feagi_npu_plasticity/

memory_neuron_array.rs

// Copyright 2025 Neuraville Inc.
// SPDX-License-Identifier: Apache-2.0

/*
 * Copyright 2025 Neuraville Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 */

//! Memory Neuron Array - Structure of Arrays for memory neurons with lifecycle management
//!
//! High-performance SoA implementation optimized for:
//! - SIMD-friendly vectorized operations
//! - Rust/RTOS compatibility
//! - Thread-safe operations
//! - Efficient memory management with index reuse

use crate::neuron_id_manager::{AllocationStats, NeuronIdManager};
use serde::{Deserialize, Serialize};
use std::collections::{HashMap, HashSet};

/// Memory neuron lifecycle configuration
#[derive(Debug, Clone, Copy)]
pub struct MemoryNeuronLifecycleConfig {
    /// Initial lifespan in bursts
    pub initial_lifespan: u32,

    /// Lifespan growth per reactivation
    pub lifespan_growth_rate: f32,

    /// Lifespan threshold for long-term memory conversion
    pub longterm_threshold: u32,

    /// Maximum reactivations before forced LTM
    pub max_reactivations: u32,
}

impl Default for MemoryNeuronLifecycleConfig {
    fn default() -> Self {
        Self {
            initial_lifespan: 20,
            lifespan_growth_rate: 3.0,
            longterm_threshold: 100,
            max_reactivations: 1000,
        }
    }
}

/// Snapshot of a single memory neuron for API / inspection (plasticity array only).
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
pub struct MemoryNeuronDetail {
    pub neuron_id: u32,
    pub cortical_area_idx: u32,
    pub pattern_hash: Option<u64>,
    pub is_longterm_memory: bool,
    pub is_active: bool,
    pub lifespan_current: u32,
    pub lifespan_initial: u32,
    pub lifespan_growth_rate: f32,
    pub creation_burst: u64,
    pub last_activation_burst: u64,
    pub activation_count: u32,
}

/// Memory neuron array statistics
#[derive(Debug, Clone, Default)]
pub struct MemoryNeuronStats {
    pub total_capacity: usize,
    pub active_neurons: usize,
    pub longterm_neurons: usize,
    pub dead_neurons: usize,
    pub reusable_indices: usize,
    pub memory_usage_bytes: usize,
    pub avg_lifespan: f64,
    pub avg_activation_count: f64,
}

/// High-performance Structure of Arrays for memory neurons
pub struct MemoryNeuronArray {
    capacity: usize,

    // Core neuron properties (SoA layout)
    neuron_ids: Vec<u32>,
    cortical_area_ids: Vec<u32>,
    is_active: Vec<bool>,

    // Lifecycle management
    lifespan_current: Vec<u32>,
    lifespan_initial: Vec<u32>,
    lifespan_growth_rate: Vec<f32>,
    is_longterm_memory: Vec<bool>,

    // Temporal tracking
    creation_burst: Vec<u64>,
    last_activation_burst: Vec<u64>,
    activation_count: Vec<u32>,

    // Pattern association (xxHash64 for fast, deterministic pattern identification)
    pattern_hash_to_index: HashMap<u64, usize>,
    index_to_pattern_hash: HashMap<usize, u64>,

    // Index management
    next_available_index: usize,
    reusable_indices: HashSet<usize>,

    // Area-specific tracking
    area_neuron_indices: HashMap<u32, HashSet<usize>>,

    // Neuron ID manager
    id_manager: NeuronIdManager,
}

impl MemoryNeuronArray {
    /// Create a new memory neuron array
    pub fn new(capacity: usize) -> Self {
        Self {
            capacity,
            neuron_ids: vec![0; capacity],
            cortical_area_ids: vec![0; capacity],
            is_active: vec![false; capacity],
            lifespan_current: vec![0; capacity],
            lifespan_initial: vec![0; capacity],
            lifespan_growth_rate: vec![0.0; capacity],
            is_longterm_memory: vec![false; capacity],
            creation_burst: vec![0; capacity],
            last_activation_burst: vec![0; capacity],
            activation_count: vec![0; capacity],
            pattern_hash_to_index: HashMap::new(),
            index_to_pattern_hash: HashMap::new(),
            next_available_index: 0,
            reusable_indices: HashSet::new(),
            area_neuron_indices: HashMap::new(),
            id_manager: NeuronIdManager::new(),
        }
    }

