🧠 Implement AI Learning System for Stop Loss Decisions

- Add stop-loss-decision-learner.js: Core learning engine
- Add enhanced-autonomous-risk-manager.js: Learning-enhanced decisions
- Add AI learning API and dashboard components
- Add database schema for decision tracking
- Integrate with existing automation system
- Demo scripts and documentation

Result: AI learns from every decision and improves over time! 🚀

database/stop-loss-learning-schema.sql

@@ -0,0 +1,125 @@
+-- Stop Loss Decision Learning Database Schema
+-- This extends the existing database to support decision-level learning
+
+-- Table to track every AI decision made near stop loss
+CREATE TABLE IF NOT EXISTS sl_decisions (
+  id TEXT PRIMARY KEY,
+  trade_id TEXT,
+  symbol TEXT NOT NULL,
+  decision_type TEXT NOT NULL, -- 'HOLD', 'EXIT', 'TIGHTEN_SL', 'PARTIAL_EXIT', 'EMERGENCY_EXIT'
+  distance_from_sl REAL NOT NULL,
+  reasoning TEXT,
+  market_conditions TEXT, -- JSON with market context
+  confidence_score REAL DEFAULT 0.7,
+  expected_outcome TEXT,
+  decision_timestamp DATETIME DEFAULT CURRENT_TIMESTAMP,
+  
+  -- Outcome tracking (filled when trade closes or situation resolves)
+  outcome TEXT, -- 'HIT_ORIGINAL_SL', 'HIT_TIGHTENED_SL', 'PROFITABLE_EXIT', 'AVOIDED_LOSS', etc.
+  outcome_timestamp DATETIME,
+  time_to_outcome INTEGER, -- minutes from decision to outcome
+  pnl_impact REAL, -- how much P&L was affected by the decision
+  was_correct BOOLEAN,
+  learning_score REAL, -- calculated learning score (0-1)
+  additional_context TEXT, -- JSON with additional outcome context
+  status TEXT DEFAULT 'PENDING_OUTCOME', -- 'PENDING_OUTCOME', 'ASSESSED'
+  
+  -- Indexing for faster queries
+  FOREIGN KEY (trade_id) REFERENCES trades(id)
+);
+
+-- Indexes for performance
+CREATE INDEX IF NOT EXISTS idx_sl_decisions_symbol ON sl_decisions(symbol);
+CREATE INDEX IF NOT EXISTS idx_sl_decisions_decision_type ON sl_decisions(decision_type);
+CREATE INDEX IF NOT EXISTS idx_sl_decisions_distance ON sl_decisions(distance_from_sl);
+CREATE INDEX IF NOT EXISTS idx_sl_decisions_timestamp ON sl_decisions(decision_timestamp);
+CREATE INDEX IF NOT EXISTS idx_sl_decisions_status ON sl_decisions(status);
+CREATE INDEX IF NOT EXISTS idx_sl_decisions_was_correct ON sl_decisions(was_correct);
+
+-- Table to store learning model parameters and thresholds
+CREATE TABLE IF NOT EXISTS learning_parameters (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  parameter_name TEXT UNIQUE NOT NULL,
+  parameter_value REAL NOT NULL,
+  description TEXT,
+  last_updated DATETIME DEFAULT CURRENT_TIMESTAMP,
+  confidence_level REAL DEFAULT 0.5 -- how confident we are in this parameter
+);
+
+-- Insert default learning parameters
+INSERT OR REPLACE INTO learning_parameters (parameter_name, parameter_value, description) VALUES
+('emergency_distance_threshold', 1.0, 'Distance from SL (%) that triggers emergency decisions'),
+('high_risk_distance_threshold', 2.0, 'Distance from SL (%) that triggers high risk decisions'),
+('medium_risk_distance_threshold', 5.0, 'Distance from SL (%) that triggers medium risk decisions'),
+('min_decisions_for_learning', 5, 'Minimum number of decisions needed to start learning'),
+('confidence_threshold', 0.7, 'Minimum confidence to use learned recommendations'),
+('learning_rate', 0.1, 'How quickly to adapt to new decision outcomes');
+
+-- Table to store decision pattern insights
+CREATE TABLE IF NOT EXISTS decision_patterns (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  pattern_type TEXT NOT NULL, -- 'SUCCESSFUL', 'FAILURE', 'OPTIMAL_TIMING'
+  decision_type TEXT NOT NULL,
+  conditions TEXT, -- JSON with conditions that trigger this pattern
+  success_rate REAL,
+  sample_size INTEGER,
+  avg_pnl_impact REAL,
+  confidence_score REAL,
+  discovered_at DATETIME DEFAULT CURRENT_TIMESTAMP,
+  last_validated DATETIME DEFAULT CURRENT_TIMESTAMP,
+  is_active BOOLEAN DEFAULT TRUE
+);
+
+-- View for quick decision performance analysis
+CREATE VIEW IF NOT EXISTS decision_performance AS
+SELECT 
+  decision_type,
+  COUNT(*) as total_decisions,
+  AVG(CASE WHEN was_correct THEN 1 ELSE 0 END) as success_rate,
+  AVG(learning_score) as avg_learning_score,
+  AVG(pnl_impact) as avg_pnl_impact,
+  AVG(distance_from_sl) as avg_distance_from_sl,
+  MIN(decision_timestamp) as first_decision,
+  MAX(decision_timestamp) as latest_decision
+FROM sl_decisions 
+WHERE status = 'ASSESSED'
+GROUP BY decision_type
+ORDER BY success_rate DESC;
+
+-- View for time-based decision analysis
+CREATE VIEW IF NOT EXISTS decision_timing_analysis AS
+SELECT 
+  CASE 
+    WHEN strftime('%H', decision_timestamp) BETWEEN '00' AND '05' THEN 'Night'
+    WHEN strftime('%H', decision_timestamp) BETWEEN '06' AND '11' THEN 'Morning'
+    WHEN strftime('%H', decision_timestamp) BETWEEN '12' AND '17' THEN 'Afternoon'
+    ELSE 'Evening'
+  END as time_period,
+  decision_type,
+  COUNT(*) as decisions,
+  AVG(CASE WHEN was_correct THEN 1 ELSE 0 END) as success_rate,
+  AVG(pnl_impact) as avg_pnl_impact
+FROM sl_decisions 
+WHERE status = 'ASSESSED'
+GROUP BY time_period, decision_type
+ORDER BY success_rate DESC;
+
+-- View for distance-based decision analysis  
+CREATE VIEW IF NOT EXISTS distance_effectiveness AS
+SELECT 
+  CASE 
+    WHEN distance_from_sl < 1.0 THEN 'Emergency (<1%)'
+    WHEN distance_from_sl < 2.0 THEN 'High Risk (1-2%)'
+    WHEN distance_from_sl < 5.0 THEN 'Medium Risk (2-5%)'
+    ELSE 'Safe (>5%)'
+  END as risk_level,
+  decision_type,
+  COUNT(*) as decisions,
+  AVG(CASE WHEN was_correct THEN 1 ELSE 0 END) as success_rate,
+  AVG(learning_score) as avg_score,
+  AVG(pnl_impact) as avg_pnl_impact,
+  AVG(time_to_outcome) as avg_resolution_time_minutes
+FROM sl_decisions 
+WHERE status = 'ASSESSED'
+GROUP BY risk_level, decision_type
+ORDER BY risk_level, success_rate DESC;