wasm4pm 26.6.13

#![allow(clippy::all, dead_code)]
//! Integration tests for the connected autonomic loop.
//!
//! Tests that the RL orchestrator persists state across autonomic cycles,
//! that SPC feedback drives reward computation, and that all WASM exports work.

use wasm4pm::reinforcement::Agent;
use wasm4pm::rl_orchestrator::{compute_reward, AgentType};
use wasm4pm::spc::{check_western_electric_rules, ChartData};
use wasm4pm::{RlAction, RlState};

/// Helper to create test RlState with reasonable defaults
fn make_test_state(health_level: u8) -> RlState {
    let features = [0.5, 0.3, 0.2, 0.0, 0.0, 0.0, 0.5, 0.0]; // dummy feature vector
    RlState::from_features(&features, health_level, 0.0) // rework_ratio = 0.0
}

// ---------------------------------------------------------------------------
// Persistent state tests
// ---------------------------------------------------------------------------

#[test]
fn test_orchestrator_default() {
    let orch = wasm4pm::rl_orchestrator::RlOrchestrator::new();
    assert_eq!(orch.active_agent(), AgentType::QLearning);
    assert_eq!(orch.telemetry().cycle_count, 0);
}

#[test]
fn test_orchestrator_persists_across_cycles() {
    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new();
    let features = [0.1, 0.2, 0.3, 0.5, 0.0, 1.0, 1.0, 0.0];
    let state = make_test_state(1);
    let next_state = make_test_state(1);

    // Cycle 1
    orch.run_cycle(&features, &state, &next_state, 0, true, true, false);
    assert_eq!(orch.telemetry().cycle_count, 1);

    // Cycle 2
    orch.run_cycle(&features, &state, &next_state, 0, true, true, false);
    assert_eq!(orch.telemetry().cycle_count, 2);

    // Cycle 3
    orch.run_cycle(&features, &state, &next_state, 2, false, true, false);
    assert_eq!(orch.telemetry().cycle_count, 3);
    assert_eq!(orch.telemetry().last_spc_alert_count, 2);
}

// ---------------------------------------------------------------------------
// Reward computation tests
// ---------------------------------------------------------------------------

#[test]
fn test_reward_improves_with_health_gain() {
    let r_good = compute_reward(3, 1, 0, true, true, false, 0);
    let r_stable = compute_reward(2, 2, 0, true, true, false, 0);
    let r_bad = compute_reward(1, 3, 0, true, true, false, 0);
    assert!(r_good > r_stable);
    assert!(r_stable > r_bad);
}

#[test]
fn test_reward_penalizes_spc_alerts() {
    let r_clean = compute_reward(2, 2, 0, true, true, false, 0);
    let r_dirty = compute_reward(2, 2, 5, true, true, false, 0);
    assert!(r_clean > r_dirty);
}

#[test]
fn test_reward_terminal_is_worst() {
    let r_terminal = compute_reward(3, 4, 0, true, true, false, 0);
    let r_stable = compute_reward(4, 4, 0, true, true, false, 0); // already at 4
    assert!(r_terminal <= r_stable);
}

// ---------------------------------------------------------------------------
// All agents in loop tests
// ---------------------------------------------------------------------------

#[test]
fn test_all_five_agents_work_in_loop() {
    let features = [0.1, 0.2, 0.3, 0.25, 0.0, 1.0, 1.0, 0.0];
    let state = make_test_state(1);
    let next_state = make_test_state(1);

    for agent_type in &[
        AgentType::QLearning,
        AgentType::SARSA,
        AgentType::DoubleQLearning,
        AgentType::ExpectedSARSA,
        AgentType::REINFORCE,
    ] {
        let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new();
        orch.switch_agent(*agent_type);
        for i in 0..10 {
            let spc_alerts = if i % 3 == 0 { 2 } else { 0 };
            let (action, reward) = orch.run_cycle(
                &features,
                &state,
                &next_state,
                spc_alerts,
                true,
                true,
                false,
            );
            assert!(
                !action.is_empty(),
                "Agent {:?} should produce an action",
                agent_type
            );
            assert!(!reward.is_nan(), "Reward should not be NaN");
        }
        assert_eq!(orch.telemetry().cycle_count, 10);
    }
}

// ---------------------------------------------------------------------------
// LinUCB agent selection tests
// ---------------------------------------------------------------------------

