foxhunt/services/trading_service/src/state.rs

//! Service state management and business logic coordination
//!
//! This module provides clean repository-based dependency injection,
//! eliminating direct database coupling from business logic.

use crate::error::{TradingServiceError, TradingServiceResult};
use crate::event_persistence::EventPersistence;
use crate::event_streaming::publisher::EventPublisher;
use crate::proto::monitoring::SystemMetrics;
use crate::repositories::*;
use crate::repository_impls::PostgresConfigRepository;
use trading_engine::trading::account_manager::AccountManager;
use trading_engine::trading::order_manager::OrderManager;
use trading_engine::trading::position_manager::PositionManager;

// Import provider traits for data connection methods
use data::providers::{MarketDataProvider, RealTimeProvider};

use futures::StreamExt;
use std::sync::Arc;
use tokio::sync::broadcast;
use tokio::sync::RwLock;

/// Central state manager for the trading service with repository pattern
///
/// This struct coordinates all business logic using repository dependency injection:
/// - Trading operations through TradingRepository
///
/// - Market data access through MarketDataRepository  
/// - Risk management through RiskRepository
///
/// - Configuration through ConfigRepository
/// - NO DIRECT DATABASE COUPLING
#[derive(Clone)]
pub struct TradingServiceState {
    /// Trading repository for orders, executions, positions
    pub trading_repository: Arc<dyn TradingRepository>,

    /// Market data repository for prices, order books
    pub market_data_repository: Arc<dyn MarketDataRepository>,

    /// Risk repository for limits, calculations, alerts
    pub risk_repository: Arc<dyn RiskRepository>,

    /// Configuration repository for settings and secrets
    pub config_repository: Arc<crate::repository_impls::PostgresConfigRepository>,

    /// Database connection pool for direct SQL queries
    pub db_pool: sqlx::PgPool,

    /// Risk management engine (business logic only)
    pub risk_engine: Arc<RwLock<RiskEngine>>,

    /// ML model registry and predictor (business logic only)
    pub ml_engine: Arc<RwLock<MLEngine>>,

    /// Market data manager (business logic only)
    pub market_data: Arc<RwLock<MarketDataManager>>,

    /// `Order` management system (business logic only)
    pub order_manager: Arc<RwLock<OrderManager>>,

    /// `Position` tracking (business logic only)
    pub position_manager: Arc<RwLock<PositionManager>>,

    /// Account management (business logic only)
    pub account_manager: Arc<RwLock<AccountManager>>,

    /// Event publisher for real-time streaming
    pub event_publisher: Arc<EventPublisher>,

    /// Event persistence for compliance and audit trail
    pub event_persistence: Arc<EventPersistence>,

    /// System metrics and monitoring
    pub metrics: Arc<RwLock<SystemMetrics>>,

    /// Kill switch system for emergency shutdown
    pub kill_switch_system: Option<Arc<crate::kill_switch_integration::TradingServiceKillSwitch>>,

    /// Ensemble coordinator for ML predictions (DQN, PPO, TFT)
    pub ensemble_coordinator: Option<Arc<crate::ensemble_coordinator::EnsembleCoordinator>>,
}

impl std::fmt::Debug for TradingServiceState {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("TradingServiceState")
            .field("trading_repository", &"<dyn TradingRepository>")
            .field("market_data_repository", &"<dyn MarketDataRepository>")
            .field("risk_repository", &"<dyn RiskRepository>")
            .field("config_repository", &self.config_repository)
            .field("db_pool", &"<PgPool>")
            .field("risk_engine", &self.risk_engine)
            .field("ml_engine", &self.ml_engine)
            .field("market_data", &self.market_data)
            .field("order_manager", &self.order_manager)
            .field("position_manager", &self.position_manager)
            .field("account_manager", &self.account_manager)
            .field("event_publisher", &self.event_publisher)
            .field("event_persistence", &self.event_persistence)
            .field("metrics", &self.metrics)
            .field("kill_switch_system", &self.kill_switch_system)
            .field("ensemble_coordinator", &self.ensemble_coordinator)
            .finish()
    }
}

impl TradingServiceState {
    /// Create new trading service state with repository dependency injection
    pub async fn new_with_repositories(
        trading_repository: Arc<dyn TradingRepository>,
        market_data_repository: Arc<dyn MarketDataRepository>,
        risk_repository: Arc<dyn RiskRepository>,
        config_repository: Arc<PostgresConfigRepository>,
        db_pool: sqlx::PgPool,
        event_persistence: Arc<EventPersistence>,
        kill_switch_system: Option<Arc<crate::kill_switch_integration::TradingServiceKillSwitch>>,
        ensemble_coordinator: Option<Arc<crate::ensemble_coordinator::EnsembleCoordinator>>,
    ) -> TradingServiceResult<Self> {
        // Initialize business logic components (no database coupling)
        let risk_engine = Arc::new(RwLock::new(RiskEngine::new()));
        let ml_engine = Arc::new(RwLock::new(MLEngine::new()));
        let market_data = Arc::new(RwLock::new(MarketDataManager::new()));
        let order_manager = Arc::new(RwLock::new(OrderManager::new()));
        let position_manager = Arc::new(RwLock::new(PositionManager::new()));
        let account_manager = Arc::new(RwLock::new(AccountManager::new()));
        // Create broadcast channel for event publishing
        let (event_sender, _) = broadcast::channel(10000);
        let event_publisher = Arc::new(EventPublisher::new(event_sender));
        let metrics = Arc::new(RwLock::new(SystemMetrics::default()));

        Ok(Self {
            trading_repository,
            market_data_repository,
            risk_repository,
            config_repository,
            db_pool,
            event_persistence,
            risk_engine,
            ml_engine,
            market_data,
            order_manager,
            position_manager,
            account_manager,
            event_publisher,
            metrics,
            kill_switch_system,
            ensemble_coordinator,
        })
    }

    /// Initialize service state with repository-based configuration
    pub async fn initialize(&self) -> TradingServiceResult<()> {
        // Initialize risk engine with repository-based configuration
        let mut risk_engine = self.risk_engine.write().await;
        risk_engine
            .initialize_with_config_repository(&self.config_repository)
            .await?;

        // Initialize ML engine with repository-based configuration
        let mut ml_engine = self.ml_engine.write().await;
        ml_engine
            .initialize_with_config_repository(&self.config_repository)
            .await?;

        // Initialize market data connections with repository-based configuration
        let mut market_data = self.market_data.write().await;
        market_data
            .initialize_with_config_repository(&self.config_repository)
            .await?;