    /// Create a new memory neuron
    pub fn create_memory_neuron(
        &mut self,
        pattern_hash: u64,
        cortical_area_id: u32,
        current_burst: u64,
        config: &MemoryNeuronLifecycleConfig,
    ) -> Option<usize> {
        // Check if pattern already exists
        if let Some(&existing_idx) = self.pattern_hash_to_index.get(&pattern_hash) {
            if self.is_active[existing_idx] {
                // Reactivate existing neuron instead
                return self.reactivate_memory_neuron_internal(existing_idx, current_burst);
            }
        }

        // Get neuron index (reuse or allocate new)
        let neuron_idx = self.get_available_index_internal()?;

        // Allocate global neuron ID
        let neuron_id = self.id_manager.allocate_memory_neuron_id()?;

        // Initialize neuron properties
        self.neuron_ids[neuron_idx] = neuron_id;
        self.cortical_area_ids[neuron_idx] = cortical_area_id;
        self.is_active[neuron_idx] = true;

        // Initialize lifecycle
        self.lifespan_current[neuron_idx] = config.initial_lifespan;
        self.lifespan_initial[neuron_idx] = config.initial_lifespan;
        self.lifespan_growth_rate[neuron_idx] = config.lifespan_growth_rate;
        self.is_longterm_memory[neuron_idx] = false;

        // Initialize temporal tracking
        self.creation_burst[neuron_idx] = current_burst;
        self.last_activation_burst[neuron_idx] = current_burst;
        self.activation_count[neuron_idx] = 1;

        // Register pattern association
        self.pattern_hash_to_index.insert(pattern_hash, neuron_idx);
        self.index_to_pattern_hash.insert(neuron_idx, pattern_hash);

        // Add to area tracking
        self.area_neuron_indices
            .entry(cortical_area_id)
            .or_default()
            .insert(neuron_idx);

        Some(neuron_idx)
    }

    /// Reactivate an existing memory neuron
    pub fn reactivate_memory_neuron(&mut self, neuron_idx: usize, current_burst: u64) -> bool {
        self.reactivate_memory_neuron_internal(neuron_idx, current_burst)
            .is_some()
    }

    /// Internal reactivate that returns Option<usize> for compatibility
    fn reactivate_memory_neuron_internal(
        &mut self,
        neuron_idx: usize,
        current_burst: u64,
    ) -> Option<usize> {
        if !self.is_valid_index(neuron_idx) || !self.is_active[neuron_idx] {
            return None;
        }

        // Update activation tracking
        self.last_activation_burst[neuron_idx] = current_burst;
        self.activation_count[neuron_idx] += 1;

        // Grow lifespan if not long-term memory
        if !self.is_longterm_memory[neuron_idx] {
            let current_lifespan = self.lifespan_current[neuron_idx];
            let growth = self.lifespan_growth_rate[neuron_idx] as u32;
            self.lifespan_current[neuron_idx] = current_lifespan.saturating_add(growth);
        }

        Some(neuron_idx)
    }

    /// Age all active memory neurons (vectorized operation)
    pub fn age_memory_neurons(&mut self, _current_burst: u64) -> Vec<usize> {
        let n = self.next_available_index;
        if n == 0 {
            return Vec::new();
        }

        let mut died_indices = Vec::new();

        // Age eligible neurons
        for i in 0..n {
            if self.is_active[i] && !self.is_longterm_memory[i] && self.lifespan_current[i] > 0 {
                self.lifespan_current[i] -= 1;

                // Check if neuron died
                if self.lifespan_current[i] == 0 {
                    self.is_active[i] = false;
                    died_indices.push(i);
                }
            }
        }

