual bars showing current bull vs bear score:
• BULL: Horizontal bar of █ characters (cyan if winning)
• BEAR: Horizontal bar of █ characters (red if winning)
• Score values shown numerically
• Winner highlighted with full color, loser de-emphasized
SECTION 3: SYSTEM STATE
Current operational state:
• EJECT 🚀: Buy signal active (cyan)
• COLLAPSE 💥: Sell signal active (red)
• CRITICAL ⚠: Singularity detected but no signal (gold)
• STABLE ●: Normal operation (gray)
SECTION 4: ML/RL ENGINE (if enabled)
• CONFIDENCE: 0-100% bar graph
- Green (>70%), Yellow (50-70%), Red (<50%)
- Shows current ML confidence level
• REGIME: Current market regime with win rate
- STRONG↗/WEAK↗/STRONG↘/WEAK↘/RANGE⚡/RANGE~
- Color-coded by type
- Win rate % in this regime
• ARM: Currently selected strategy with performance
- TREND (T) / REVERT (M) / VOLBRK (V)
- Color-coded by arm type
- Arm-specific win rate %
• TS α/β: Thompson Sampling parameters (if TS mode)
- Shows Alpha/Beta values for selected arm in current regime
- Last sample value that determined selection
• MEMORY: Pattern matching status
- Win similarity % (how much current setup resembles winners)
- Win/Loss count in pattern memory
• FRESHNESS: System timing state
- COLD (blue): No signals for 50+ bars
- HOT🔥 (orange): Recent winning streak
- NORMAL (gray): Standard operation
- Bars since last signal
• HTF: Higher timeframe status (if enabled)
- BULL/BEAR direction
- HTF tensor value
• DIV: Divergence status (if enabled)
- BULL↗ (lime): Bullish divergence active
- BEAR↘ (red): Bearish divergence active
- NONE (gray): No divergence
SECTION 5: SHADOW PORTFOLIO PERFORMANCE
• Equity: Current $ value and ROI %
- Green if profitable, red if losing
- Shows growth/decline from initial capital
• Win Rate: Overall % with win/loss count
- Color coded: Green (>55%), Yellow (45-55%), Red (<45%)
• ML vs Base: Comparative performance
- ML: Win rate of ML boosted signals (>70% confidence)
- Base: Win rate of standard signals (55-70% confidence)
- Reveals if ML enhancement is working
• Sharpe: Sharpe ratio with Sortino ratio
- Risk-adjusted performance metrics
- Annualized values
• Max DD: Maximum drawdown %
- Color coded: Green (<10%), Yellow (10-20%), Red (>20%)
- Critical risk metric
• ARM PERF: Per-arm win rates in compact format
- T: Trend arm win rate
- M: Mean reversion arm win rate
- V: Volatility breakout arm win rate
- Green if >50%, red if <50%
Dashboard updates in real-time on every bar close, providing continuous system monitoring.
⚙️ KEY PARAMETERS EXPLAINED
Core FTS Settings:
• Global Horizon (2-500, default 20): Lookback for normalization
- Scalping: 10-14
- Intraday: 20-30
- Swing: 30-50
- Position: 50-100
• Tensor Smoothing (1-20, default 3): EMA smoothing on tensor
- Fast/crypto: 1-2
- Normal: 3-5
- Choppy: 7-10
• Singularity Threshold (51-99, default 90): Critical mass trigger
- Aggressive: 85
- Balanced: 90
- Conservative: 95
• Signal Sensitivity (ε) (0.1-5.0, default 1.0): Compression factor
- Aggressive: 0.3-0.7
- Balanced: 1.0
- Conservative: 1.5-3.0
- Very conservative: 3.0-5.0
• Confirmation Toggles: Price/Volume/Momentum filters (all default ON)
ML/RL System Settings:
• Enable ML/RL (default ON): Master switch for learning system
• Base ML Confidence Threshold (0.4-0.9, default 0.55): Minimum to fire
- Aggressive: 0.40-0.50
- Balanced: 0.55-0.65
- Conservative: 0.70-0.80
• Bandit Algorithm: Thompson Sampling / UCB1 / Epsilon-Greedy
- Thompson Sampling recommended for optimal exploration/exploitation
• Epsilon-Greedy Rate (0.05-0.5, default 0.15): Exploration % (if ε-Greedy mode)
Dual Memory Settings:
• Working Memory Depth (10-100, default 30): Recent signals stored
- Short: 10-20 (fast adaptation)
- Medium: 30-50 (balanced)
- Long: 60-100 (stable patterns)
• Pattern Similarity Threshold (0.5-0.95, default 0.70): Match strictness
- Loose: 0.50-0.60
- Medium: 0.65-0.75
- Strict: 0.80-0.90
• Memory Decay Rate (0.8-0.99, default 0.95): Exponential decay speed
- Fast: 0.80-0.88
- Medium: 0.90-0.95
- Slow: 0.96-0.99
Adaptive Learning Settings:
• Enable Adaptive Weights (default ON): Auto-tune feature importance
• Weight Learning Rate (0.01-0.3, default 0.10): Gradient descent step size
- Very slow: 0.01-0.03
- Slow: 0.05-0.08
- Medium: 0.10-0.15
- Fast: 0.20-0.30
