foxhunt/AGENT_165_QUICK_REFERENCE.md

AGENT 165: Ensemble Integration Tests - Quick Reference

File: /home/jgrusewski/Work/foxhunt/ml/tests/ensemble_integration_tests.rs Status: ✅ Complete (650 lines, code-only)

---

⚡ Quick Commands

# Run all tests
cargo test -p ml --test ensemble_integration_tests -- --nocapture

# Run specific test
cargo test -p ml --test ensemble_integration_tests test_01_all_models_loaded -- --nocapture

# Run with release mode (accurate latency benchmarks)
cargo test -p ml --test ensemble_integration_tests --release -- --nocapture

# Run with coverage
cargo llvm-cov test -p ml --test ensemble_integration_tests --html

---

📋 Test Checklist

#	Test Name	Purpose	Pass Criteria
1	test_01_all_models_loaded	6 models registered	model_count = 6, weights sum = 1.0
2	test_02_model_registry_state	Registry stability	model_count stable after update
3	test_03_ensemble_prediction_aggregation	Weighted voting	signal/confidence in range, avg_conf > 0.5
4	test_04_trading_action_determination	Buy/Sell/Hold logic	action distribution validated
5	test_05_model_disagreement_handling	High disagreement	disagreement_rate ≥ 40%
6	test_06_confidence_calculation	Ensemble confidence	min/max/avg in [0,1], avg > 0.5
7	test_07_fallback_on_model_error	Graceful degradation	5 models continue when 1 fails
8	test_08_adaptive_strategy_integration	Regime detection	different signals per regime
9	test_09_performance_latency	<100μs P99 target	P99 < 100μs in --release mode
10	test_10_full_e2e_pipeline	Full integration	500 predictions in <5 seconds

---

📊 Model Weights (Production)

Model	Weight	Confidence	Behavior
DQN	20%	0.78	Aggressive
PPO	20%	0.82	Most Aggressive
MAMBA-2	20%	0.85	Moderate
TFT	15%	0.75	Conservative
Liquid	15%	0.80	Adaptive
TLOB	10%	0.72	Very Conservative

Total: 100% (±1e-6 tolerance)

---

🎯 Performance Targets (--release mode)

Metric	Target	Expected
Average	<20μs	~12μs
P50	<15μs	~10μs
P95	<50μs	~18μs
P99	<100μs	~24μs
Throughput	>10K/s	~83K/s

---

🔧 Mock Predictors

create_dqn_mock()      // DQN:     multiplier 0.80, confidence 0.78
create_ppo_mock()      // PPO:     multiplier 0.90, confidence 0.82
create_mamba2_mock()   // MAMBA-2: multiplier 0.75, confidence 0.85
create_tft_mock()      // TFT:     multiplier 0.70, confidence 0.75
create_liquid_mock()   // Liquid:  multiplier 0.85, confidence 0.80
create_tlob_mock()     // TLOB:    multiplier 0.65, confidence 0.72
create_failing_mock()  // Error:   always fails (testing fallback)

Formula: signal = tanh(mean(features) × multiplier)

---

📐 Ensemble Formulas

Weighted Signal

weighted_signal = Σ(model_value × model_confidence × model_weight)
                  / Σ(model_weight × model_confidence)

Ensemble Confidence

ensemble_confidence = Σ(model_confidence × model_weight) / Σ(model_weight)

Disagreement Rate

disagreement_rate = count(sign(model_value) ≠ sign(mean_signal)) / total_models

Trading Action

if signal > 0.3  → Buy
if signal < -0.3 → Sell
else             → Hold

---

⚠️ Known Issues

1. Mock Implementation: Tests use mock predictors, not real models

   • Currently only 3 models (DQN, PPO, TFT) in coordinator
   • Update assertion in Test 3 after coordinator integration

2. Debug Mode Latency: P99 may exceed 100μs

   • Solution: Always run with --release flag

3. Real Checkpoints: Not loaded (future work after Wave 160)

---

🔗 Integration Points

EnsembleCoordinator Update Required

// ml/src/ensemble/coordinator.rs:122-132
fn mock_model_prediction(&self, model_id: &str, features: &Features) -> f64 {
    match model_id {
        "DQN" => (feature_mean * 0.8).tanh(),
        "PPO" => (feature_mean * 0.9).tanh(),
        "TFT" => (feature_mean * 0.7).tanh(),
        "MAMBA-2" => (feature_mean * 0.75).tanh(),  // ADD
        "Liquid" => (feature_mean * 0.85).tanh(),    // ADD
        "TLOB" => (feature_mean * 0.65).tanh(),      // ADD
        _ => 0.0,
    }
}

---

📚 Related Files

• Coordinator: /ml/src/ensemble/coordinator.rs
• Decision Types: /ml/src/ensemble/decision.rs
• E2E Tests: /ml/tests/e2e_ensemble_integration.rs
• Summary: AGENT_165_SUMMARY.md

---

🚀 Next Steps

1. ✅ Tests created (650 lines, code-only)
2. ⏳ Run after ML training (Wave 160)
3. ⏳ Update coordinator with 6-model mocks
4. ⏳ Replace mocks with real models
5. ⏳ Benchmark on RTX 3050 Ti GPU

---

Last Updated: 2025-10-15 (Agent 165) Status: ✅ READY FOR EXECUTION (after coordinator update)