        // Clean up dead neurons after iteration
        for &i in &died_indices {
            self.cleanup_dead_neuron_internal(i);
        }

        died_indices
    }

    /// Check for neurons ready for long-term memory conversion
    pub fn check_longterm_conversion(&mut self, longterm_threshold: u32) -> Vec<usize> {
        let n = self.next_available_index;
        if n == 0 {
            return Vec::new();
        }

        let mut converted_indices = Vec::new();

        for i in 0..n {
            if self.is_active[i]
                && !self.is_longterm_memory[i]
                && self.lifespan_current[i] >= longterm_threshold
            {
                self.is_longterm_memory[i] = true;
                converted_indices.push(i);
            }
        }

        converted_indices
    }

    /// Check for neurons ready for long-term conversion using per-area thresholds.
    ///
    /// Areas without an explicit lifecycle config use `default_threshold`.
    pub fn check_longterm_conversion_by_area(
        &mut self,
        lifecycle_configs: &HashMap<u32, MemoryNeuronLifecycleConfig>,
        default_threshold: u32,
    ) -> Vec<usize> {
        let n = self.next_available_index;
        if n == 0 {
            return Vec::new();
        }

        let area_indices: Vec<(u32, Vec<usize>)> = self
            .area_neuron_indices
            .iter()
            .map(|(&area, indices)| (area, indices.iter().copied().collect()))
            .collect();

        let mut converted_indices = Vec::new();
        for (area_id, indices) in area_indices {
            let threshold = lifecycle_configs
                .get(&area_id)
                .map(|config| config.longterm_threshold)
                .unwrap_or(default_threshold);

            for neuron_idx in indices {
                if self.is_active[neuron_idx]
                    && !self.is_longterm_memory[neuron_idx]
                    && self.lifespan_current[neuron_idx] >= threshold
                {
                    self.is_longterm_memory[neuron_idx] = true;
                    converted_indices.push(neuron_idx);
                }
            }
        }

        converted_indices
    }

    /// Get all active neuron IDs for a cortical area
    pub fn get_active_neurons_by_area(&self, cortical_area_id: u32) -> Vec<u32> {
        if let Some(indices) = self.area_neuron_indices.get(&cortical_area_id) {
            indices
                .iter()
                .filter(|&&idx| self.is_valid_index(idx) && self.is_active[idx])
                .map(|&idx| self.neuron_ids[idx])
                .collect()
        } else {
            Vec::new()
        }
    }

    /// Active short-term neurons (not marked LTM) in the area.
    pub fn count_short_term_in_area(&self, cortical_area_id: u32) -> usize {
        self.area_neuron_indices
            .get(&cortical_area_id)
            .map(|indices| {
                indices
                    .iter()
                    .filter(|&&idx| {
                        self.is_valid_index(idx)
                            && self.is_active[idx]
                            && !self.is_longterm_memory[idx]
                    })
                    .count()
            })
            .unwrap_or(0)
    }

    /// Active long-term memory neurons in the area.
    pub fn count_long_term_in_area(&self, cortical_area_id: u32) -> usize {
        self.area_neuron_indices
            .get(&cortical_area_id)
            .map(|indices| {
                indices
                    .iter()
                    .filter(|&&idx| {
                        self.is_valid_index(idx)
                            && self.is_active[idx]
                            && self.is_longterm_memory[idx]
                    })
                    .count()
            })
            .unwrap_or(0)
    }

    /// Stable-sorted active neuron IDs for the area, with pagination `(page_ids, total_count)`.
    pub fn paginated_neuron_ids_in_area(
        &self,
        cortical_area_id: u32,
        offset: usize,
        limit: usize,
    ) -> (Vec<u32>, usize) {
        let mut ids = self.get_active_neurons_by_area(cortical_area_id);
        ids.sort_unstable();
        let total = ids.len();
        let page = ids.into_iter().skip(offset).take(limit).collect();
        (page, total)
    }

    fn find_neuron_index_by_global_id(&self, neuron_id: u32) -> Option<usize> {
        (0..self.next_available_index).find(|&i| self.neuron_ids[i] == neuron_id)
    }