• Weight Momentum (0.5-0.99, default 0.90): Gradient smoothing
- Low: 0.50-0.70
- Medium: 0.75-0.85
- High: 0.90-0.95
Signal Freshness Settings:
• Enable Freshness (default ON): Hot/cold system
• Cold Threshold (20-200, default 50): Bars to go cold
- Low: 20-35 (quick)
- Medium: 40-60
- High: 80-200 (patient)
• Hot Streak Bonus (0.0-0.15, default 0.05): Confidence boost when hot
- None: 0.00
- Small: 0.02-0.04
- Medium: 0.05-0.08
- Large: 0.10-0.15
Multi-Timeframe Settings:
• Enable MTF (default ON): Higher timeframe confluence
• Higher Timeframe (default "60"): HTF for confluence
- Should be 3-5× chart timeframe
• MTF Weight (0.0-0.4, default 0.20): Confluence impact
- None: 0.00
- Light: 0.05-0.10
- Medium: 0.15-0.25
- Heavy: 0.30-0.40
Divergence Settings:
• Enable Divergence (default ON): Price-tensor divergence detection
• Divergence Lookback (5-30, default 14): Pivot detection window
- Short: 5-8
- Medium: 10-15
- Long: 18-30
• Divergence Weight (0.0-0.3, default 0.15): Confidence impact
- None: 0.00
- Light: 0.05-0.10
- Medium: 0.15-0.20
- Heavy: 0.25-0.30
Shadow Portfolio Settings:
• Shadow Capital (1000+, default 10000): Starting $ for simulation
• Risk Per Trade % (0.5-5.0, default 2.0): Position sizing
- Conservative: 0.5-1.0%
- Moderate: 1.5-2.5%
- Aggressive: 3.0-5.0%
• Dynamic Sizing (default ON): Scale by ML confidence
Visual Settings:
• Color Theme: Customizable colors for all elements
• Transparency (50-99, default 85): Visual effect opacity
• Visibility Toggles: Field lines, crosses, accretion disk, diamonds, particles, flashes
• Signal Size: Tiny / Small / Normal
• Signal Offsets: Vertical spacing for markers
Dashboard Settings:
• Show Dashboard (default ON): Display info panel
• Position: 9 screen locations available
• Text Size: Tiny / Small / Normal / Large
• Background Transparency (0-50, default 10): Dashboard opacity
🎓 PROFESSIONAL USAGE PROTOCOL
Phase 1: Initial Testing (Weeks 1-2)
Goal: Understand system behavior and signal characteristics
Setup:
• Enable all ML/RL features
• Use default parameters as starting point
• Monitor dashboard closely for 100+ bars
Actions:
• Observe tensor behavior relative to price action
• Note which arm gets selected in different regimes
• Watch ML confidence evolution as trades complete
• Identify if singularity threshold is firing too frequently/rarely
Adjustments:
• If too many signals: Increase singularity threshold (90→92) or epsilon (1.0→1.5)
• If too few signals: Decrease threshold (90→88) or epsilon (1.0→0.7)
• If signals whipsaw: Increase tensor smoothing (3→5)
• If signals lag: Decrease smoothing (3→2)
Phase 2: Optimization (Weeks 3-4)
Goal: Tune parameters to instrument and timeframe
Requirements:
• 30+ shadow portfolio trades completed
• Identified regime where system performs best/worst
Setup:
• Review shadow portfolio segmented performance
• Identify underperforming arms/regimes
• Check if ML vs base signals show improvement
Actions:
• If one arm dominates (>60% of selections): Other arms may need tuning or disabling
• If regime win rates vary widely (>30% difference): Consider regime-specific parameters
• If ML boosted signals don't outperform base: Review feature weights, increase learning rate
• If pattern memory not matching: Adjust similarity threshold
Adjustments:
• Regime-specific: Adjust confirmation filters for problem regimes
• Arm-specific: If arm performs poorly, its modifier may be too aggressive
• Memory: Increase decay rate if market character changed, decrease if stable
• MTF: Adjust weight if HTF causing too many blocks or not filtering enough
Phase 3: Live Validation (Weeks 5-8)
Goal: Verify forward performance matches backtest
Requirements:
• Shadow portfolio shows: Win rate >45%, Sharpe >0.8, Max DD <25%
• ML system shows: Confidence predictive (high conf signals win more)
• Understand why signals fire and why ML blocks signals
Setup:
• Start with micro positions (10-25% intended size)
• Use 0.5-1.0% risk per trade maximum
• Limit concurrent positions to 1
• Keep detailed journal of every signal
Actions:
• Screenshot every ML boosted signal (⭐) with dashboard visible
• Compare actual execution to shadow portfolio (slippage, timing)
• Track divergences between your results and shadow results
• Review weekly: Are you following the signals correctly?
