hedgequantx 2.5.35 → 2.5.37

package/src/services/ai/strategy-supervisor.js

@@ -0,0 +1,675 @@
+/**
+ * AI Strategy Supervisor
+ * 
+ * Observes, learns from, and optimizes the HQX Ultra Scalping strategy in real-time.
+ * 
+ * FUNCTIONS:
+ * 1. OBSERVE - Receive all market data, signals, and trades in real-time
+ * 2. LEARN - Analyze winning/losing trades to identify patterns
+ * 3. OPTIMIZE - Suggest and apply parameter improvements
+ * 4. SUPERVISE - Monitor risk and intervene when necessary
+ * 
+ * In CONSENSUS mode (2+ agents), ALL agents must agree before applying changes.
+ */
+
+const { analyzePerformance, getMarketAdvice, callAI } = require('./client');
+
+// Singleton supervisor state
+let supervisorState = {
+  active: false,
+  agents: [],
+  strategy: null,
+  service: null,
+  accountId: null,
+  
+  // Real-time data (synced with strategy)
+  ticks: [],
+  signals: [],
+  trades: [],
+  
+  // Learning data
+  winningPatterns: [],
+  losingPatterns: [],
+  
+  // Performance tracking
+  performance: {
+    trades: 0,
+    wins: 0,
+    losses: 0,
+    totalPnL: 0,
+    maxDrawdown: 0,
+    currentDrawdown: 0,
+    peakPnL: 0,
+    winStreak: 0,
+    lossStreak: 0,
+    maxWinStreak: 0,
+    maxLossStreak: 0
+  },
+  
+  // Optimization state
+  optimizations: [],
+  lastOptimizationTime: 0,
+  optimizationInterval: 60000, // Analyze every 60 seconds
+  
+  // Current recommendations
+  currentAdvice: {
+    action: 'NORMAL',
+    sizeMultiplier: 1.0,
+    reason: 'Starting'
+  }
+};
+
+// Analysis interval
+let analysisInterval = null;
+
+/**
+ * Initialize supervisor with strategy and agents
+ */
+const initialize = (strategy, agents, service, accountId) => {
+  supervisorState = {
+    ...supervisorState,
+    active: true,
+    agents: agents || [],
+    strategy,
+    service,
+    accountId,
+    ticks: [],
+    signals: [],
+    trades: [],
+    winningPatterns: [],
+    losingPatterns: [],
+    performance: {
+      trades: 0,
+      wins: 0,
+      losses: 0,
+      totalPnL: 0,
+      maxDrawdown: 0,
+      currentDrawdown: 0,
+      peakPnL: 0,
+      winStreak: 0,
+      lossStreak: 0,
+      maxWinStreak: 0,
+      maxLossStreak: 0
+    },
+    optimizations: [],
+    lastOptimizationTime: Date.now()
+  };
+  
+  // Start continuous analysis loop
+  if (analysisInterval) clearInterval(analysisInterval);
+  analysisInterval = setInterval(analyzeAndOptimize, supervisorState.optimizationInterval);
+  
+  return {
+    success: true,
+    agents: agents.length,
+    mode: agents.length >= 2 ? 'CONSENSUS' : 'INDIVIDUAL'
+  };
+};
+
+/**
+ * Stop supervisor
+ */
+const stop = () => {
+  if (analysisInterval) {
+    clearInterval(analysisInterval);
+    analysisInterval = null;
+  }
+  
+  const summary = {
+    ...supervisorState.performance,
+    optimizationsApplied: supervisorState.optimizations.length,
+    winningPatterns: supervisorState.winningPatterns.length,
+    losingPatterns: supervisorState.losingPatterns.length
+  };
+  
+  supervisorState.active = false;
+  
+  return summary;
+};
+
+/**
+ * Feed tick data (called on every market tick)
+ */
+const feedTick = (tick) => {
+  if (!supervisorState.active) return;
+  
+  supervisorState.ticks.push({
+    ...tick,
+    timestamp: Date.now()
+  });
+  
+  // Keep last 5000 ticks for pattern analysis
+  if (supervisorState.ticks.length > 5000) {
+    supervisorState.ticks = supervisorState.ticks.slice(-5000);
+  }
+};
+
+/**
+ * Feed signal data (called when strategy generates a signal)
+ */
+const feedSignal = (signal) => {
+  if (!supervisorState.active) return;
+  
+  const signalData = {
+    ...signal,
+    timestamp: Date.now(),
+    ticksContext: supervisorState.ticks.slice(-50) // Last 50 ticks before signal
+  };
+  
+  supervisorState.signals.push(signalData);
+  
+  // Keep last 100 signals
+  if (supervisorState.signals.length > 100) {
+    supervisorState.signals = supervisorState.signals.slice(-100);
+  }
+};
+
+/**
+ * Feed trade result (called when a trade completes)
+ * This is where LEARNING happens
+ */
+const feedTradeResult = (trade) => {
+  if (!supervisorState.active) return;
+  
+  const tradeData = {
+    ...trade,
+    timestamp: Date.now(),
+    // Capture context at time of trade