        // Start event processing for market data providers
        market_data.start_event_processing().await?;

        Ok(())
    }
    /// Create new trading service state for testing purposes
    ///
    /// This creates a minimal state with mock repositories suitable for integration tests.
    ///
    /// Note: This function is only available when building tests.
    pub async fn new_for_testing() -> TradingServiceResult<Self> {
        // For testing, create a minimal repository setup using PostgreSQL implementations
        use crate::event_persistence::EventPersistence;
        use crate::repository_impls::{
            PostgresConfigRepository, PostgresMarketDataRepository, PostgresRiskRepository,
            PostgresTradingRepository,
        };
        use sqlx::PgPool;

        // Get database URL for testing
        let db_url = std::env::var("DATABASE_URL").unwrap_or_else(|_| {
            "postgresql://foxhunt:foxhunt_dev_password@localhost:5432/foxhunt".to_string()
        });

        let pool = PgPool::connect(&db_url).await.map_err(|e| {
            crate::error::TradingServiceError::Internal {
                message: format!("Failed to create test database pool: {}", e),
            }
        })?;

        // Create PostgreSQL repositories for testing
        let trading_repository =
            Arc::new(PostgresTradingRepository::new(pool.clone())) as Arc<dyn TradingRepository>;
        let market_data_repository = Arc::new(PostgresMarketDataRepository::new(pool.clone()))
            as Arc<dyn MarketDataRepository>;
        let risk_repository =
            Arc::new(PostgresRiskRepository::new(pool.clone())) as Arc<dyn RiskRepository>;

        // Create config repository using the same pool
        let config_repository = Arc::new(PostgresConfigRepository::new(pool.clone()));

        // Create event persistence with a test directory
        let event_persistence =
            Arc::new(EventPersistence::new_for_testing().await.map_err(|e| {
                crate::error::TradingServiceError::Internal {
                    message: format!("Failed to create event persistence: {}", e),
                }
            })?);

        // Create state without kill switch, model cache, or ensemble for simplicity
        Self::new_with_repositories(
            trading_repository,
            market_data_repository,
            risk_repository,
            config_repository,
            pool,
            event_persistence,
            None, // kill_switch_system
            None, // ensemble_coordinator
        )
        .await
    }

    /// Get health status of all components
    pub async fn get_health_status(&self) -> HealthStatus {
        // Check all components and return overall health
        let market_data_health = self.market_data.read().await.get_provider_health().await;

        // Check if any providers are unhealthy
        let has_unhealthy_providers = market_data_health
            .iter()
            .any(|(_, status)| !status.connected);

        // Check ensemble coordinator health
        let ensemble_health = self.check_ensemble_health().await;

        if has_unhealthy_providers {
            HealthStatus::Degraded
        } else if market_data_health.is_empty() {
            HealthStatus::Critical // No providers available
        } else if matches!(ensemble_health, EnsembleHealth::Unhealthy) {
            HealthStatus::Degraded // Ensemble issues but trading can continue
        } else {
            HealthStatus::Healthy
        }
    }

    /// Check ensemble coordinator health
    async fn check_ensemble_health(&self) -> EnsembleHealth {
        use tracing::{error, info};

        let ensemble = match &self.ensemble_coordinator {
            Some(coordinator) => coordinator,
            None => {
                info!("Ensemble coordinator not configured");
                return EnsembleHealth::NotConfigured;
            },
        };

        // Check if models are loaded
        let model_count = ensemble.model_count().await;
        if model_count == 0 {
            error!("Ensemble coordinator has no models loaded");
            return EnsembleHealth::Unhealthy;
        }

        // Mock health check - in production, this would:
        // 1. Verify all 6 models are loaded (DQN, PPO, TFT x 2 each)
        // 2. Check inference latency (<50μs P99)
        // 3. Check model staleness (last updated)
        // 4. Verify checkpoint integrity

        info!("Ensemble coordinator healthy with {} models", model_count);
        EnsembleHealth::Healthy
    }

    /// Get detailed ensemble health report
    pub async fn get_ensemble_health_report(&self) -> EnsembleHealthReport {
        let ensemble = match &self.ensemble_coordinator {
            Some(coordinator) => coordinator,
            None => {
                return EnsembleHealthReport {
                    status: EnsembleHealth::NotConfigured,
                    models_loaded: 0,
                    expected_models: 6, // DQN, PPO, TFT (2 checkpoints each)
                    inference_latency_us: None,
                    last_prediction: None,
                    model_details: vec![],
                };
            },
        };

        let model_count = ensemble.model_count().await;
        let status = if model_count == 0 {
            EnsembleHealth::Unhealthy
        } else if model_count < 6 {
            EnsembleHealth::Degraded
        } else {
            EnsembleHealth::Healthy
        };

        EnsembleHealthReport {
            status,
            models_loaded: model_count,
            expected_models: 6,
            inference_latency_us: Some(42.0), // Mock latency
            last_prediction: Some(std::time::SystemTime::now()),
            model_details: vec![
                ModelHealthDetail {
                    model_id: "DQN".to_string(),
                    checkpoint: "epoch_30".to_string(),
                    loaded: true,
                    last_inference: std::time::SystemTime::now(),
                },
                ModelHealthDetail {
                    model_id: "PPO".to_string(),
                    checkpoint: "epoch_30".to_string(),
                    loaded: true,
                    last_inference: std::time::SystemTime::now(),
                },
                ModelHealthDetail {
                    model_id: "TFT".to_string(),
                    checkpoint: "epoch_30".to_string(),
                    loaded: true,
                    last_inference: std::time::SystemTime::now(),
                },
            ],
        }
    }

    /// Subscribe to market data for trading symbols
    pub async fn subscribe_to_market_data(
        &self,
        symbols: Vec<common::types::Symbol>,
    ) -> TradingServiceResult<()> {
        let mut market_data = self.market_data.write().await;
        market_data.subscribe_to_symbols(symbols).await
    }

    /// Get market data event stream
    pub async fn get_market_data_stream(
        &self,
    ) -> tokio::sync::broadcast::Receiver<trading_engine::trading::data_interface::MarketDataEvent>
    {
        let market_data = self.market_data.read().await;
        market_data.get_event_receiver()
    }

    /// Get ensemble coordinator reference (if initialized)
    pub fn ensemble_coordinator(
        &self,
    ) -> Option<&Arc<crate::ensemble_coordinator::EnsembleCoordinator>> {
        self.ensemble_coordinator.as_ref()
    }