    /// Full detail for an active memory neuron by global neuron id, if present.
    pub fn get_memory_neuron_detail(&self, neuron_id: u32) -> Option<MemoryNeuronDetail> {
        let idx = self.find_neuron_index_by_global_id(neuron_id)?;
        if !self.is_valid_index(idx) || !self.is_active[idx] {
            return None;
        }
        Some(MemoryNeuronDetail {
            neuron_id: self.neuron_ids[idx],
            cortical_area_idx: self.cortical_area_ids[idx],
            pattern_hash: self.index_to_pattern_hash.get(&idx).copied(),
            is_longterm_memory: self.is_longterm_memory[idx],
            is_active: self.is_active[idx],
            lifespan_current: self.lifespan_current[idx],
            lifespan_initial: self.lifespan_initial[idx],
            lifespan_growth_rate: self.lifespan_growth_rate[idx],
            creation_burst: self.creation_burst[idx],
            last_activation_burst: self.last_activation_burst[idx],
            activation_count: self.activation_count[idx],
        })
    }

    /// Find neuron index by pattern hash
    pub fn find_neuron_by_pattern(&self, pattern_hash: &u64) -> Option<usize> {
        self.pattern_hash_to_index
            .get(pattern_hash)
            .copied()
            .filter(|&idx| self.is_valid_index(idx) && self.is_active[idx])
    }

    /// Get neuron ID at index
    pub fn get_neuron_id(&self, neuron_idx: usize) -> Option<u32> {
        if self.is_valid_index(neuron_idx) {
            Some(self.neuron_ids[neuron_idx])
        } else {
            None
        }
    }

    /// Get cortical area ID at index
    pub fn get_cortical_area_id(&self, neuron_idx: usize) -> Option<u32> {
        if self.is_valid_index(neuron_idx) {
            Some(self.cortical_area_ids[neuron_idx])
        } else {
            None
        }
    }

    pub fn get_pattern_hash(&self, neuron_idx: usize) -> Option<u64> {
        self.index_to_pattern_hash.get(&neuron_idx).copied()
    }

    /// Get comprehensive statistics
    pub fn get_stats(&self) -> MemoryNeuronStats {
        let n = self.next_available_index;

        if n == 0 {
            return MemoryNeuronStats {
                total_capacity: self.capacity,
                ..Default::default()
            };
        }

        let active_count = self.is_active[..n].iter().filter(|&&x| x).count();
        let longterm_count = (0..n)
            .filter(|&i| self.is_active[i] && self.is_longterm_memory[i])
            .count();
        let dead_count = n - active_count;

        // Calculate averages for active neurons
        let (avg_lifespan, avg_activation_count) = if active_count > 0 {
            let total_lifespan: u32 = (0..n)
                .filter(|&i| self.is_active[i])
                .map(|i| self.lifespan_current[i])
                .sum();
            let total_activations: u32 = (0..n)
                .filter(|&i| self.is_active[i])
                .map(|i| self.activation_count[i])
                .sum();

            (
                total_lifespan as f64 / active_count as f64,
                total_activations as f64 / active_count as f64,
            )
        } else {
            (0.0, 0.0)
        };

        // Estimate memory usage
        let memory_usage = self.capacity * (
            std::mem::size_of::<u32>() * 4 +  // uint32 arrays
            std::mem::size_of::<f32>() +       // float32 array
            std::mem::size_of::<u64>() * 2 +   // uint64 arrays
            std::mem::size_of::<bool>() * 2    // bool arrays
        ) + self.pattern_hash_to_index.len() * (8 + 8)  // Pattern hash mappings (u64 + usize)
          + self.area_neuron_indices.len() * 64; // Area tracking overhead

