//! Agent 37: Kelly Criterion Position Sizing Tests
//!
//! Test suite for Kelly criterion integration in DQN training.
//! Tests verify that position sizes adapt dynamically based on edge estimates.

use anyhow::Result;
use ml::trainers::dqn::{DQNHyperparameters, DQNTrainer};

/// Test 1: Positive edge → Non-zero position size
///
/// When DQN estimates positive expected value (edge), Kelly should recommend
/// a non-zero position size proportional to the edge.
#[tokio::test]
async fn test_positive_edge_kelly_sizing() -> Result<()> {
    // Create DQN trainer with Kelly enabled
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_fraction = 0.25; // Conservative 25% Kelly
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Simulate historical rewards with 60% win rate and 2:1 win/loss ratio
    // Kelly = (p * b - q) / b = (0.6 * 2 - 0.4) / 2 = 0.4
    // Position size = 0.4 * 0.25 * max_position = 0.1 * max_position = 10
    let win_rate = 0.6;
    let avg_win = 2.0;
    let avg_loss = 1.0;

    let position_size = trainer.calculate_kelly_position_size(win_rate, avg_win, avg_loss)?;

    assert!(position_size > 0.0, "Positive edge should yield non-zero position");
    assert!(
        position_size <= 100.0,
        "Position size should not exceed max position"
    );

    // Expected: ~10.0 (with some tolerance)
    let expected = 10.0;
    assert!(
        (position_size - expected).abs() < 1.0,
        "Position size {:.2} should be close to expected {:.2}",
        position_size,
        expected
    );

    Ok(())
}

/// Test 2: Negative edge → Zero position size
///
/// When expected value is negative (losing strategy), Kelly should recommend
/// zero position size to avoid the trade.
#[tokio::test]
async fn test_negative_edge_zero_position() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_fraction = 0.25;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Losing strategy: 40% win rate, 1:2 win/loss ratio
    // Kelly = (0.4 * 0.5 - 0.6) / 0.5 = -0.8 (negative)
    let win_rate = 0.4;
    let avg_win = 1.0;
    let avg_loss = 2.0;

    let position_size = trainer.calculate_kelly_position_size(win_rate, avg_win, avg_loss)?;

    assert_eq!(
        position_size, 0.0,
        "Negative edge should yield zero position"
    );

    Ok(())
}

/// Test 3: Conservative Kelly fraction prevents over-leveraging
///
/// Even with strong edge, conservative Kelly fraction (0.25-0.50) should
/// prevent excessive position sizes.
#[tokio::test]
async fn test_conservative_kelly_prevents_overleveraging() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_fraction = 0.25; // 25% of full Kelly
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Very strong edge: 80% win rate, 3:1 win/loss ratio
    // Full Kelly = (0.8 * 3 - 0.2) / 3 = 0.733
    // Conservative Kelly = 0.733 * 0.25 = 0.183
    // Position = 0.183 * 100 = 18.3
    let win_rate = 0.8;
    let avg_win = 3.0;
    let avg_loss = 1.0;

    let position_size = trainer.calculate_kelly_position_size(win_rate, avg_win, avg_loss)?;

    // Should be capped at ~18% of max position (not 73%)
    let max_allowed = 20.0;
    assert!(
        position_size <= max_allowed,
        "Position size {:.2} should be capped by conservative fraction (max: {:.2})",
        position_size,
        max_allowed
    );

    Ok(())
}

/// Test 4: Position sizing adapts to Q-value edge estimates
///
/// Position sizes should dynamically adjust as DQN learns and Q-value
/// estimates change over training.
#[tokio::test]
async fn test_position_sizing_adapts_to_edge() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_fraction = 0.5;
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Scenario 1: Early training, poor edge
    let position_size_early = trainer.calculate_kelly_position_size(0.51, 1.1, 1.0)?;

    // Scenario 2: After learning, stronger edge
    let position_size_late = trainer.calculate_kelly_position_size(0.65, 1.5, 1.0)?;

    assert!(
        position_size_late > position_size_early,
        "Better edge ({:.2} vs {:.2}) should yield larger position",
        position_size_late,
        position_size_early
    );