    /// Get trading signal from ensemble coordinator
    ///
    /// This method orchestrates the full prediction flow:
    /// 1. Extract features from market data
    /// 2. Call ensemble coordinator for prediction
    /// 3. Convert ensemble decision to trading signal
    /// 4. Apply risk adjustments (position sizing)
    /// 5. Return signal with ensemble attribution
    ///
    /// Returns `Err(MLModel)` if no real prediction is available (ensemble not
    /// initialised, feature extraction fails, or inference fails).
    pub async fn get_ensemble_trading_signal(
        &self,
        symbol: &str,
    ) -> TradingServiceResult<EnsembleTradingSignal> {
        use tracing::{info, warn};

        // Check if ensemble coordinator is available
        let ensemble = match &self.ensemble_coordinator {
            Some(coordinator) => coordinator,
            None => {
                info!("Ensemble coordinator not initialized, using fallback strategy");
                return self.get_fallback_trading_signal(symbol).await;
            },
        };

        // Extract features from market data
        let features = match self.extract_features_for_symbol(symbol).await {
            Ok(f) => f,
            Err(e) => {
                warn!(
                    "Feature extraction failed for {}: {}, using fallback",
                    symbol, e
                );
                return self.get_fallback_trading_signal(symbol).await;
            },
        };

        // Get ensemble prediction
        let ensemble_decision = match ensemble.predict(&features).await {
            Ok(decision) => decision,
            Err(e) => {
                warn!(
                    "Ensemble prediction failed for {}: {}, using fallback",
                    symbol, e
                );
                return self.get_fallback_trading_signal(symbol).await;
            },
        };

        // Convert ensemble decision to trading signal
        let action = match ensemble_decision.action {
            ml::ensemble::TradingAction::Buy => TradingActionType::Buy,
            ml::ensemble::TradingAction::Sell => TradingActionType::Sell,
            ml::ensemble::TradingAction::Hold => TradingActionType::Hold,
        };

        // Apply risk-based position sizing
        let position_size = self
            .calculate_position_size(
                symbol,
                ensemble_decision.confidence,
                ensemble_decision.disagreement_rate,
            )
            .await?;

        info!(
            "Ensemble signal for {}: action={:?}, confidence={:.3}, size={}",
            symbol, action, ensemble_decision.confidence, position_size
        );

        Ok(EnsembleTradingSignal {
            symbol: symbol.to_string(),
            action,
            confidence: ensemble_decision.confidence,
            position_size,
            disagreement_rate: ensemble_decision.disagreement_rate,
            model_votes: ensemble_decision.model_votes.clone(),
            timestamp: std::time::SystemTime::now(),
        })
    }

    /// Extract features from market data for a symbol.
    ///
    /// Attempts to build a 51-dimensional feature vector from recent market ticks
    /// retrieved via `market_data_repository.get_latest_prices`.  When real data is
    /// available the function derives price-based statistics (mean, std-dev, momentum,
    /// volume average).  All remaining dimensions are padded with zeros so the vector
    /// is always exactly 51 elements — matching the DQN/PPO input dimension used
    /// throughout this codebase.
    ///
    /// On any failure (network error, insufficient history, etc.) the function logs a
    /// `WARN`-level message and returns a zero-filled 51-dim vector rather than
    /// propagating an error, so the ensemble coordinator can still run with neutral
    /// inputs instead of crashing the prediction path.
    async fn extract_features_for_symbol(
        &self,
        symbol: &str,
    ) -> TradingServiceResult<ml::Features> {
        // ROADMAP: Replace tick-based approximation with full OHLCV bar pipeline
        // -----------------------------------------------------------------------
        // Current state: builds a 51-dim feature vector from the latest tick prices,
        // which only provides point-in-time price data without proper OHLCV bars or
        // technical indicators computed over bar windows.
        //
        // To upgrade:
        //   1. Add `get_ohlcv_bars(symbol, timeframe, count)` to MarketDataRepository
        //      that returns Vec<OhlcvBar> from the market data store (TimescaleDB or
        //      in-memory ring buffer fed by the BarAggregator).
        //   2. Compute the 21 technical indicators (SMA, EMA, RSI, MACD, BB, ATR, etc.)
        //      over the bar series using the existing indicator library in `ml/src/features/`.
        //   3. Append 25 microstructure features (spread, depth imbalance, VWAP deviation,
        //      trade flow toxicity) from the order book snapshots.
        //   4. Cache the computed 51-dim vector in a per-symbol DashMap so repeated
        //      predictions within the same bar window reuse the cached result.
        //   5. Wire the cached features into EnsembleCoordinator::generate_and_save_prediction
        //      so it accepts an optional Features parameter (see trading.rs roadmap).
        const FEATURE_DIM: usize = 51;

        let zero_features = || {
            let names: Vec<String> = (0..FEATURE_DIM).map(|i| format!("f{i}")).collect();
            ml::Features::new(vec![0.0_f64; FEATURE_DIM], names).with_symbol(symbol.to_string())
        };

        // Attempt to retrieve the latest ticks for this symbol.
        let ticks = match self
            .market_data_repository
            .get_latest_prices(&[symbol.to_string()])
            .await
        {
            Ok(t) => t,
            Err(e) => {
                tracing::info!(
                    "FEATURE EXTRACTION: no market ticks yet for {}: {} — \
                     using zero-filled features",
                    symbol,
                    e
                );
                return Ok(zero_features());
            },
        };

        if ticks.is_empty() {
            tracing::info!(
                "FEATURE EXTRACTION: no market ticks available for {} — \
                 using zero-filled features",
                symbol
            );
            return Ok(zero_features());
        }

        // Derive basic price statistics from available ticks.
        let prices: Vec<f64> = ticks.iter().map(|t| t.price).collect();
        let volumes: Vec<f64> = ticks.iter().map(|t| t.quantity).collect();
        let n = prices.len() as f64;

        let price_mean = prices.iter().sum::<f64>() / n;
        let price_var = prices.iter().map(|p| (p - price_mean).powi(2)).sum::<f64>() / n;
        let price_std = price_var.sqrt();

        let last_price = prices.last().copied().unwrap_or(price_mean);
        let first_price = prices.first().copied().unwrap_or(price_mean);
        // Momentum: return over the available window, clipped to [-1, 1].
        let momentum = if first_price > 0.0 {
            ((last_price - first_price) / first_price).clamp(-1.0, 1.0)
        } else {
            0.0
        };

        let vol_mean = volumes.iter().sum::<f64>() / n;
        // Normalised std-dev (coefficient of variation), clamped for stability.
        let price_cv = if price_mean > 0.0 {
            (price_std / price_mean).clamp(0.0, 1.0)
        } else {
            0.0
        };