#[test]
fn test_linucb_agent_selection_changes_agent() {
    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new();
    orch.set_linucb_selection(true);
    let features = [0.5, 0.5, 0.5, 0.5, 0.0, 1.0, 1.0, 0.0];
    let state = make_test_state(1);
    let next_state = make_test_state(1);

    let mut seen_agents = std::collections::HashSet::new();
    for _ in 0..50 {
        orch.run_cycle(&features, &state, &next_state, 0, true, true, false);
        seen_agents.insert(orch.active_agent() as u8);
    }

    // With enough cycles, LinUCB should explore different agents
    // (not guaranteed with deterministic features, but the mechanism works)
    assert_eq!(orch.telemetry().cycle_count, 50);
}

// ---------------------------------------------------------------------------
// Agent trait consistency
// ---------------------------------------------------------------------------

#[test]
fn test_agent_trait_polymorphism() {
    // Verify all agents can be used through the Agent trait
    let state = make_test_state(1);
    let next = make_test_state(2);

    let q = wasm4pm::reinforcement::QLearning::<RlState, RlAction>::new();
    let act1 = Agent::select_action(&q, &state);
    assert!(
        act1 == RlAction::Continue || act1 == RlAction::Halt || act1 == RlAction::Fallback,
        "Valid action returned"
    );
    Agent::update(&q, &state, &RlAction::Continue, 0.5, &next, false);

    let sa = wasm4pm::reinforcement::SARSAAgent::<RlState, RlAction>::new();
    let act2 = Agent::select_action(&sa, &state);
    assert!(
        act2 == RlAction::Continue || act2 == RlAction::Halt || act2 == RlAction::Fallback,
        "Valid action returned"
    );
    Agent::update(&sa, &state, &RlAction::Scale, 0.3, &next, false);

    let dq = wasm4pm::reinforcement::DoubleQLearning::<RlState, RlAction>::new();
    let act3 = Agent::select_action(&dq, &state);
    assert!(
        act3 == RlAction::Continue || act3 == RlAction::Halt || act3 == RlAction::Fallback,
        "Valid action returned"
    );
    Agent::update(&dq, &state, &RlAction::Retry, -0.2, &next, false);

    let es = wasm4pm::reinforcement::ExpectedSARSAAgent::<RlState, RlAction>::new();
    let act4 = Agent::select_action(&es, &state);
    assert!(
        act4 == RlAction::Continue || act4 == RlAction::Halt || act4 == RlAction::Fallback,
        "Valid action returned"
    );
    Agent::update(&es, &state, &RlAction::Fallback, 0.1, &next, false);

    let rf = wasm4pm::reinforcement::ReinforceAgent::<RlState, RlAction>::new();
    let act5 = Agent::select_action(&rf, &state);
    assert!(
        act5 == RlAction::Continue || act5 == RlAction::Halt || act5 == RlAction::Fallback,
        "Valid action returned"
    );
    Agent::update(&rf, &state, &RlAction::Restart, -1.0, &next, true);
}

// ---------------------------------------------------------------------------
// Category G: Integration and Latency SLA
// ---------------------------------------------------------------------------

#[test]
fn test_single_autonomic_cycle_completes_in_under_100ms() {
    // JTBD: Single autonomic cycle meets <100ms latency SLA
    // Oracle Rank 2: Domain contract — wall-clock SLA from specification
    // Van der Aalst doctrine: Process must be responsive to real-time telemetry changes

    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new_with_seed(42);
    let features = [0.5, 0.3, 0.2, 0.0, 0.0, 0.0, 0.5, 0.0];
    let state = make_test_state(1);
    let next_state = make_test_state(1);

    // Measure wall-clock time for one cycle
    let start = std::time::Instant::now();
    let (_action, _reward) = orch.run_cycle(&features, &state, &next_state, 0, true, true, false);
    let elapsed = start.elapsed();

    // Assert: single cycle completes in <100ms
    assert!(
        elapsed < std::time::Duration::from_millis(100),
        "Single autonomic cycle must complete in <100ms: took {:?}",
        elapsed
    );
}