+    ticksBefore: supervisorState.ticks.slice(-100),
+    signalUsed: supervisorState.signals[supervisorState.signals.length - 1] || null
+  };
+  
+  supervisorState.trades.push(tradeData);
+  
+  // Update performance metrics
+  const perf = supervisorState.performance;
+  perf.trades++;
+  perf.totalPnL += trade.pnl || 0;
+  
+  if (trade.pnl > 0) {
+    perf.wins++;
+    perf.winStreak++;
+    perf.lossStreak = 0;
+    perf.maxWinStreak = Math.max(perf.maxWinStreak, perf.winStreak);
+    
+    // Learn from winning trade
+    learnFromTrade(tradeData, 'win');
+  } else if (trade.pnl < 0) {
+    perf.losses++;
+    perf.lossStreak++;
+    perf.winStreak = 0;
+    perf.maxLossStreak = Math.max(perf.maxLossStreak, perf.lossStreak);
+    
+    // Learn from losing trade
+    learnFromTrade(tradeData, 'loss');
+  }
+  
+  // Update drawdown
+  if (perf.totalPnL > perf.peakPnL) {
+    perf.peakPnL = perf.totalPnL;
+    perf.currentDrawdown = 0;
+  } else {
+    perf.currentDrawdown = perf.peakPnL - perf.totalPnL;
+    perf.maxDrawdown = Math.max(perf.maxDrawdown, perf.currentDrawdown);
+  }
+  
+  // Trigger immediate analysis after losing streaks
+  if (perf.lossStreak >= 3) {
+    analyzeAndOptimize();
+  }
+};
+
+/**
+ * Learn from a completed trade
+ * Extracts patterns from winning and losing trades
+ */
+const learnFromTrade = (trade, result) => {
+  const pattern = {
+    timestamp: trade.timestamp,
+    result,
+    pnl: trade.pnl,
+    direction: trade.direction || trade.side,
+    
+    // Market context before trade
+    priceAction: analyzePriceAction(trade.ticksBefore),
+    volumeProfile: analyzeVolume(trade.ticksBefore),
+    volatility: calculateVolatility(trade.ticksBefore),
+    
+    // Signal characteristics
+    signalConfidence: trade.signalUsed?.confidence || null,
+    entryPrice: trade.price || trade.signalUsed?.entry,
+    stopLoss: trade.signalUsed?.stopLoss,
+    takeProfit: trade.signalUsed?.takeProfit
+  };
+  
+  if (result === 'win') {
+    supervisorState.winningPatterns.push(pattern);
+    // Keep last 50 winning patterns
+    if (supervisorState.winningPatterns.length > 50) {
+      supervisorState.winningPatterns = supervisorState.winningPatterns.slice(-50);
+    }
+  } else {
+    supervisorState.losingPatterns.push(pattern);
+    // Keep last 50 losing patterns
+    if (supervisorState.losingPatterns.length > 50) {
+      supervisorState.losingPatterns = supervisorState.losingPatterns.slice(-50);
+    }
+  }
+};
+
+/**
+ * Analyze price action from ticks
+ */
+const analyzePriceAction = (ticks) => {
+  if (!ticks || ticks.length < 2) return { trend: 'unknown', strength: 0 };
+  
+  const prices = ticks.map(t => t.price).filter(Boolean);
+  if (prices.length < 2) return { trend: 'unknown', strength: 0 };
+  
+  const first = prices[0];
+  const last = prices[prices.length - 1];
+  const change = last - first;
+  const range = Math.max(...prices) - Math.min(...prices);
+  
+  return {
+    trend: change > 0 ? 'up' : change < 0 ? 'down' : 'flat',
+    strength: range > 0 ? Math.abs(change) / range : 0,
+    range,
+    change
+  };
+};
+
+/**
+ * Analyze volume from ticks
+ */
+const analyzeVolume = (ticks) => {
+  if (!ticks || ticks.length === 0) return { total: 0, avg: 0, trend: 'unknown' };
+  
+  const volumes = ticks.map(t => t.volume || 0);
+  const total = volumes.reduce((a, b) => a + b, 0);
+  const avg = total / volumes.length;
+  
+  // Compare first half vs second half
+  const mid = Math.floor(volumes.length / 2);
+  const firstHalf = volumes.slice(0, mid).reduce((a, b) => a + b, 0);
+  const secondHalf = volumes.slice(mid).reduce((a, b) => a + b, 0);
+  
+  return {
+    total,
+    avg,
+    trend: secondHalf > firstHalf * 1.2 ? 'increasing' : secondHalf < firstHalf * 0.8 ? 'decreasing' : 'stable'
+  };
+};
+
+/**
+ * Calculate volatility from ticks
+ */
+const calculateVolatility = (ticks) => {
+  if (!ticks || ticks.length < 2) return 0;
+  
+  const prices = ticks.map(t => t.price).filter(Boolean);
+  if (prices.length < 2) return 0;
+  
+  const returns = [];
+  for (let i = 1; i < prices.length; i++) {
+    returns.push((prices[i] - prices[i-1]) / prices[i-1]);
+  }
+  
+  const mean = returns.reduce((a, b) => a + b, 0) / returns.length;