        MemoryNeuronStats {
            total_capacity: self.capacity,
            active_neurons: active_count,
            longterm_neurons: longterm_count,
            dead_neurons: dead_count,
            reusable_indices: self.reusable_indices.len(),
            memory_usage_bytes: memory_usage,
            avg_lifespan,
            avg_activation_count,
        }
    }

    /// Get ID manager allocation statistics
    pub fn get_id_allocation_stats(&self) -> AllocationStats {
        self.id_manager.get_allocation_stats()
    }

    /// Get available index (reuse or allocate new)
    fn get_available_index_internal(&mut self) -> Option<usize> {
        // Try to reuse a dead neuron index first
        if let Some(&idx) = self.reusable_indices.iter().next() {
            self.reusable_indices.remove(&idx);
            return Some(idx);
        }

        // Allocate new index if capacity allows
        if self.next_available_index < self.capacity {
            let idx = self.next_available_index;
            self.next_available_index += 1;
            Some(idx)
        } else {
            None
        }
    }

    /// Clean up associations for a dead neuron
    fn cleanup_dead_neuron_internal(&mut self, neuron_idx: usize) {
        // Deallocate global neuron ID
        let neuron_id = self.neuron_ids[neuron_idx];
        self.id_manager.deallocate_memory_neuron_id(neuron_id);

        // Remove pattern association
        if let Some(pattern_hash) = self.index_to_pattern_hash.remove(&neuron_idx) {
            self.pattern_hash_to_index.remove(&pattern_hash);
        }

        // Remove from area tracking
        let area_id = self.cortical_area_ids[neuron_idx];
        if let Some(indices) = self.area_neuron_indices.get_mut(&area_id) {
            indices.remove(&neuron_idx);
        }

        // Add to reusable indices
        self.reusable_indices.insert(neuron_idx);
    }

    /// Check if neuron index is valid
    fn is_valid_index(&self, neuron_idx: usize) -> bool {
        neuron_idx < self.next_available_index
    }

    /// Reset array state (for testing)
    /// Deactivate all memory neurons in a specific cortical area
    pub fn reset_cortical_area(&mut self, cortical_area_id: u32) -> usize {
        let Some(neuron_indices) = self.area_neuron_indices.get(&cortical_area_id) else {
            return 0;
        };

        let indices_to_reset: Vec<usize> = neuron_indices.iter().copied().collect();
        let mut reset_count = 0;

        for neuron_idx in indices_to_reset {
            if !self.is_valid_index(neuron_idx) {
                continue;
            }

            // Deactivate neuron
            self.is_active[neuron_idx] = false;

            // Remove pattern mapping
            if let Some(pattern_hash) = self.index_to_pattern_hash.remove(&neuron_idx) {
                self.pattern_hash_to_index.remove(&pattern_hash);
            }

            // Clear properties (optional but clean)
            self.lifespan_current[neuron_idx] = 0;
            self.activation_count[neuron_idx] = 0;

            // Mark as reusable
            self.reusable_indices.insert(neuron_idx);

            reset_count += 1;
        }

        // Remove area tracking
        self.area_neuron_indices.remove(&cortical_area_id);

        reset_count
    }

    pub fn reset(&mut self) {
        self.neuron_ids.fill(0);
        self.cortical_area_ids.fill(0);
        self.is_active.fill(false);
        self.lifespan_current.fill(0);
        self.lifespan_initial.fill(0);
        self.lifespan_growth_rate.fill(0.0);
        self.is_longterm_memory.fill(false);
        self.creation_burst.fill(0);
        self.last_activation_burst.fill(0);
        self.activation_count.fill(0);

        self.pattern_hash_to_index.clear();
        self.index_to_pattern_hash.clear();
        self.area_neuron_indices.clear();

        self.next_available_index = 0;
        self.reusable_indices.clear();

        self.id_manager.reset();
    }
}

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

    #[test]
    fn test_lifecycle_config_default() {
        let config = MemoryNeuronLifecycleConfig::default();
        assert_eq!(config.initial_lifespan, 20);
        assert_eq!(config.lifespan_growth_rate, 3.0);
        assert_eq!(config.longterm_threshold, 100);
        assert_eq!(config.max_reactivations, 1000);
    }