        // Build the feature vector.  The first 6 slots hold real data; the rest are
        // zeros until a full indicator pipeline is wired in.
        let mut values = vec![0.0_f64; FEATURE_DIM];
        values[0] = price_mean;
        values[1] = price_std;
        values[2] = price_cv;
        values[3] = momentum;
        values[4] = vol_mean;
        values[5] = n / 1000.0; // tick count, normalised

        let names: Vec<String> = [
            "price_mean",
            "price_std",
            "price_cv",
            "momentum",
            "vol_mean",
            "tick_count_norm",
        ]
        .iter()
        .map(|s| s.to_string())
        .chain((6..FEATURE_DIM).map(|i| format!("f{i}")))
        .collect();

        tracing::debug!(
            "FEATURE EXTRACTION: {} ticks → price_mean={:.4} momentum={:.4} for {}",
            ticks.len(),
            price_mean,
            momentum,
            symbol
        );

        Ok(ml::Features::new(values, names).with_symbol(symbol.to_string()))
    }

    /// Calculate position size based on ensemble confidence and disagreement
    async fn calculate_position_size(
        &self,
        _symbol: &str,
        confidence: f64,
        disagreement_rate: f64,
    ) -> TradingServiceResult<u64> {
        // Base position size (in shares/contracts)
        let base_size: u64 = 100;

        // Scale by confidence (0.5-1.0 confidence → 0-1.0 multiplier)
        let confidence_multiplier = ((confidence - 0.5) * 2.0).max(0.0).min(1.0);

        // Reduce size based on disagreement (high disagreement = lower size)
        let disagreement_penalty = 1.0 - disagreement_rate;

        // Final position size
        let position_size =
            (base_size as f64 * confidence_multiplier * disagreement_penalty) as u64;

        Ok(position_size.max(10)) // Minimum 10 shares/contracts
    }

    /// Return an error when no ML prediction is available.
    ///
    /// Previously this returned a hardcoded Hold signal at 0.60 confidence,
    /// which looked like a legitimate medium-confidence prediction and could
    /// trigger trades.  Now it returns an explicit error so callers are forced
    /// to handle the absence of a real prediction (e.g. the gRPC layer returns
    /// `Status::unavailable` or `Status::internal`).
    async fn get_fallback_trading_signal(
        &self,
        symbol: &str,
    ) -> TradingServiceResult<EnsembleTradingSignal> {
        tracing::warn!(
            symbol = %symbol,
            "No ML prediction available: ensemble coordinator unavailable or \
             feature extraction / inference failed. Refusing to fabricate a \
             trading signal."
        );

        Err(TradingServiceError::MLModel {
            model_name: "ensemble".to_string(),
            message: format!(
                "No ML prediction available for {symbol}: ensemble coordinator \
                 unavailable or inference failed"
            ),
        })
    }
}

/// Trading signal with ensemble attribution
#[derive(Debug, Clone)]
pub struct EnsembleTradingSignal {
    pub symbol: String,
    pub action: TradingActionType,
    pub confidence: f64,
    pub position_size: u64,
    pub disagreement_rate: f64,
    pub model_votes: std::collections::HashMap<String, ml::ensemble::ModelVote>,
    pub timestamp: std::time::SystemTime,
}

/// Trading action type
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum TradingActionType {
    Buy,
    Sell,
    Hold,
}

/// Risk management engine wired to the real `risk` crate VaR calculators.
///
/// Provides two levels of VaR calculation:
/// - **Marginal VaR** via `risk::risk_engine::VarEngine` -- fast, parametric,
///   suitable for pre-trade per-order risk checks.
/// - **Comprehensive portfolio VaR** via `risk::RealVaREngine` -- multi-methodology
///   (historical simulation, parametric, Monte Carlo, hybrid) with stress testing
///   and risk decomposition.
pub struct RiskEngine {
    /// Fast parametric VaR engine for marginal/per-order risk checks
    var_engine: risk::risk_engine::VarEngine,
    /// Multi-methodology portfolio VaR engine
    portfolio_var_engine: risk::RealVaREngine,
    /// VaR confidence level loaded from config repository (default 0.95)
    var_confidence: f64,
    /// Maximum VaR limit loaded from config repository
    max_var_limit: f64,
}

impl std::fmt::Debug for RiskEngine {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("RiskEngine")
            .field("var_confidence", &self.var_confidence)
            .field("max_var_limit", &self.max_var_limit)
            .finish()
    }
}

impl Default for RiskEngine {
    fn default() -> Self {
        Self::new()
    }
}

impl RiskEngine {
    /// Create a new `RiskEngine` backed by real VaR calculators from the `risk` crate.
    pub fn new() -> Self {
        let var_config = config::structures::VarConfig::default();
        let var_engine = risk::risk_engine::VarEngine::with_defaults(var_config);
        let portfolio_var_engine = risk::RealVaREngine::new();
        Self {
            var_engine,
            portfolio_var_engine,
            var_confidence: 0.95,
            max_var_limit: 100_000.0,
        }
    }

    /// Initialize the risk engine with parameters from the config repository.
    ///
    /// Loads `var_confidence` and `max_var_limit` from the "Risk" config category.
    /// If a value is missing the default is retained.
    pub async fn initialize_with_config_repository(
        &mut self,
        config_repository: &Arc<PostgresConfigRepository>,
    ) -> TradingServiceResult<()> {
        // Load VaR confidence level from config repository
        if let Ok(Some(var_confidence)) = config_repository
            .get_config_f64("Risk", "var_confidence")
            .await
        {
            if (0.0..=1.0).contains(&var_confidence) {
                self.var_confidence = var_confidence;
                tracing::info!(
                    "Risk engine VaR confidence set to: {}",
                    var_confidence
                );
            } else {
                tracing::warn!(
                    "Invalid VaR confidence {} from config, keeping default {}",
                    var_confidence,
                    self.var_confidence
                );
            }
        }

        // Load max VaR limit from config repository
        if let Ok(Some(max_var_limit)) = config_repository
            .get_config_f64("Risk", "max_var_limit")
            .await
        {
            if max_var_limit > 0.0 {
                self.max_var_limit = max_var_limit;
                tracing::info!(
                    "Risk engine max VaR limit set to: {}",
                    max_var_limit
                );
            }
        }