+  const variance = returns.reduce((sum, r) => sum + Math.pow(r - mean, 2), 0) / returns.length;
+  
+  return Math.sqrt(variance);
+};
+
+/**
+ * Main analysis and optimization loop
+ * Called periodically and after significant events
+ */
+const analyzeAndOptimize = async () => {
+  if (!supervisorState.active || supervisorState.agents.length === 0) return;
+  
+  const perf = supervisorState.performance;
+  
+  // Skip if not enough data
+  if (perf.trades < 3) return;
+  
+  // Prepare performance data for AI analysis
+  const performanceData = {
+    trades: perf.trades,
+    wins: perf.wins,
+    losses: perf.losses,
+    winRate: perf.trades > 0 ? perf.wins / perf.trades : 0,
+    pnl: perf.totalPnL,
+    maxDrawdown: perf.maxDrawdown,
+    currentDrawdown: perf.currentDrawdown,
+    winStreak: perf.winStreak,
+    lossStreak: perf.lossStreak,
+    maxWinStreak: perf.maxWinStreak,
+    maxLossStreak: perf.maxLossStreak,
+    
+    // Calculate averages
+    avgWin: perf.wins > 0 ? 
+      supervisorState.trades.filter(t => t.pnl > 0).reduce((s, t) => s + t.pnl, 0) / perf.wins : 0,
+    avgLoss: perf.losses > 0 ? 
+      Math.abs(supervisorState.trades.filter(t => t.pnl < 0).reduce((s, t) => s + t.pnl, 0) / perf.losses) : 0,
+    
+    // Recent trades for context
+    recentTrades: supervisorState.trades.slice(-10).map(t => ({
+      side: t.direction || t.side,
+      qty: t.qty,
+      price: t.price,
+      pnl: t.pnl
+    })),
+    
+    // Pattern summaries
+    winningPatternCount: supervisorState.winningPatterns.length,
+    losingPatternCount: supervisorState.losingPatterns.length,
+    
+    // Common characteristics of losing trades
+    losingTradeAnalysis: analyzeLosingPatterns()
+  };
+  
+  // Get optimization suggestions from all agents
+  const suggestions = [];
+  
+  for (const agent of supervisorState.agents) {
+    try {
+      const suggestion = await getOptimizationFromAgent(agent, performanceData);
+      if (suggestion) {
+        suggestions.push({
+          agentId: agent.id,
+          agentName: agent.name,
+          ...suggestion
+        });
+      }
+    } catch (e) {
+      // Silent fail for individual agent
+    }
+  }
+  
+  if (suggestions.length === 0) return;
+  
+  // Process suggestions based on mode
+  const isConsensus = supervisorState.agents.length >= 2;
+  
+  if (isConsensus) {
+    // CONSENSUS MODE: All agents must agree
+    const consensusResult = buildConsensus(suggestions);
+    
+    if (consensusResult.isUnanimous && consensusResult.optimizations.length > 0) {
+      // Apply unanimous optimizations
+      for (const opt of consensusResult.optimizations) {
+        applyOptimization(opt);
+      }
+    }
+    
+    // Update current advice based on consensus
+    if (consensusResult.action) {
+      supervisorState.currentAdvice = {
+        action: consensusResult.action,
+        sizeMultiplier: consensusResult.sizeMultiplier || 1.0,
+        reason: consensusResult.reason || 'Consensus recommendation'
+      };
+    }
+  } else {
+    // INDIVIDUAL MODE: Apply single agent's suggestions
+    const suggestion = suggestions[0];
+    
+    if (suggestion.optimizations) {
+      for (const opt of suggestion.optimizations) {
+        applyOptimization(opt);
+      }
+    }
+    
+    if (suggestion.action) {
+      supervisorState.currentAdvice = {
+        action: suggestion.action,
+        sizeMultiplier: suggestion.sizeMultiplier || 1.0,
+        reason: suggestion.reason || 'Agent recommendation'
+      };
+    }
+  }
+  
+  supervisorState.lastOptimizationTime = Date.now();
+};
+
+/**
+ * Analyze patterns in losing trades
+ */
+const analyzeLosingPatterns = () => {
+  const patterns = supervisorState.losingPatterns;
+  if (patterns.length === 0) return null;
+  
+  // Find common characteristics
+  const trends = patterns.map(p => p.priceAction?.trend).filter(Boolean);
+  const volatilities = patterns.map(p => p.volatility).filter(Boolean);
+  const confidences = patterns.map(p => p.signalConfidence).filter(Boolean);
+  
+  const trendCounts = {};
+  for (const t of trends) {
+    trendCounts[t] = (trendCounts[t] || 0) + 1;
+  }
+  
+  const avgVolatility = volatilities.length > 0 ? 
