foxhunt/WAVE_13.2_AGENT_11_QUICK_REFERENCE.md

Agent 11 Quick Reference: ML Order Service

Status: ✅ ALREADY IMPLEMENTED - No changes required

---

What Was Found

The ML order submission service is fully operational in the Trading Service. All three gRPC methods are implemented:

1. ✅ SubmitMLOrder - Lines 649-747 in trading.rs
2. ✅ GetMLPredictions - Lines 749-902 in trading.rs
3. ✅ GetMLPerformance - Lines 904-1289 in trading.rs

---

Key Implementation Details

SubmitMLOrder Logic

1. Validate 26 features (5 OHLCV + 21 technical indicators)
2. Generate ensemble prediction via EnsembleCoordinator
3. Check confidence threshold (60% minimum)
4. Execute order if BUY/SELL action + high confidence
5. Store prediction in ensemble_predictions table
6. Return MLOrderResponse with order_id and prediction_id

Database Schema

Table: ensemble_predictions (migration 022)

• Stores all ML predictions with per-model attribution
• Links predictions to orders via order_id
• Tracks P&L, confidence, disagreement rate
• TimescaleDB hypertable for fast queries

Ensemble Models

• DQN - Deep Q-Network
• PPO - Proximal Policy Optimization
• MAMBA-2 - State space model (200-epoch trained, 70.6% loss reduction)
• TFT - Temporal Fusion Transformer

---

Proto Definitions

Request:

message MLOrderRequest {
  string symbol = 1;                    // ES.FUT, NQ.FUT, etc.
  string account_id = 2;                // Trading account
  bool use_ensemble = 3;                // Always true
  repeated double features = 5;         // 26 features (OHLCV + technicals)
}

Response:

message MLOrderResponse {
  string order_id = 1;                  // Order ID (if executed)
  string prediction_id = 2;             // UUID in ensemble_predictions
  string action = 3;                    // BUY, SELL, HOLD
  double confidence = 4;                // 0.0-1.0
  string message = 5;                   // Status message
  bool executed = 6;                    // True if order placed
}

---

Key Files

File	Lines	Purpose
services/trading_service/src/services/trading.rs	649-1289	ML order implementation
services/trading_service/proto/trading.proto	45-258	Proto definitions
migrations/022_create_ensemble_tables.sql	-	Database schema
services/trading_service/src/ensemble_coordinator.rs	-	Ensemble orchestration

---

Performance Targets

Operation	Target	Typical
SubmitMLOrder	<100ms	60-560ms
GetMLPredictions	<100ms	5-50ms
GetMLPerformance	<400ms	40-400ms

---

Next Steps

Agent 7: API Gateway proxy for TLI integration Agent 14: Integration tests and database verification

---

TLI Usage (Future)

# Submit ML order
tli ml order --symbol ES.FUT --account paper_001 --ensemble

# Get prediction history
tli ml predictions --symbol ES.FUT --limit 100

# Get model performance
tli ml performance --model MAMBA2

---

Confidence Threshold

• Minimum: 60% confidence required
• BUY: ensemble_signal > 0.6 AND confidence >= 0.60
• SELL: ensemble_signal < 0.4 AND confidence >= 0.60
• HOLD: confidence < 0.60 OR 0.4 <= signal <= 0.6

---

Audit Trail

Every prediction stored with:

• Per-model signals and confidences
• Order linkage (if executed)
• P&L tracking (populated after order fills)
• Feature snapshot (for reproducibility)
• Model checkpoint IDs (version tracking)

---

Agent: 11/20 in Wave 13.2 Completion: 2025-10-16 Status: ✅ COMPLETE