    #[test]
    fn test_create_memory_neuron() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        let pattern_hash = 0x0101010101010101u64;
        let neuron_idx = array.create_memory_neuron(pattern_hash, 100, 0, &config);

        assert!(neuron_idx.is_some());
        let idx = neuron_idx.unwrap();
        assert!(array.is_active[idx]);
        assert_eq!(array.cortical_area_ids[idx], 100);
        assert_eq!(array.lifespan_current[idx], config.initial_lifespan);
        assert_eq!(array.activation_count[idx], 1);
        assert_eq!(array.creation_burst[idx], 0);
    }

    #[test]
    fn test_create_duplicate_pattern() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        let pattern_hash = 0x0101010101010101u64;
        let idx1 = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        // Creating with same pattern should reactivate existing neuron
        let idx2 = array
            .create_memory_neuron(pattern_hash, 100, 1, &config)
            .unwrap();

        assert_eq!(idx1, idx2);
        assert_eq!(array.activation_count[idx1], 2); // Should have been reactivated
    }

    #[test]
    fn test_multiple_neurons() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        let mut neurons = Vec::new();
        for i in 0..10 {
            let pattern_hash = i as u64;
            let idx = array
                .create_memory_neuron(pattern_hash, 100, 0, &config)
                .unwrap();
            neurons.push(idx);
        }

        assert_eq!(neurons.len(), 10);
        assert_eq!(array.next_available_index, 10);

        let stats = array.get_stats();
        assert_eq!(stats.active_neurons, 10);
    }

    #[test]
    fn test_reactivate_memory_neuron() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        let initial_count = array.activation_count[idx];
        let initial_lifespan = array.lifespan_current[idx];

        assert!(array.reactivate_memory_neuron(idx, 1));

        assert_eq!(array.activation_count[idx], initial_count + 1);
        assert_eq!(array.last_activation_burst[idx], 1);

        // Lifespan should have grown
        let expected_lifespan = initial_lifespan + config.lifespan_growth_rate as u32;
        assert_eq!(array.lifespan_current[idx], expected_lifespan);
    }

    #[test]
    fn test_reactivate_invalid_neuron() {
        let mut array = MemoryNeuronArray::new(1000);

        // Try to reactivate non-existent neuron
        assert!(!array.reactivate_memory_neuron(0, 1));
        assert!(!array.reactivate_memory_neuron(999, 1));
    }

    #[test]
    fn test_age_memory_neurons() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 2,
            ..Default::default()
        };

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        // Age once
        let died = array.age_memory_neurons(1);
        assert!(died.is_empty());
        assert_eq!(array.lifespan_current[idx], 1);

        // Age again - should die
        let died = array.age_memory_neurons(2);
        assert_eq!(died.len(), 1);
        assert_eq!(died[0], idx);
        assert!(!array.is_active[idx]);

        // Pattern should no longer be findable
        let found = array.find_neuron_by_pattern(&pattern_hash);
        assert!(found.is_none());
    }

    #[test]
    fn test_age_multiple_neurons() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 5,
            ..Default::default()
        };

        let mut neurons = Vec::new();
        for i in 0..10 {
            let pattern_hash = i as u64;
            let idx = array
                .create_memory_neuron(pattern_hash, 100, 0, &config)
                .unwrap();
            neurons.push(idx);
        }

        // Age 5 times - all should die
        for burst in 1..=5 {
            let died = array.age_memory_neurons(burst);
            if burst < 5 {
                assert_eq!(died.len(), 0);
            } else {
                assert_eq!(died.len(), 10);
            }
        }

        let stats = array.get_stats();
        assert_eq!(stats.active_neurons, 0);
        assert_eq!(stats.dead_neurons, 10);
    }

    #[test]
    fn test_longterm_memory_no_aging() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 100,
            ..Default::default()
        };