+    volatilities.reduce((a, b) => a + b, 0) / volatilities.length : 0;
+  
+  const avgConfidence = confidences.length > 0 ?
+    confidences.reduce((a, b) => a + b, 0) / confidences.length : 0;
+  
+  return {
+    commonTrend: Object.entries(trendCounts).sort((a, b) => b[1] - a[1])[0]?.[0] || 'unknown',
+    avgVolatility,
+    avgConfidence,
+    count: patterns.length
+  };
+};
+
+/**
+ * Get optimization suggestion from a single agent
+ */
+const getOptimizationFromAgent = async (agent, performanceData) => {
+  const systemPrompt = `You are an AI supervisor for HQX Ultra Scalping, a professional futures trading strategy.
+
+The strategy uses:
+- Order flow analysis (delta, absorption, imbalance)
+- Statistical models (z-score, standard deviation)
+- Dynamic risk management (Kelly criterion)
+
+ANALYZE the performance data and LEARN from the losing trades.
+Suggest SPECIFIC optimizations to improve win rate and reduce losses.
+
+Respond in JSON:
+{
+  "assessment": "brief assessment",
+  "action": "AGGRESSIVE|NORMAL|CAUTIOUS|PAUSE",
+  "sizeMultiplier": 0.5-1.5,
+  "optimizations": [
+    {"param": "name", "direction": "increase|decrease", "amount": "10%", "reason": "why"}
+  ],
+  "learnings": "what we learned from losing trades",
+  "confidence": 0-100
+}`;
+
+  const prompt = `STRATEGY PERFORMANCE ANALYSIS
+
+Stats:
+- Trades: ${performanceData.trades} (${performanceData.wins}W / ${performanceData.losses}L)
+- Win Rate: ${(performanceData.winRate * 100).toFixed(1)}%
+- P&L: $${performanceData.pnl.toFixed(2)}
+- Max Drawdown: $${performanceData.maxDrawdown.toFixed(2)}
+- Current Streak: ${performanceData.winStreak > 0 ? performanceData.winStreak + ' wins' : performanceData.lossStreak + ' losses'}
+
+Losing Trade Analysis:
+${performanceData.losingTradeAnalysis ? `
+- Common trend at entry: ${performanceData.losingTradeAnalysis.commonTrend}
+- Avg volatility: ${(performanceData.losingTradeAnalysis.avgVolatility * 100).toFixed(3)}%
+- Avg signal confidence: ${(performanceData.losingTradeAnalysis.avgConfidence * 100).toFixed(1)}%
+- Total losing patterns: ${performanceData.losingTradeAnalysis.count}
+` : 'Not enough data'}
+
+Recent Trades:
+${performanceData.recentTrades.map(t => `${t.side} @ ${t.price} → $${t.pnl?.toFixed(2)}`).join('\n')}
+
+What should we LEARN and OPTIMIZE?`;
+
+  try {
+    const response = await callAI(agent, prompt, systemPrompt);
+    if (!response) return null;
+    
+    const jsonMatch = response.match(/\{[\s\S]*\}/);
+    if (jsonMatch) {
+      return JSON.parse(jsonMatch[0]);
+    }
+    return null;
+  } catch (e) {
+    return null;
+  }
+};
+
+/**
+ * Build consensus from multiple agent suggestions
+ */
+const buildConsensus = (suggestions) => {
+  if (suggestions.length === 0) return { isUnanimous: false };
+  
+  // Check action consensus
+  const actions = suggestions.map(s => s.action).filter(Boolean);
+  const allSameAction = actions.length === suggestions.length && 
+    actions.every(a => a === actions[0]);
+  
+  // Check optimization consensus
+  const allOptimizations = suggestions.flatMap(s => s.optimizations || []);
+  const paramGroups = {};
+  
+  for (const opt of allOptimizations) {
+    if (!opt.param) continue;
+    const key = `${opt.param}_${opt.direction}`;
+    if (!paramGroups[key]) {
+      paramGroups[key] = { ...opt, count: 0 };
+    }
+    paramGroups[key].count++;
+  }
+  
+  // Find unanimous optimizations (all agents agree)