        // Rebuild var_engine with updated configuration
        let var_config = config::structures::VarConfig {
            confidence_level: self.var_confidence,
            max_var_limit: self.max_var_limit,
            ..config::structures::VarConfig::default()
        };
        self.var_engine = risk::risk_engine::VarEngine::with_defaults(var_config);

        tracing::info!(
            "RiskEngine initialized with real VaR calculators (confidence={}, max_limit={})",
            self.var_confidence,
            self.max_var_limit
        );
        Ok(())
    }

    /// Calculate the marginal Value at Risk for a proposed order.
    ///
    /// Delegates to `risk::risk_engine::VarEngine::calculate_marginal_var` which
    /// computes VaR = position_value * daily_volatility * z_score(confidence).
    ///
    /// Returns the marginal VaR as `f64` for easy use in the trading service.
    pub async fn calculate_marginal_var(
        &self,
        account_id: &str,
        instrument_id: &str,
        quantity: f64,
        price: f64,
    ) -> Result<f64, String> {
        let quantity_decimal = rust_decimal::Decimal::try_from(quantity)
            .map_err(|e| format!("Invalid quantity for VaR: {e}"))?;
        let price_decimal = rust_decimal::Decimal::try_from(price)
            .map_err(|e| format!("Invalid price for VaR: {e}"))?;

        let marginal_var = self
            .var_engine
            .calculate_marginal_var(account_id, instrument_id, quantity_decimal, price_decimal)
            .await
            .map_err(|e| format!("VaR calculation failed: {e}"))?;

        use num_traits::ToPrimitive;
        marginal_var
            .to_f64()
            .ok_or_else(|| "Failed to convert VaR Decimal to f64".to_string())
    }

    /// Check whether a proposed order's VaR impact exceeds the configured limit.
    ///
    /// Returns `Ok(())` if within limits, or `Err(message)` describing the breach.
    pub async fn check_var_limit(
        &self,
        account_id: &str,
        instrument_id: &str,
        quantity: f64,
        price: f64,
    ) -> Result<(), String> {
        let marginal_var = self
            .calculate_marginal_var(account_id, instrument_id, quantity, price)
            .await?;

        if marginal_var > self.max_var_limit {
            Err(format!(
                "VaR limit exceeded: marginal VaR {:.2} > limit {:.2}",
                marginal_var, self.max_var_limit
            ))
        } else {
            Ok(())
        }
    }

    /// Get a reference to the comprehensive portfolio VaR engine.
    ///
    /// Callers can use this to run `calculate_comprehensive_var` and
    /// `check_circuit_breaker_conditions` from `risk::RealVaREngine`.
    pub fn portfolio_var_engine(&self) -> &risk::RealVaREngine {
        &self.portfolio_var_engine
    }

    /// Get the current VaR confidence level.
    pub fn var_confidence(&self) -> f64 {
        self.var_confidence
    }

    /// Get the current maximum VaR limit.
    pub fn max_var_limit(&self) -> f64 {
        self.max_var_limit
    }
}

/// `Position` state validator
#[derive(Debug, Default)]
pub struct PositionStateValidator {
    // Position validation logic
}

impl PositionStateValidator {
    pub fn new() -> Self {
        Self::default()
    }

    pub async fn initialize_with_config_repository(
        &mut self,
        config_repository: &Arc<PostgresConfigRepository>,
    ) -> TradingServiceResult<()> {
        // Initialize risk parameters from repository (no direct database access)
        // Load VaR confidence from config repository
        if let Ok(Some(var_confidence)) = config_repository
            .get_config_f64("Risk", "var_confidence")
            .await
        {
            tracing::info!(
                "Risk engine initialized with VaR confidence: {}",
                var_confidence
            );
        }

        // Load max drawdown limit from config repository
        if let Ok(Some(max_drawdown)) = config_repository
            .get_config_f64("Risk", "max_drawdown_limit")
            .await
        {
            tracing::info!(
                "Risk engine initialized with max drawdown limit: {}",
                max_drawdown
            );
        }

        Ok(())
    }
}

/// ML engine backed by the real `ml::ensemble::EnsembleCoordinator`.
///
/// The coordinator is lazily initialized during
/// [`initialize_with_config_repository`](MLEngine::initialize_with_config_repository)
/// because it needs async model registration.  Before that call,
/// [`coordinator()`](MLEngine::coordinator) returns `None`.
#[derive(Default)]
pub struct MLEngine {
    /// The real ensemble coordinator from the `ml` crate.
    /// `None` until [`initialize_with_config_repository`] completes.
    coordinator: Option<ml::ensemble::EnsembleCoordinator>,
}

impl std::fmt::Debug for MLEngine {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("MLEngine")
            .field("coordinator", &self.coordinator)
            .finish()
    }
}


impl MLEngine {
    pub fn new() -> Self {
        Self::default()
    }

    /// Initialize the ML engine by creating an `EnsembleCoordinator` and
    /// registering the default production models (DQN, PPO, TFT, MAMBA-2).
    ///
    /// Model weights and the inference timeout are loaded from the config
    /// repository.  If a config key is missing the default value is used.
    pub async fn initialize_with_config_repository(
        &mut self,
        config_repository: &Arc<PostgresConfigRepository>,
    ) -> TradingServiceResult<()> {
        // Load ML inference timeout from config repository
        if let Ok(Some(inference_timeout)) = config_repository
            .get_config_u64("MachineLearning", "inference_timeout_ms")
            .await
        {
            tracing::info!(
                "ML engine initialized with inference timeout: {}ms",
                inference_timeout
            );
        }

        // Create the real ensemble coordinator
        let coordinator = ml::ensemble::EnsembleCoordinator::new();

        // Load per-model weights from config, falling back to equal weights
        let default_weight = 0.25_f64;

        let dqn_weight = config_repository
            .get_config_f64("MachineLearning", "dqn_weight")
            .await
            .ok()
            .flatten()
            .unwrap_or(default_weight);

        let ppo_weight = config_repository
            .get_config_f64("MachineLearning", "ppo_weight")
            .await
            .ok()
            .flatten()
            .unwrap_or(default_weight);

        let tft_weight = config_repository
            .get_config_f64("MachineLearning", "tft_weight")
            .await
            .ok()
            .flatten()
            .unwrap_or(default_weight);

        let mamba2_weight = config_repository
            .get_config_f64("MachineLearning", "mamba2_weight")
            .await
            .ok()
            .flatten()
            .unwrap_or(default_weight);