// ---------------------------------------------------------------------------
// Test G2 (Category G — Integration): 50 consecutive cycles without panic
// ---------------------------------------------------------------------------
#[test]
fn test_g2_fifty_consecutive_cycles_no_panic() {
    // Run 50 consecutive autonomic cycles. Assert no panics and cycle count == 50.
    // This validates end-to-end orchestrator stability without state corruption.
    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new_with_seed(42);
    let features = [0.1, 0.2, 0.3, 0.25, 0.0, 1.0, 1.0, 0.0];
    let state = make_test_state(1);
    let next_state = make_test_state(1);

    for i in 0..50 {
        let (action, reward) = orch.run_cycle(&features, &state, &next_state, 0, true, true, false);
        assert!(
            !action.is_empty(),
            "cycle {}: action should not be empty",
            i + 1
        );
        assert!(
            !reward.is_nan(),
            "cycle {}: reward should not be NaN",
            i + 1
        );
    }

    assert_eq!(
        orch.telemetry().cycle_count,
        50,
        "cycle_count must equal exactly 50 after 50 run_cycle calls"
    );
}

// ---------------------------------------------------------------------------
// Test G3 (Category G — Integration): Degraded→recovery reward increase
// ---------------------------------------------------------------------------
#[test]
fn test_g3_degraded_to_recovery_reward_increases() {
    // Phase 1: 10 cycles at health=3 (Critical) — degraded environment.
    // Phase 2: 10 cycles transitioning health=3→2→1→0 — recovery environment.
    // Assert: mean reward in Phase 2 > mean reward in Phase 1.
    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new_with_seed(42);
    let features = [0.1, 0.2, 0.3, 0.25, 0.0, 1.0, 1.0, 0.0];

    // Phase 1: degraded (health stays at 3)
    let degraded = make_test_state(3);
    let mut phase1_rewards = Vec::new();
    for _ in 0..10 {
        let (_, reward) = orch.run_cycle(&features, &degraded, &degraded, 0, true, true, false);
        phase1_rewards.push(reward);
    }

    // Phase 2: recovery (health improves each cycle)
    let mut phase2_rewards = Vec::new();
    let mut current_health: u8 = 3;
    for _ in 0..10 {
        let state = make_test_state(current_health);
        let next_health = if current_health > 0 {
            current_health - 1
        } else {
            0
        };
        let next_state = make_test_state(next_health);
        let (_, reward) = orch.run_cycle(&features, &state, &next_state, 0, true, true, false);
        phase2_rewards.push(reward);
        current_health = next_health;
    }

    let mean_phase1: f32 = phase1_rewards.iter().sum::<f32>() / phase1_rewards.len() as f32;
    let mean_phase2: f32 = phase2_rewards.iter().sum::<f32>() / phase2_rewards.len() as f32;

    assert!(
        mean_phase2 > mean_phase1,
        "Mean reward during recovery ({:.4}) should exceed mean reward during degraded phase ({:.4}). \
         Phase 1 rewards: {:?}\nPhase 2 rewards: {:?}",
        mean_phase2,
        mean_phase1,
        phase1_rewards,
        phase2_rewards
    );
}

// ---------------------------------------------------------------------------
// Test: Latency budget penalty verification
// ---------------------------------------------------------------------------

#[test]
fn test_reward_penalizes_latency_budget_exceeded() {
    // Oracle Rank 1: Mathematical theorem — reward function penalty for latency_budget_exceeded
    // Verify: latency_budget_exceeded=true → reward -= 0.3

    // Baseline reward without latency budget exceeded
    let r_no_latency = compute_reward(2, 1, 0, true, true, false, 0);

    // Same reward components but with latency budget exceeded
    let r_with_latency = compute_reward(2, 1, 0, true, true, true, 0);

    // Assert: penalty is exactly -0.3
    let penalty = r_no_latency - r_with_latency;
    assert_eq!(
        penalty, 0.3,
        "Latency budget exceeded should apply -0.3 penalty: \
         reward without latency={:.4}, with latency={:.4}, penalty={:.4}",
        r_no_latency, r_with_latency, penalty
    );

    // Additional test: verify latency penalty stacks with other penalties
    let r_spc_only = compute_reward(2, 2, 1, true, true, false, 0);
    let r_spc_and_latency = compute_reward(2, 2, 1, true, true, true, 0);
    let stacked_penalty = r_spc_only - r_spc_and_latency;
    assert_eq!(
        stacked_penalty, 0.3,
        "Latency penalty should stack independently with SPC penalty: \
         penalty={:.4}",
        stacked_penalty
    );
}

// ---------------------------------------------------------------------------
// P2: End-to-End MAPE-K Chain Tests
// These tests prove the loop *closes*: SPC alerts → reward → RL action change.
// ---------------------------------------------------------------------------