+  const unanimousOptimizations = Object.values(paramGroups)
+    .filter(g => g.count === suggestions.length)
+    .map(g => ({
+      param: g.param,
+      direction: g.direction,
+      amount: g.amount,
+      reason: `Unanimous (${suggestions.length} agents)`
+    }));
+  
+  // Average size multiplier
+  const multipliers = suggestions.map(s => s.sizeMultiplier || 1.0);
+  const avgMultiplier = multipliers.reduce((a, b) => a + b, 0) / multipliers.length;
+  
+  // Average confidence
+  const confidences = suggestions.map(s => s.confidence || 50);
+  const avgConfidence = Math.round(confidences.reduce((a, b) => a + b, 0) / confidences.length);
+  
+  return {
+    isUnanimous: allSameAction && unanimousOptimizations.length > 0,
+    action: allSameAction ? actions[0] : 'CAUTIOUS',
+    sizeMultiplier: allSameAction ? avgMultiplier : 0.5,
+    optimizations: unanimousOptimizations,
+    confidence: avgConfidence,
+    reason: allSameAction ? 
+      `${suggestions.length} agents agree` : 
+      'Agents disagree - being cautious',
+    votes: actions.reduce((acc, a) => { acc[a] = (acc[a] || 0) + 1; return acc; }, {})
+  };
+};
+
+/**
+ * Apply an optimization to the strategy
+ */
+const applyOptimization = (optimization) => {
+  const strategy = supervisorState.strategy;
+  if (!strategy) return false;
+  
+  // Record the optimization
+  supervisorState.optimizations.push({
+    timestamp: Date.now(),
+    ...optimization
+  });
+  
+  // Try to apply to strategy if it supports it
+  try {
+    if (typeof strategy.applyAIOptimization === 'function') {
+      strategy.applyAIOptimization(optimization);
+      return true;
+    }
+    
+    if (typeof strategy.setParameter === 'function') {
+      strategy.setParameter(optimization.param, optimization.direction, optimization.amount);
+      return true;
+    }
+  } catch (e) {
+    // Strategy doesn't support this optimization
+  }
+  
+  return false;
+};
+
+/**
+ * Get current advice for the strategy
+ * Called before each trade decision
+ */
+const getCurrentAdvice = () => {
+  if (!supervisorState.active) {
+    return { action: 'NORMAL', sizeMultiplier: 1.0, reason: 'No supervision' };
+  }
+  
+  return supervisorState.currentAdvice;
+};
+
+/**
+ * Get supervision status
+ */
+const getStatus = () => {
+  return {
+    active: supervisorState.active,
+    agents: supervisorState.agents.length,
+    mode: supervisorState.agents.length >= 2 ? 'CONSENSUS' : 'INDIVIDUAL',
+    performance: supervisorState.performance,
+    currentAdvice: supervisorState.currentAdvice,
+    optimizationsApplied: supervisorState.optimizations.length,
+    patternsLearned: {
+      winning: supervisorState.winningPatterns.length,
+      losing: supervisorState.losingPatterns.length
+    },
+    lastOptimization: supervisorState.lastOptimizationTime
+  };
+};
+
+/**
+ * Check if should proceed with trade based on AI advice
+ */
+const shouldTrade = () => {
+  if (!supervisorState.active) return { proceed: true, multiplier: 1.0 };
+  
+  const advice = supervisorState.currentAdvice;
+  
+  if (advice.action === 'PAUSE') {
+    return { proceed: false, reason: advice.reason };
+  }
+  
+  return {
+    proceed: true,
+    multiplier: advice.sizeMultiplier || 1.0,
+    action: advice.action
+  };
+};
+
+module.exports = {
+  initialize,
+  stop,
+  feedTick,
+  feedSignal,
+  feedTradeResult,
+  getCurrentAdvice,
+  shouldTrade,
+  getStatus,
+  analyzeAndOptimize
+};