    Ok(())
}

/// Test 5: Kelly sizing respects max position limits
///
/// Even with infinite edge, position size should never exceed absolute max.
#[tokio::test]
async fn test_kelly_respects_max_position() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_fraction = 1.0; // Full Kelly (aggressive)
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Extreme edge: 99% win rate, 10:1 win/loss ratio
    // Full Kelly would be huge, but should be capped
    let win_rate = 0.99;
    let avg_win = 10.0;
    let avg_loss = 1.0;

    let position_size = trainer.calculate_kelly_position_size(win_rate, avg_win, avg_loss)?;

    assert!(
        position_size <= 100.0,
        "Position size {:.2} must not exceed max position 100.0",
        position_size
    );

    Ok(())
}

/// Test 6: Kelly sizing disabled by default (backward compatibility)
///
/// When Kelly sizing is disabled, system should fall back to fixed position
/// sizing based on exposure levels.
#[tokio::test]
async fn test_kelly_disabled_fallback() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = false; // Disabled
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Even with great edge, should use fixed sizing
    let win_rate = 0.8;
    let avg_win = 3.0;
    let avg_loss = 1.0;

    let position_size = trainer.calculate_kelly_position_size(win_rate, avg_win, avg_loss)?;

    // Should return fixed position size (50% default exposure when Kelly disabled)
    assert!(
        position_size > 0.0,
        "Should still return valid position size when Kelly disabled"
    );

    // Should be exactly 50% of max position
    assert_eq!(
        position_size,
        50.0,
        "Fixed sizing should be 50% of max position when Kelly disabled"
    );

    Ok(())
}

/// Test 7: Insufficient historical data handling
///
/// When insufficient data exists to calculate reliable Kelly fractions,
/// system should fall back to conservative default sizing.
#[tokio::test]
async fn test_insufficient_data_fallback() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_min_samples = 30; // Require 30 trades minimum
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Only 10 trades in history (insufficient) - use calculate_kelly_position_size_with_check
    let position_size = trainer.calculate_kelly_position_size_with_check()?;

    // Should use conservative default (25% of max position)
    assert!(
        position_size > 0.0,
        "Should use conservative default with insufficient data"
    );

    assert_eq!(
        position_size,
        25.0,
        "Conservative default should be 25% of max position"
    );

    Ok(())
}

/// Test 8: Kelly stats tracking
///
/// Verify that Kelly statistics are properly tracked and accessible.
#[tokio::test]
async fn test_kelly_stats_tracking() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;
    hyperparams.kelly_fraction = 0.25;
    hyperparams.kelly_min_samples = 20;
    hyperparams.max_position = 100.0;

    let trainer = DQNTrainer::new(hyperparams)?;

    let stats = trainer.get_kelly_stats();

    assert!(stats.enabled, "Kelly should be enabled");
    assert_eq!(stats.total_trades, 0, "No trades yet");
    assert_eq!(stats.win_count, 0, "No wins yet");
    assert_eq!(stats.win_rate, 0.0, "Win rate should be 0");
    assert_eq!(stats.kelly_fraction, 0.25, "Kelly fraction should be 0.25");
    assert!(!stats.sufficient_data, "Insufficient data initially");

    Ok(())
}

/// Test 9: Input validation
///
/// Verify that invalid inputs are properly rejected.
#[tokio::test]
async fn test_input_validation() -> Result<()> {
    let mut hyperparams = DQNHyperparameters::conservative();
    hyperparams.enable_kelly_sizing = true;

    let trainer = DQNTrainer::new(hyperparams)?;

    // Invalid win rate (> 1.0)
    let result = trainer.calculate_kelly_position_size(1.5, 2.0, 1.0);
    assert!(result.is_err(), "Should reject invalid win rate");

    // Invalid win rate (< 0.0)
    let result = trainer.calculate_kelly_position_size(-0.5, 2.0, 1.0);
    assert!(result.is_err(), "Should reject negative win rate");

    // Zero average win
    let result = trainer.calculate_kelly_position_size(0.6, 0.0, 1.0);
    assert!(result.is_err(), "Should reject zero average win");

    // Negative average loss
    let result = trainer.calculate_kelly_position_size(0.6, 2.0, -1.0);
    assert!(result.is_err(), "Should reject negative average loss");

    Ok(())
}