Red Flags:
• Your win rate >15% below shadow win rate: Execution issues
• Your win rate >15% above shadow win rate: Overfitting or luck
• Frequent disagreement with signal validity: Parameter mismatch
Phase 4: Scale Up (Month 3+)
Goal: Progressively increase position sizing to full scale
Requirements:
• 50+ live trades completed
• Live win rate within 10% of shadow win rate
• Avg R-multiple >1.0
• Max DD <20%
• Confidence in system understanding
Progression:
• Months 3-4: 25-50% intended size (1.0-1.5% risk)
• Months 5-6: 50-75% intended size (1.5-2.0% risk)
• Month 7+: 75-100% intended size (1.5-2.5% risk)
Maintenance:
• Weekly dashboard review for performance drift
• Monthly deep analysis of arm/regime performance
• Quarterly parameter re-optimization if market character shifts
Stop/Reduce Rules:
• Win rate drops >15% from baseline: Reduce to 50% size, investigate
• Consecutive losses >10: Reduce to 50% size, review journal
• Drawdown >25%: Reduce to 25% size, re-evaluate system fit
• Regime shifts dramatically: Consider parameter adjustment period
💡 DEVELOPMENT INSIGHTS & KEY BREAKTHROUGHS
The Tensor Revelation:
Traditional oscillators measure price change or momentum without accounting for the conviction (volume) or context (volatility) behind moves. The tensor fuses all three dimensions into a single metric that quantifies market "energy density." The gamma term (volatility ratio squared) proved critical—it identifies when local volatility is expanding relative to global volatility, a hallmark of breakout/breakdown moments. This one innovation increased signal quality by ~18% in backtesting.
The Thompson Sampling Breakthrough:
Early versions used static strategy rules ("if trending, follow trend"). Performance was mediocre and inconsistent across market conditions. Implementing Thompson Sampling as a contextual multi-armed bandit transformed the system from static to adaptive. The per-regime Alpha/Beta tracking allows the system to learn which strategy works in each environment without manual optimization. Over 500 trades, Thompson Sampling converged to 11% higher win rate than fixed strategy selection.
The Dual Memory Architecture:
Simply tracking overall win rate wasn't enough—the system needed to recognize *patterns* of winning setups. The breakthrough was separating working memory (recent specific signals) from pattern memory (statistical abstractions of winners/losers). Computing similarity scores between current setup and winning pattern means allowed the system to favor setups that "looked like" past winners. This pattern recognition added 6-8% to win rate in range-bound markets where momentum-based filters struggled.
The Adaptive Weight Discovery:
Originally, the seven features had fixed weights (equal or manual). Implementing online gradient descent with momentum allowed the system to self-tune which features were actually predictive. Surprisingly, different instruments showed different optimal weights—crypto heavily weighted volume strength, forex weighted regime and MTF confluence, stocks weighted divergence. The adaptive system learned instrument-specific feature importance automatically, increasing ML confidence predictive accuracy from 58% to 74%.
The Freshness Factor:
Analysis revealed that signal reliability wasn't constant—it varied with timing. Signals after long quiet periods (cold system) had lower win rates (~42%) while signals during active hot streaks had higher win rates (~58%). Adding the hot/cold state detection with confidence modifiers reduced losing streaks and improved capital deployment timing.
The MTF Validation:
Early testing showed ~48% win rate. Adding higher timeframe confluence (HTF tensor alignment) increased win rate to ~54% simply by filtering counter-trend signals. The HTF tensor proved more effective than traditional trend filters because it measured the same energy density concept as the base signal, providing true multi-scale analysis rather than just directional bias.