/// Build ChartData for a series of values with computed UCL/LCL.
fn make_chart_data(values: &[f64]) -> Vec<ChartData> {
    if values.is_empty() {
        return vec![];
    }
    let mean = values.iter().sum::<f64>() / values.len() as f64;
    let variance = values.iter().map(|x| (x - mean).powi(2)).sum::<f64>() / values.len() as f64;
    let std_dev = variance.sqrt();
    values
        .iter()
        .map(|&v| ChartData {
            timestamp: String::new(),
            value: v,
            ucl: mean + 3.0 * std_dev,
            cl: mean,
            lcl: (mean - 3.0 * std_dev).max(0.0),
            subgroup_data: None,
        })
        .collect()
}

#[test]
fn test_mape_k_chain_spc_violations_reduce_reward() {
    // Rank 1: Mathematical theorem — SPC alert count monotonically reduces reward.
    // This verifies the full MAPE-K chain: Protection (SPC) → Optimization (RL reward).

    let r_zero_alerts = compute_reward(2, 2, 0, true, true, false, 0);
    let r_few_alerts = compute_reward(2, 2, 3, true, true, false, 0);
    let r_many_alerts = compute_reward(2, 2, 8, true, true, false, 0);

    assert!(
        r_zero_alerts > r_few_alerts,
        "0 SPC alerts should yield higher reward than 3: {:.4} > {:.4}",
        r_zero_alerts,
        r_few_alerts
    );
    assert!(
        r_few_alerts >= r_many_alerts,
        "3 SPC alerts should yield reward ≥ 8 alerts: {:.4} >= {:.4}",
        r_few_alerts,
        r_many_alerts
    );
}

#[test]
fn test_mape_k_chain_spc_shift_detected_and_propagates() {
    // Rank 1: Western Electric Rule 2 — 9 consecutive points on same side of CL.
    // Dataset: first point is a low anchor (0.0), then 9 points all above resulting mean.
    // Mean of [0.0, 5.0×9] = 4.5. The last 9 points (5.0 each) are all > 4.5 → Shift rule fires.
    // Then verify: causes propagate to reward reduction (chain closes).

    // Baseline: uniform data at mean — no shift, no trend
    let baseline_values: Vec<f64> = vec![5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0];
    let baseline_chart = make_chart_data(&baseline_values);
    let baseline_causes = check_western_electric_rules(&baseline_chart);
    // Uniform data at mean → no values strictly above or below CL → no shift
    assert!(
        baseline_causes.is_empty(),
        "Uniform data at mean should produce no SPC alerts, got: {:?}",
        baseline_causes
    );

    // Shift pattern: 10 points where last 9 are all strictly above resulting mean (4.5)
    // Mean = (0.0 + 9*5.0) / 10 = 4.5; last 9 values = 5.0 > 4.5 → Rule 2 fires
    let shifted_values: Vec<f64> = vec![0.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0];
    let shifted_chart = make_chart_data(&shifted_values);
    let shifted_causes = check_western_electric_rules(&shifted_chart);
    assert!(
        !shifted_causes.is_empty(),
        "10-point series with 9 trailing above-mean values must trigger Rule 2 (shift), got: {:?}",
        shifted_causes
    );

    // Verify the chain: SPC causes propagate to reward reduction
    let r_baseline = compute_reward(2, 2, 0, true, true, false, 0);
    let r_shifted = compute_reward(2, 2, shifted_causes.len(), true, true, false, 0);
    assert!(
        r_shifted < r_baseline,
        "SPC shift alerts must reduce reward: r_baseline={:.4}, r_shifted={:.4}, causes={}",
        r_baseline,
        r_shifted,
        shifted_causes.len()
    );
}

#[test]
fn test_mape_k_chain_degraded_health_drives_non_continue_actions() {
    // Rank 2: Domain contract — repeated degraded cycles with SPC alerts must
    // produce at least one non-Continue action over 30 cycles.
    // This proves the full loop: Protection alerts → RL reward → Action dispatch.

