# ================================================================================================
# Paper Trading Configuration for Ensemble ML Models
# ================================================================================================
# Last Updated: 2025-10-14
# Status: PRODUCTION READY
# Purpose: Configure paper trading deployment for 3-model ensemble (DQN + 2x PPO)
# ================================================================================================

# ================================================================================================
# PAPER TRADING SETTINGS
# ================================================================================================
paper_trading:
  # Trading mode (paper = simulated execution, live = real capital)
  mode: paper  # DO NOT change to 'live' without completing 5-phase rollout

  # Virtual capital for paper trading
  initial_capital: 100000  # $100K virtual capital

  # Risk limits for paper trading phase
  max_position_size: 10000  # $10K per position (conservative for testing)
  max_daily_loss: 2000  # $2K max loss per day (circuit breaker trigger)
  max_open_positions: 3  # Maximum 3 positions at once

  # Symbols to trade (start with 2 liquid futures)
  symbols:
    - ES.FUT  # E-mini S&P 500 (highly liquid)
    - NQ.FUT  # Nasdaq-100 futures (highly liquid)

  # Transaction cost simulation (realistic slippage + commission)
  simulated_commission_per_contract: 4.50  # $4.50 per contract (typical futures commission)
  simulated_slippage_bps: 2  # 2 basis points slippage (0.02%)

# ================================================================================================
# ENSEMBLE MODEL CONFIGURATION
# ================================================================================================
ensemble:
  # Model definitions (3-model ensemble based on checkpoint analysis)
  models:
    # DQN: Epoch 30 (Sharpe 1.63, explained_variance 0.84, stable)
    - name: DQN_epoch30
      type: DQN
      checkpoint: ml/trained_models/production/dqn/dqn_epoch_30.safetensors
      weight: 0.4  # 40% weight (highest performing single model)
      enabled: true

    # PPO: Epoch 130 (Sharpe 1.59, explained_variance 0.69, value function healthy)
    - name: PPO_epoch130
      type: PPO
      checkpoint_actor: ml/trained_models/production/ppo/ppo_actor_epoch_130.safetensors
      checkpoint_critic: ml/trained_models/production/ppo/ppo_critic_epoch_130.safetensors
      weight: 0.4  # 40% weight (diversity champion)
      enabled: true

    # PPO: Epoch 420 (Sharpe 1.48, explained_variance 0.67, mature training)
    - name: PPO_epoch420
      type: PPO
      checkpoint_actor: ml/trained_models/production/ppo/ppo_actor_epoch_420.safetensors
      checkpoint_critic: ml/trained_models/production/ppo/ppo_critic_epoch_420.safetensors
      weight: 0.2  # 20% weight (stability anchor)
      enabled: true

  # Voting strategy
  voting_strategy: weighted  # Options: weighted, majority, unanimous
  consensus_threshold: 0.6  # 60% agreement required for trade execution

  # Performance thresholds
  prediction_timeout_ms: 50  # 50ms max latency per model
  min_confidence_threshold: 0.55  # Only trade when confidence > 55%
  max_disagreement_threshold: 0.70  # Reduce position size if disagreement > 70%

  # Dynamic weight adjustment (adaptive performance weighting)
  adaptive_weights:
    enabled: true
    adjustment_window_hours: 24  # Recalculate weights every 24 hours
    min_weight: 0.1  # Minimum weight per model (10%)
    max_weight: 0.6  # Maximum weight per model (60%)
    adjustment_alpha: 0.1  # Exponential moving average smoothing (10% learning rate)

# ================================================================================================
# RISK MANAGEMENT
# ================================================================================================
risk_management:
  # Per-model circuit breakers (halt if single model fails)
  per_model_circuit_breaker:
    enabled: true
    max_consecutive_errors: 3  # Disable model after 3 consecutive prediction errors
    error_rate_threshold: 0.10  # Disable model if error rate > 10% over 1 hour

  # Cascade failure protection (halt entire ensemble if multiple models fail)
  cascade_failure_threshold: 2  # If 2 models fail, stop trading

  # Cooldown periods after circuit breaker triggers
  cooldown_period_minutes: 60  # 1 hour cooldown before re-enabling failed models