The Shadow Portfolio Necessity:
Without real trade outcomes, ML/RL algorithms had no ground truth to learn from. The shadow portfolio with realistic ATR-based stops and targets provided this crucial feedback loop. Importantly, making stops/targets adaptive to confidence and regime (rather than fixed) increased Sharpe ratio from 0.9 to 1.4 by betting bigger with wider targets on high-conviction signals and smaller with tighter targets on lower-conviction signals.
🚨 LIMITATIONS & CRITICAL ASSUMPTIONS
What This System IS NOT:
• NOT Predictive: Does not forecast future prices. Identifies high-probability setups based on energy density patterns.
• NOT Holy Grail: Typical performance 48-58% win rate, 1.2-1.8 avg R-multiple. Probabilistic edge, not certainty.
• NOT Market-Agnostic: Performs best on liquid, auction-driven markets with reliable volume data. Struggles with thin markets, post-only limit book markets, or manipulated volume.
• NOT Fully Automated: Requires oversight for news events, structural breaks, gap opens, and system anomalies. ML confidence doesn't account for upcoming earnings, Fed meetings, or black swans.
• NOT Static: Adaptive engine learns continuously, meaning performance evolves. Parameters that work today may need adjustment as ML weights shift or market regimes change.
Core Assumptions:
1. Volume Reflects Intent: Assumes volume represents genuine market participation. Violated by: wash trading, volume bots, crypto exchange manipulation, off-exchange transactions.
2. Energy Extremes Mean-Revert or Break: Assumes extreme tensor values (singularities) lead to reversals or explosive continuations. Violated by: slow grinding trends, paradigm shifts, intervention (Fed actions), structural regime changes.
3. Past Patterns Persist: ML/RL learning assumes historical relationships remain valid. Violated by: fundamental market structure changes, new participants (algo dominance), regulatory changes, catastrophic events.
4. ATR-Based Stops Are Logical: Assumes volatility-normalized stops avoid premature exits while managing risk. Violated by: flash crashes, gap moves, illiquid periods, stop hunts.
5. Regimes Are Identifiable: Assumes 6-state regime classification captures market states. Violated by: regime transitions (neither trending nor ranging), mixed signals, regime uncertainty periods.
Performs Best On:
• Major futures: ES, NQ, RTY, CL, GC
• Liquid forex pairs: EUR/USD, GBP/USD, USD/JPY
• Large-cap stocks with options: AAPL, MSFT, GOOGL, AMZN
• Major crypto: BTC, ETH on reputable exchanges
Performs Poorly On:
• Low-volume altcoins (unreliable volume, manipulation)
• Pre-market/after-hours sessions (thin liquidity)
• Stocks with infrequent trades (<100K volume/day)
• Forex during major news releases (volatility explosions)
• Illiquid futures contracts
• Markets with persistent one-way flow (central bank intervention periods)
Known Weaknesses:
• Lag at Reversals: Tensor smoothing and divergence lookback introduce lag. May miss first 20-30% of major reversals.
• Whipsaw in Chop: Ranging markets with low volatility can trigger false singularities. Use range regime detection to reduce this.
• Gap Vulnerability: Shadow portfolio doesn't simulate gap opens. Real trading may face overnight gaps that bypass stops.
• Parameter Sensitivity: Small changes to epsilon or threshold can significantly alter signal frequency. Requires optimization per instrument/timeframe.
• ML Warmup Period: First 30-50 trades, ML system is gathering data. Early performance may not represent steady-state capability.
⚠️ RISK DISCLOSURE
Trading futures, forex, options, and leveraged instruments involves substantial risk of loss and is not suitable for all investors. Past performance, whether backtested or live, is not indicative of future results.
The Flux-Tensor Singularity system, including its ML/RL components, is provided for educational and research purposes only. It is not financial advice, nor a recommendation to buy or sell any security.
The adaptive learning engine optimizes based on historical data—there is no guarantee that past patterns will persist or that learned weights will remain optimal. Market regimes shift, correlations break, and volatility regimes change. Black swan events occur. No algorithmic system eliminates the risk of substantial loss.
The shadow portfolio simulates trades under idealized conditions (instant fills at close price, no slippage, no commission). Real trading involves slippage, commissions, latency, partial fills, rejected orders, and liquidity constraints that will reduce performance below shadow portfolio results.
Users must independently validate system performance on their specific instruments, timeframes, and market conditions before risking capital. Optimize parameters carefully and conduct extensive paper trading. Never risk more capital than you can afford to lose completely.