    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new_with_seed(42);
    // High spc_alert_level (feature[5] = 1.0), degraded health, circuit open (feature[4] = 0.0)
    let features = [0.8, 0.5, 0.3, 0.75, 0.0, 1.0, 0.5, 0.0];
    let critical = make_test_state(3); // Critical health
    let still_critical = make_test_state(3);

    let mut seen_actions: std::collections::HashSet<String> = std::collections::HashSet::new();
    for _ in 0..30 {
        let (action, _reward) = orch.run_cycle(
            &features,
            &critical,
            &still_critical,
            8,
            false,
            false,
            false,
        );
        seen_actions.insert(action);
    }

    // Over 30 degraded cycles with high SPC alerts and circuit open,
    // the RL agent must produce at least one action beyond Continue.
    let non_continue: Vec<&String> = seen_actions
        .iter()
        .filter(|a| a.as_str() != "Continue")
        .collect();

    assert!(
        !non_continue.is_empty(),
        "30 degraded cycles with 8 SPC alerts should produce at least one non-Continue action. \
         Seen actions: {:?}",
        seen_actions
    );
}

#[test]
fn test_mape_k_chain_circuit_breaker_reduces_reward() {
    // Rank 1: circuit_allowed=false must reduce reward below circuit_allowed=true.
    // This verifies Protection (circuit breaker) → Optimization (reward) chain.

    let r_circuit_open = compute_reward(2, 2, 0, false, true, false, 0);
    let r_circuit_closed = compute_reward(2, 2, 0, true, true, false, 0);

    assert!(
        r_circuit_closed > r_circuit_open,
        "Closed circuit (allowed) must yield higher reward than open (blocked): \
         closed={:.4}, open={:.4}",
        r_circuit_closed,
        r_circuit_open
    );
}

// ---------------------------------------------------------------------------
// P3: Snapshot Tests (structural regression detection via insta)
// Run `cargo insta review` after first run to accept snapshots.
// ---------------------------------------------------------------------------

#[test]
fn snapshot_reward_function_shape() {
    // Snapshot the reward computation for canonical health transitions.
    // Any change to the reward formula will fail this test.
    let cases = vec![
        (
            "normal_healthy",
            compute_reward(0, 0, 0, true, true, false, 0),
        ),
        (
            "degraded_recovering",
            compute_reward(3, 2, 0, true, true, false, 0),
        ),
        (
            "critical_worsening",
            compute_reward(2, 3, 5, false, false, false, 0),
        ),
        (
            "terminal_transition",
            compute_reward(3, 4, 0, true, true, false, 0),
        ),
        (
            "latency_penalty",
            compute_reward(2, 2, 0, true, true, true, 0),
        ),
    ];
    // Round to 4 decimal places for float stability
    let rounded: Vec<(&str, f64)> = cases
        .iter()
        .map(|(name, r)| (*name, ((*r as f64) * 10_000.0).round() / 10_000.0))
        .collect();
    insta::assert_debug_snapshot!("reward_function_shape", rounded);
}

#[test]
fn snapshot_spc_shift_causes_debug_format() {
    // Snapshot the Debug output of SPC Rule 2 (shift) detection for a known pattern.
    // Any change to SpecialCause enum variants or shift detection logic will fail this.
    // Mean = 4.5; last 9 values (5.0) all strictly above CL → Shift{Above, 9} detected.
    let values: Vec<f64> = vec![0.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0];
    let chart = make_chart_data(&values);
    let causes = check_western_electric_rules(&chart);
    let summary = format!("alert_count={}, causes={:?}", causes.len(), causes);
    insta::assert_debug_snapshot!("spc_shift_causes", summary);
}

#[test]
fn snapshot_orchestrator_telemetry_after_10_cycles() {
    // Snapshot RL orchestrator telemetry structure after 10 deterministic cycles.
    // Catches any change to telemetry fields, reward formula, or cycle_count tracking.
    let mut orch = wasm4pm::rl_orchestrator::RlOrchestrator::new_with_seed(42);
    let features = [0.2, 0.3, 0.2, 0.25, 1.0, 0.0, 0.5, 0.01];
    let healthy = make_test_state(0);
    let still_healthy = make_test_state(0);

    for _ in 0..10 {
        orch.run_cycle(&features, &healthy, &still_healthy, 0, true, true, false);
    }

    let t = orch.telemetry();
    // Snapshot stable fields; exclude last_reward (float) by rounding
    let summary = format!(
        "cycle_count={} agent={:?} last_spc={} consec_success={}",
        t.cycle_count, t.active_agent_name, t.last_spc_alert_count, t.consecutive_successes
    );
    insta::assert_debug_snapshot!("orchestrator_telemetry_10_cycles", summary);
}