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        // Convert to long-term memory
        let converted = array.check_longterm_conversion(100);
        assert_eq!(converted.len(), 1);
        assert!(array.is_longterm_memory[idx]);

        let initial_lifespan = array.lifespan_current[idx];

        // Age many times - should not affect long-term memory
        for burst in 1..=50 {
            array.age_memory_neurons(burst);
        }

        assert!(array.is_active[idx]);
        assert_eq!(array.lifespan_current[idx], initial_lifespan); // Should not change
    }

    #[test]
    fn test_longterm_conversion() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 100,
            ..Default::default()
        };

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        let converted = array.check_longterm_conversion(100);
        assert_eq!(converted.len(), 1);
        assert_eq!(converted[0], idx);
        assert!(array.is_longterm_memory[idx]);

        // Second check should not convert again
        let converted2 = array.check_longterm_conversion(100);
        assert_eq!(converted2.len(), 0);
    }

    #[test]
    fn test_longterm_conversion_threshold() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 50,
            ..Default::default()
        };

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        // Should not convert below threshold
        let converted = array.check_longterm_conversion(100);
        assert_eq!(converted.len(), 0);
        assert!(!array.is_longterm_memory[idx]);

        // Grow lifespan through reactivations
        for burst in 1..=20 {
            array.reactivate_memory_neuron(idx, burst);
        }

        // Now should convert
        let converted = array.check_longterm_conversion(100);
        assert_eq!(converted.len(), 1);
        assert!(array.is_longterm_memory[idx]);
    }

    #[test]
    fn test_find_neuron_by_pattern() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        let found = array.find_neuron_by_pattern(&pattern_hash);
        assert_eq!(found, Some(idx));

        // Different pattern should not be found
        let pattern_hash2 = 0x0202020202020202u64;
        let found2 = array.find_neuron_by_pattern(&pattern_hash2);
        assert_eq!(found2, None);
    }

    #[test]
    fn test_get_active_neurons_by_area() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        // Create neurons in different areas
        for area in [100, 200] {
            for i in 0..5 {
                let pattern_hash = ((area as u64) << 32) | (i as u64);
                array.create_memory_neuron(pattern_hash, area, 0, &config);
            }
        }

        let area100_neurons = array.get_active_neurons_by_area(100);
        let area200_neurons = array.get_active_neurons_by_area(200);

        assert_eq!(area100_neurons.len(), 5);
        assert_eq!(area200_neurons.len(), 5);

        // Non-existent area
        let area999_neurons = array.get_active_neurons_by_area(999);
        assert_eq!(area999_neurons.len(), 0);
    }

    #[test]
    fn test_get_neuron_id() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        let neuron_id = array.get_neuron_id(idx);
        assert!(neuron_id.is_some());

        // Invalid index
        let invalid_id = array.get_neuron_id(999);
        assert!(invalid_id.is_none());
    }

    #[test]
    fn test_index_reuse() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 1,
            ..Default::default()
        };

        let pattern_hash = 0x0101010101010101u64;
        let idx1 = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();
        assert_eq!(idx1, 0);

        // Let it die
        array.age_memory_neurons(1);
        assert!(!array.is_active[idx1]);

        // Create new neuron - should reuse index
        let pattern_hash2 = 0x0202020202020202u64;
        let idx2 = array
            .create_memory_neuron(pattern_hash2, 100, 2, &config)
            .unwrap();
        assert_eq!(idx2, 0); // Should reuse index 0
        assert_eq!(array.next_available_index, 1); // Should not have advanced
    }

    #[test]
    fn test_get_stats() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        // Create some neurons
        for i in 0..10 {
            let pattern_hash = i as u64;
            array.create_memory_neuron(pattern_hash, 100, 0, &config);
        }

        let stats = array.get_stats();
        assert_eq!(stats.total_capacity, 1000);
        assert_eq!(stats.active_neurons, 10);
        assert_eq!(stats.longterm_neurons, 0);
        assert_eq!(stats.dead_neurons, 0);
        assert!(stats.avg_lifespan > 0.0);
        assert!(stats.avg_activation_count >= 1.0);
        assert!(stats.memory_usage_bytes > 0);
    }