  # Position sizing based on confidence
  confidence_based_sizing:
    enabled: true
    min_confidence: 0.55  # No trade below 55% confidence
    max_confidence: 0.90  # Cap confidence at 90% (avoid overconfidence)
    min_position_multiplier: 0.25  # 25% position at min confidence
    max_position_multiplier: 1.00  # 100% position at max confidence

  # Disagreement-based risk reduction
  disagreement_penalty:
    enabled: true
    low_disagreement: 0.30  # <30% disagreement = no penalty
    medium_disagreement: 0.50  # 30-50% = 25% position reduction
    high_disagreement: 0.70  # >70% = 50% position reduction

# ================================================================================================
# MONITORING & ALERTING
# ================================================================================================
monitoring:
  # Logging configuration
  log_all_predictions: true  # Log every ensemble prediction (for audit and debugging)
  log_level: info  # Options: debug, info, warn, error

  # Prometheus metrics export
  prometheus_metrics:
    enabled: true
    port: 9092  # Metrics exposed on http://localhost:9092/metrics
    push_interval_seconds: 5  # Push metrics every 5 seconds

  # Grafana dashboard integration
  grafana_dashboard:
    enabled: true
    dashboard_id: ensemble-ml-prod  # Dashboard UID in Grafana
    url: http://localhost:3000/d/ensemble-ml-prod

  # Alert thresholds (trigger PagerDuty/Slack alerts)
  alerts:
    # Sharpe ratio drop alert (model degradation)
    sharpe_drop_threshold: 0.5  # Alert if Sharpe drops by 50% (e.g., 1.6 -> 0.8)
    sharpe_check_window_hours: 24

    # High disagreement alert (market regime shift)
    high_disagreement_rate: 0.70  # Alert if disagreement > 70% for 10+ predictions
    high_disagreement_count: 10

    # Latency spike alert (performance degradation)
    latency_p99_threshold_us: 100  # Alert if P99 latency > 100μs

    # Circuit breaker alert (model failure)
    circuit_breaker_triggered: true  # Always alert on circuit breaker

    # Daily P&L alert (risk management)
    daily_loss_alert_threshold: -1500  # Alert if daily loss exceeds $1,500

# ================================================================================================
# AUDIT & COMPLIANCE
# ================================================================================================
audit:
  # PostgreSQL audit logging
  database_logging:
    enabled: true
    table: ensemble_predictions  # Primary audit table
    batch_size: 100  # Batch insert 100 predictions at once
    flush_interval_seconds: 10  # Flush batch every 10 seconds

  # Feature snapshot storage (for reproducibility)
  feature_snapshots:
    enabled: true  # Store all input features in JSONB column
    compression: true  # Compress feature snapshots (save storage)

  # Compliance tracking
  compliance:
    enabled: true
    track_user_id: true  # Track which user initiated trade
    track_session_id: true  # Track session for audit trail
    track_request_id: true  # Track request ID for distributed tracing

# ================================================================================================
# A/B TESTING (OPTIONAL - NOT USED IN PAPER TRADING PHASE)
# ================================================================================================
ab_testing:
  # A/B testing is DISABLED during paper trading phase
  # Enable in Phase 2 (1% capital) for ensemble vs single-model comparison
  enabled: false

  # A/B test configuration (for future phases)
  test_id: null  # Set when starting A/B test
  control_model: DQN_epoch30  # Baseline: single model
  treatment_model: ensemble  # Treatment: 3-model ensemble
  traffic_split: 0.5  # 50/50 split
  min_sample_size: 1000  # 1,000 predictions per group
  significance_level: 0.05  # 95% confidence (p < 0.05)
  max_duration_hours: 168  # 1 week maximum

# ================================================================================================
# DEPLOYMENT METADATA
# ================================================================================================
metadata:
  deployment_phase: paper_trading  # Phase 1 of 5-phase rollout
  deployment_date: 2025-10-14
  version: 1.0.0
  deployed_by: ml-team@foxhunt.trading
  notes: |
    Paper trading deployment for 3-model ensemble:
    - DQN epoch 30 (Sharpe 1.63, weight 0.4)
    - PPO epoch 130 (Sharpe 1.59, weight 0.4)
    - PPO epoch 420 (Sharpe 1.48, weight 0.2)

    Success criteria for Phase 2 advancement:
    - Sharpe ratio > 1.5 over 7 days
    - Win rate > 52%
    - Max drawdown < 10%
    - Zero critical errors or rollbacks
    - Latency P99 < 50μs