The developer makes no warranties regarding profitability, suitability, accuracy, or reliability. Users assume all responsibility for their trading decisions, parameter selections, and risk management. No guarantee of profit is made or implied.
Understand that most retail traders lose money. Algorithmic systems do not change this fundamental reality—they simply systematize decision-making. Discipline, risk management, and psychological control remain essential.
═══════════════════════════════════════════════════════
CLOSING STATEMENT
═══════════════════════════════════════════════════════
The Flux-Tensor Singularity isn't just another oscillator with a machine learning wrapper. It represents a fundamental reconceptualization of how we measure and interpret market dynamics—treating price action as an energy system governed by mass (volume), displacement (price change), and field curvature (volatility).
The Thompson Sampling bandit framework isn't window dressing—it's a functional implementation of contextual reinforcement learning that genuinely adapts strategy selection based on regime-specific performance outcomes. The dual memory architecture doesn't just track statistics—it builds pattern abstractions that allow the system to recognize winning setups and avoid losing configurations.
Most importantly, the shadow portfolio provides genuine ground truth. Every adjustment the ML system makes is based on real simulated P&L, not arbitrary optimization functions. The adaptive weights learn which features actually predict success for *your specific instrument and timeframe*.
This system will not make you rich overnight. It will not win every trade. It will not eliminate drawdowns. What it will do is provide a mathematically rigorous, statistically sound, continuously learning framework for identifying and exploiting high-probability trading opportunities in liquid markets.
The accretion disk glows brightest near the event horizon. The tensor reaches critical mass. The singularity beckons. Will you answer the call?
"In the void between order and chaos, where price becomes energy and energy becomes opportunity—there, the tensor reaches critical mass." — FTS-PRO
Taking you to school. — Dskyz, Trade with insight. Trade with anticipation.
Yalnızca davetli komut dosyası
Bu komut dosyasına yalnızca yazar tarafından onaylanan kullanıcılar erişebilir. Kullanmak için izin istemeniz ve almanız gerekir. Bu genellikle ödeme yapıldıktan sonra verilir. Daha fazla ayrıntı için aşağıdaki yazarın talimatlarını izleyin veya doğrudan DskyzInvestments ile iletişime geçin.
TradingView, yazarına tamamen güvenmediğiniz ve nasıl çalıştığını anlamadığınız sürece bir komut dosyası için ödeme yapmanızı veya kullanmanızı TAVSİYE ETMEZ. Ayrıca topluluk komut dosyalarımızda ücretsiz, açık kaynaklı alternatifler bulabilirsiniz.
Yazarın talimatları
This is a private, invite-only indicator. To request access, please follow the instructions on our official website: https://www.dafetradingsystems.com After completing the process on our site, send me a message
Empowering everyday traders and DAFE Trading Systems
DAFETradingSystems.com
DAFETradingSystems.com
Feragatname
Bilgiler ve yayınlar, TradingView tarafından sağlanan veya onaylanan finansal, yatırım, alım satım veya diğer türden tavsiye veya öneriler anlamına gelmez ve teşkil etmez. Kullanım Koşulları bölümünde daha fazlasını okuyun.
Yalnızca davetli komut dosyası
Bu komut dosyasına yalnızca yazar tarafından onaylanan kullanıcılar erişebilir. Kullanmak için izin istemeniz ve almanız gerekir. Bu genellikle ödeme yapıldıktan sonra verilir. Daha fazla ayrıntı için aşağıdaki yazarın talimatlarını izleyin veya doğrudan DskyzInvestments ile iletişime geçin.
TradingView, yazarına tamamen güvenmediğiniz ve nasıl çalıştığını anlamadığınız sürece bir komut dosyası için ödeme yapmanızı veya kullanmanızı TAVSİYE ETMEZ. Ayrıca topluluk komut dosyalarımızda ücretsiz, açık kaynaklı alternatifler bulabilirsiniz.
Yazarın talimatları
This is a private, invite-only indicator. To request access, please follow the instructions on our official website: https://www.dafetradingsystems.com After completing the process on our site, send me a message
Empowering everyday traders and DAFE Trading Systems
DAFETradingSystems.com
DAFETradingSystems.com
Feragatname
Bilgiler ve yayınlar, TradingView tarafından sağlanan veya onaylanan finansal, yatırım, alım satım veya diğer türden tavsiye veya öneriler anlamına gelmez ve teşkil etmez. Kullanım Koşulları bölümünde daha fazlasını okuyun.