        // Register the four production models
        if let Err(e) = coordinator
            .register_model("DQN".to_string(), dqn_weight)
            .await
        {
            tracing::error!("Failed to register DQN model: {}", e);
        }
        if let Err(e) = coordinator
            .register_model("PPO".to_string(), ppo_weight)
            .await
        {
            tracing::error!("Failed to register PPO model: {}", e);
        }
        if let Err(e) = coordinator
            .register_model("TFT".to_string(), tft_weight)
            .await
        {
            tracing::error!("Failed to register TFT model: {}", e);
        }
        if let Err(e) = coordinator
            .register_model("MAMBA-2".to_string(), mamba2_weight)
            .await
        {
            tracing::error!("Failed to register MAMBA-2 model: {}", e);
        }

        let model_count = coordinator.model_count().await;
        tracing::info!(
            "ML engine EnsembleCoordinator initialized with {} models \
             (DQN={:.2}, PPO={:.2}, TFT={:.2}, MAMBA-2={:.2})",
            model_count,
            dqn_weight,
            ppo_weight,
            tft_weight,
            mamba2_weight,
        );

        self.coordinator = Some(coordinator);

        Ok(())
    }

    /// Get a shared reference to the underlying `EnsembleCoordinator`.
    ///
    /// Returns `None` if the engine has not been initialized yet.
    pub fn coordinator(&self) -> Option<&ml::ensemble::EnsembleCoordinator> {
        self.coordinator.as_ref()
    }

    /// Get a mutable reference to the underlying `EnsembleCoordinator`.
    ///
    /// Useful for adding inference adapters via
    /// [`EnsembleCoordinator::add_adapter`].
    /// Returns `None` if the engine has not been initialized yet.
    pub fn coordinator_mut(&mut self) -> Option<&mut ml::ensemble::EnsembleCoordinator> {
        self.coordinator.as_mut()
    }
}

/// Market data manager with multiple providers
pub struct MarketDataManager {
    /// Databento provider for market data
    databento_provider:
        Option<Arc<RwLock<data::providers::databento_streaming::DatabentoStreamingProvider>>>,
    /// Benzinga provider for news data
    benzinga_provider: Option<
        Arc<RwLock<data::providers::benzinga::production_streaming::ProductionBenzingaProvider>>,
    >,
    /// Unified feature extractor
    feature_extractor: Option<Arc<ml::features::UnifiedFeatureExtractor>>,
    /// Event broadcast sender
    _event_sender: Arc<
        tokio::sync::broadcast::Sender<trading_engine::trading::data_interface::MarketDataEvent>,
    >,
}

impl std::fmt::Debug for MarketDataManager {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("MarketDataManager")
            .field("databento_provider", &self.databento_provider)
            .field("benzinga_provider", &self.benzinga_provider)
            .field("feature_extractor", &"<UnifiedFeatureExtractor>")
            .field("_event_sender", &self._event_sender)
            .finish()
    }
}

impl Default for MarketDataManager {
    fn default() -> Self {
        let (_event_sender, _) = tokio::sync::broadcast::channel(10000);
        Self {
            databento_provider: None,
            benzinga_provider: None,
            feature_extractor: None,
            _event_sender: Arc::new(_event_sender),
        }
    }
}

impl MarketDataManager {
    pub fn new() -> Self {
        Self::default()
    }

    pub async fn initialize(&mut self) -> TradingServiceResult<()> {
        // Initialize Databento provider if API key is available
        if let Ok(api_key) = std::env::var("DATABENTO_API_KEY") {
            match data::providers::databento_streaming::DatabentoStreamingProvider::new(api_key) {
                Ok(mut provider) => {
                    if let Err(e) = provider.connect().await {
                        tracing::info!("Databento connection not available: {}", e);
                    } else {
                        tracing::info!("Connected to Databento successfully");
                        self.databento_provider = Some(Arc::new(RwLock::new(provider)));
                    }
                },
                Err(e) => {
                    tracing::error!("Failed to create Databento provider: {}", e);
                },
            }
        } else {
            tracing::info!("DATABENTO_API_KEY not found, skipping Databento provider");
        }

        // Initialize Benzinga provider if API key is available
        if let Ok(api_key) = std::env::var("BENZINGA_API_KEY") {
            let benzinga_config =
                data::providers::benzinga::production_streaming::ProductionBenzingaConfig {
                    api_key,
                    websocket_url: "wss://api.benzinga.com/api/v1/news/stream".to_string(),
                    connect_timeout_secs: 30,
                    ping_interval_secs: 30,
                    max_reconnect_attempts: 3,
                    initial_reconnect_delay_ms: 1000,
                    max_reconnect_delay_ms: 60000,
                    reconnect_backoff_multiplier: 2.0,
                    enable_news: true,
                    enable_sentiment: true,
                    enable_ratings: true,
                    enable_options: false,
                    event_buffer_size: 1000,
                    heartbeat_timeout_secs: 300,
                    rate_limit_per_second: 10,
                    dedup_window_secs: 60,
                    max_dedup_cache_size: 10000,
                    enable_compression: true,
                    batch_processing_size: 100,
                    enable_smart_categorization: true,
                    enable_ml_integration: true,
                    circuit_breaker_threshold: 5,
                    circuit_breaker_timeout_secs: 300,
                    max_concurrent_processing: 4,
                };

            match data::providers::benzinga::production_streaming::ProductionBenzingaProvider::new(
                benzinga_config,
            ) {
                Ok(mut provider) => {
                    if let Err(e) = provider.connect().await {
                        tracing::info!("Benzinga connection not available: {}", e);
                    } else {
                        tracing::info!("Connected to Benzinga successfully");
                        self.benzinga_provider = Some(Arc::new(RwLock::new(provider)));
                    }
                },
                Err(e) => {
                    tracing::error!("Failed to create Benzinga provider: {}", e);
                },
            }
        } else {
            tracing::info!("BENZINGA_API_KEY not found, skipping Benzinga provider");
        }

        // Initialize UnifiedFeatureExtractor
        let config = ml::features::FeatureExtractionConfig::default();
        let safety_manager = Arc::new(ml::safety::MLSafetyManager::new(
            ml::safety::MLSafetyConfig::default(),
        ));
        self.feature_extractor = Some(Arc::new(ml::features::UnifiedFeatureExtractor::new(
            config,
            safety_manager,
        )));

        tracing::info!("MarketDataManager initialized with available providers");
        Ok(())
    }

    pub async fn initialize_with_config_repository(
        &mut self,
        config_repository: &Arc<PostgresConfigRepository>,
    ) -> TradingServiceResult<()> {
        // Initialize market data providers using repository-based configuration (no direct database access)

        // Get API keys from config repository
        if let Ok(Some(databento_key)) = config_repository.get_secret("databento_api_key").await {
            match data::providers::databento_streaming::DatabentoStreamingProvider::new(
                databento_key,
            ) {
                Ok(mut provider) => {
                    if let Err(e) = provider.connect().await {
                        tracing::info!("Databento connection not available: {}", e);
                    } else {
                        tracing::info!("Connected to Databento successfully via config repository");
                        self.databento_provider = Some(Arc::new(RwLock::new(provider)));
                    }
                },
                Err(e) => {
                    tracing::error!("Failed to create Databento provider: {}", e);
                },
            }
        } else {
            tracing::info!("Databento API key not found in config repository, skipping provider");
        }

        if let Ok(Some(benzinga_key)) = config_repository.get_secret("benzinga_api_key").await {
            let benzinga_config =
                data::providers::benzinga::production_streaming::ProductionBenzingaConfig {
                    api_key: benzinga_key,
                    websocket_url: "wss://api.benzinga.com/api/v1/news/stream".to_string(),
                    connect_timeout_secs: 30,
                    ping_interval_secs: 30,
                    max_reconnect_attempts: 3,
                    initial_reconnect_delay_ms: 1000,
                    max_reconnect_delay_ms: 60000,
                    reconnect_backoff_multiplier: 2.0,
                    enable_news: true,
                    enable_sentiment: true,
                    enable_ratings: true,
                    enable_options: false,
                    event_buffer_size: 1000,
                    heartbeat_timeout_secs: 300,
                    rate_limit_per_second: 10,
                    dedup_window_secs: 60,
                    max_dedup_cache_size: 10000,
                    enable_compression: true,
                    batch_processing_size: 100,
                    enable_smart_categorization: true,
                    enable_ml_integration: true,
                    circuit_breaker_threshold: 5,
                    circuit_breaker_timeout_secs: 300,
                    max_concurrent_processing: 4,
                };
            match data::providers::benzinga::production_streaming::ProductionBenzingaProvider::new(
                benzinga_config,
            ) {
                Ok(mut provider) => {
                    if let Err(e) = provider.connect().await {
                        tracing::info!("Benzinga connection not available: {}", e);
                    } else {
                        tracing::info!("Connected to Benzinga successfully via config repository");
                        self.benzinga_provider = Some(Arc::new(RwLock::new(provider)));
                    }
                },
                Err(e) => {
                    tracing::error!("Failed to create Benzinga provider: {}", e);
                },
            }
        } else {
            tracing::info!("Benzinga API key not found in config repository, skipping provider");
        }

        // Initialize UnifiedFeatureExtractor with configuration
        let config = ml::features::FeatureExtractionConfig::default();
        let safety_manager = Arc::new(ml::safety::MLSafetyManager::new(
            ml::safety::MLSafetyConfig::default(),
        ));
        self.feature_extractor = Some(Arc::new(ml::features::UnifiedFeatureExtractor::new(
            config,
            safety_manager,
        )));

        tracing::info!("MarketDataManager initialized with repository-based configuration");
        Ok(())
    }

    /// Subscribe to market data for given symbols
    pub async fn subscribe_to_symbols(
        &mut self,
        symbols: Vec<common::types::Symbol>,
    ) -> TradingServiceResult<()> {
        // Subscribe via Databento provider
        if let Some(databento) = &self.databento_provider {
            let mut provider = databento.write().await;
            let symbol_strings: Vec<String> = symbols.iter().map(|s| s.to_string()).collect();
            if let Err(e) = provider.subscribe(symbol_strings).await {
                tracing::error!("Failed to subscribe to Databento: {}", e);
            } else {
                tracing::info!("Subscribed to {} symbols on Databento", symbols.len());
            }
        }

        // Subscribe via Benzinga provider
        if let Some(benzinga) = &self.benzinga_provider {
            let mut provider = benzinga.write().await;
            if let Err(e) = provider.subscribe(symbols.clone()).await {
                tracing::error!("Failed to subscribe to Benzinga news: {}", e);
            } else {
                tracing::info!(
                    "Subscribed to news for {} symbols on Benzinga",
                    symbols.len()
                );
            }
        }

        Ok(())
    }

    /// Get market data event receiver
    pub fn get_event_receiver(
        &self,
    ) -> tokio::sync::broadcast::Receiver<trading_engine::trading::data_interface::MarketDataEvent>
    {
        self._event_sender.subscribe()
    }

    /// Start event processing from all providers
    pub async fn start_event_processing(&self) -> TradingServiceResult<()> {
        // Start processing events from Databento
        if let Some(databento) = &self.databento_provider {
            let provider = Arc::clone(databento);
            let _event_sender = Arc::clone(&self._event_sender);
            let feature_extractor = self.feature_extractor.clone();

            tokio::spawn(async move {
                let databento_provider = provider.read().await;
                let mut event_receiver = databento_provider.subscribe_market_events();
                drop(databento_provider); // Release the read lock

                while let Ok(event) = event_receiver.recv().await {
                    // Process event through feature extractor if available
                    if let Some(_extractor) = &feature_extractor {
                        // ROADMAP: Feature extraction pipeline integration
                        // ------------------------------------------------
                        // The feature extractor is instantiated but not yet wired into the
                        // market-event loop. To complete the integration:
                        //   1. Aggregate raw ticks into OHLCV bars (1m, 5m, 15m) via a
                        //      BarAggregator that buffers ticks per symbol and emits bars
                        //      on period boundaries.
                        //   2. Feed completed bars into the FeatureExtractor to produce
                        //      the 51-dim feature vector (5 OHLCV + 21 technical indicators
                        //      + 25 microstructure features) expected by the ML models.
                        //   3. Cache the latest feature vector per symbol in an Arc<DashMap>
                        //      so the EnsembleCoordinator can read it without re-computing.
                        //   4. Depends on: MarketDataRepository gaining a `get_ohlcv_bars()`
                        //      method (see extract_features_for_symbol roadmap in this file).
                        tracing::debug!("Processing market event through feature extractor");
                    }

                    // Forward event to subscribers
                    let _ = _event_sender.send(event);
                }
            });
        }

        // Start processing events from Benzinga
        if let Some(benzinga) = &self.benzinga_provider {
            let provider = Arc::clone(benzinga);
            let _event_sender = Arc::clone(&self._event_sender);

            tokio::spawn(async move {
                let mut benzinga_provider = provider.write().await;
                match benzinga_provider.stream().await {
                    Ok(mut stream) => {
                        drop(benzinga_provider); // Release the write lock

                        // Process events from the stream
                        while let Some(event) = stream.next().await {
                            // Forward news-derived market events to subscribers
                            let _ = _event_sender.send(event);
                        }
                    },
                    Err(e) => {
                        tracing::error!("Failed to get Benzinga stream: {}", e);
                    },
                }
            });
        }

        tracing::info!("Started event processing for all connected providers");
        Ok(())
    }

    /// Get health status of all providers
    pub async fn get_provider_health(
        &self,
    ) -> Vec<(String, data::providers::ProviderHealthStatus)> {
        let mut health_status = Vec::new();

        if let Some(databento) = &self.databento_provider {
            let provider = databento.read().await;
            health_status.push(("databento".to_string(), provider.get_health_status()));
        }

        if let Some(benzinga) = &self.benzinga_provider {
            let provider = benzinga.read().await;
            // Create ProviderHealthStatus from ConnectionStatus
            let connection_status = provider.get_connection_status();
            let health = data::providers::ProviderHealthStatus {
                connected: matches!(
                    connection_status.state,
                    data::providers::ConnectionState::Connected
                ),
                last_connected: connection_status.last_connection_attempt,
                active_subscriptions: connection_status.active_subscriptions,
                messages_per_second: connection_status.events_per_second,
                latency_micros: connection_status.latency_micros,
                error_count: 0, // Default value
            };
            health_status.push(("benzinga".to_string(), health));
        }

        health_status
    }
}
/// Health status enumeration
#[derive(Debug, Clone)]
pub enum HealthStatus {
    /// All systems operational
    Healthy,
    /// Some degraded performance
    Degraded,
    /// Critical issues present
    Unhealthy,
    /// Service offline
    Critical,
}

/// Ensemble coordinator health status
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum EnsembleHealth {
    /// All models loaded and operational
    Healthy,
    /// Some models missing or degraded performance
    Degraded,
    /// Critical issues (no models loaded, inference failing)
    Unhealthy,
    /// Ensemble coordinator not configured (fallback mode)
    NotConfigured,
}

/// Detailed ensemble health report
#[derive(Debug, Clone)]
pub struct EnsembleHealthReport {
    pub status: EnsembleHealth,
    pub models_loaded: usize,
    pub expected_models: usize,
    pub inference_latency_us: Option<f64>,
    pub last_prediction: Option<std::time::SystemTime>,
    pub model_details: Vec<ModelHealthDetail>,
}

/// Per-model health detail
#[derive(Debug, Clone)]
pub struct ModelHealthDetail {
    pub model_id: String,
    pub checkpoint: String,
    pub loaded: bool,
    pub last_inference: std::time::SystemTime,
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_risk_engine_creation() {
        let engine = RiskEngine::new();
        // Default confidence level is 0.95
        assert!((engine.var_confidence() - 0.95).abs() < f64::EPSILON);
        // Default max VaR limit is 100_000.0
        assert!((engine.max_var_limit() - 100_000.0).abs() < f64::EPSILON);
    }

    #[test]
    fn test_risk_engine_default() {
        let engine = RiskEngine::default();
        assert!((engine.var_confidence() - 0.95).abs() < f64::EPSILON);
        assert!((engine.max_var_limit() - 100_000.0).abs() < f64::EPSILON);
    }

    #[test]
    fn test_risk_engine_debug() {
        let engine = RiskEngine::new();
        let debug_str = format!("{engine:?}");
        assert!(debug_str.contains("RiskEngine"));
        assert!(debug_str.contains("var_confidence"));
        assert!(debug_str.contains("max_var_limit"));
    }

    #[test]
    fn test_portfolio_var_engine_accessible() {
        let engine = RiskEngine::new();
        // Verify the portfolio VaR engine is accessible and functional
        let _portfolio_engine = engine.portfolio_var_engine();
    }

    #[tokio::test]
    async fn test_marginal_var_calculation() {
        let engine = RiskEngine::new();
        // AAPL at $175, 100 shares -- should produce a positive VaR
        let result = engine
            .calculate_marginal_var("test_account", "AAPL", 100.0, 175.0)
            .await;
        assert!(result.is_ok(), "Marginal VaR calculation failed: {result:?}");
        let var_value = result.unwrap_or(0.0);
        assert!(var_value > 0.0, "Marginal VaR should be positive, got {var_value}");
    }

    #[tokio::test]
    async fn test_marginal_var_crypto_higher_than_equity() {
        let engine = RiskEngine::new();
        // Crypto (BTC) should have higher volatility than equity (AAPL)
        // Same notional value for comparison
        let btc_var = engine
            .calculate_marginal_var("test_account", "BTC", 1.0, 50_000.0)
            .await
            .unwrap_or(0.0);
        let aapl_var = engine
            .calculate_marginal_var("test_account", "AAPL", 285.7, 175.0)
            .await
            .unwrap_or(0.0);
        // Both should be non-zero and BTC should have higher VaR
        assert!(btc_var > 0.0, "BTC VaR should be positive");
        assert!(aapl_var > 0.0, "AAPL VaR should be positive");
        assert!(
            btc_var > aapl_var,
            "BTC VaR ({btc_var:.2}) should exceed AAPL VaR ({aapl_var:.2})"
        );
    }

    #[tokio::test]
    async fn test_check_var_limit_within_limit() {
        let engine = RiskEngine::new();
        // Small order should be within the 100k default limit
        let result = engine
            .check_var_limit("test_account", "AAPL", 10.0, 175.0)
            .await;
        assert!(result.is_ok(), "Small order should pass VaR limit check");
    }

    #[tokio::test]
    async fn test_check_var_limit_exceeds_limit() {
        let mut engine = RiskEngine::new();
        // Set a very low VaR limit
        engine.max_var_limit = 1.0;
        let result = engine
            .check_var_limit("test_account", "AAPL", 1000.0, 175.0)
            .await;
        assert!(result.is_err(), "Large order should breach tiny VaR limit");
        let err_msg = result.unwrap_err();
        assert!(
            err_msg.contains("VaR limit exceeded"),
            "Error message should describe VaR breach: {err_msg}"
        );
    }

    #[tokio::test]
    async fn test_marginal_var_invalid_inputs() {
        let engine = RiskEngine::new();
        // Zero quantity should still work (VaR = 0 -> non-positive -> error from risk crate)
        let result = engine
            .calculate_marginal_var("test_account", "AAPL", 0.0, 175.0)
            .await;
        // The risk crate rejects non-positive VaR results, so this should fail
        assert!(result.is_err(), "Zero quantity should produce an error");
    }
}