    #[test]
    fn test_capacity_exhaustion() {
        let mut array = MemoryNeuronArray::new(5);
        let config = MemoryNeuronLifecycleConfig::default();

        // Create neurons up to capacity
        for i in 0..5 {
            let pattern_hash = i as u64;
            let idx = array.create_memory_neuron(pattern_hash, 100, 0, &config);
            assert!(idx.is_some());
        }

        // Try to create beyond capacity
        let pattern_hash = 0x6363636363636363u64;
        let idx = array.create_memory_neuron(pattern_hash, 100, 0, &config);
        assert!(idx.is_none());
    }

    #[test]
    fn test_reset_cortical_area() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        // Create neurons in area 5
        let pattern1 = 0x0101010101010101u64;
        let pattern2 = 0x0202020202020202u64;
        array.create_memory_neuron(pattern1, 5, 0, &config);
        array.create_memory_neuron(pattern2, 5, 0, &config);

        // Create neurons in area 6
        let pattern3 = 0x0303030303030303u64;
        array.create_memory_neuron(pattern3, 6, 0, &config);

        // Verify initial state
        assert_eq!(array.get_active_neurons_by_area(5).len(), 2);
        assert_eq!(array.get_active_neurons_by_area(6).len(), 1);

        // Reset area 5
        let reset_count = array.reset_cortical_area(5);
        assert_eq!(reset_count, 2);

        // Verify area 5 is empty
        assert_eq!(array.get_active_neurons_by_area(5).len(), 0);

        // Verify area 6 is unchanged
        assert_eq!(array.get_active_neurons_by_area(6).len(), 1);

        // Verify pattern hashes are cleared for area 5
        assert!(!array.pattern_hash_to_index.contains_key(&pattern1));
        assert!(!array.pattern_hash_to_index.contains_key(&pattern2));

        // Verify area 6 pattern still exists
        assert!(array.pattern_hash_to_index.contains_key(&pattern3));
    }

    #[test]
    fn test_reset() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        // Create some neurons
        for i in 0..5 {
            let pattern_hash = i as u64;
            array.create_memory_neuron(pattern_hash, 100, 0, &config);
        }

        assert_eq!(array.next_available_index, 5);

        array.reset();

        assert_eq!(array.next_available_index, 0);
        let stats = array.get_stats();
        assert_eq!(stats.active_neurons, 0);
    }

    #[test]
    fn test_lifespan_growth_on_reactivation() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig {
            initial_lifespan: 10,
            lifespan_growth_rate: 5.0,
            longterm_threshold: 100,
            max_reactivations: 1000,
        };

        let pattern_hash = 0x0101010101010101u64;
        let idx = array
            .create_memory_neuron(pattern_hash, 100, 0, &config)
            .unwrap();

        assert_eq!(array.lifespan_current[idx], 10);

        array.reactivate_memory_neuron(idx, 1);
        assert_eq!(array.lifespan_current[idx], 15);

        array.reactivate_memory_neuron(idx, 2);
        assert_eq!(array.lifespan_current[idx], 20);
    }

    #[test]
    fn test_st_ltm_counts_and_pagination_and_detail() {
        let mut array = MemoryNeuronArray::new(1000);
        let config = MemoryNeuronLifecycleConfig::default();

        for i in 0..5 {
            array
                .create_memory_neuron(i as u64, 7, 0, &config)
                .expect("create");
        }
        assert_eq!(array.count_short_term_in_area(7), 5);
        assert_eq!(array.count_long_term_in_area(7), 0);

        array.is_longterm_memory[0] = true;
        assert_eq!(array.count_short_term_in_area(7), 4);
        assert_eq!(array.count_long_term_in_area(7), 1);

        let (page0, total) = array.paginated_neuron_ids_in_area(7, 0, 2);
        assert_eq!(total, 5);
        assert_eq!(page0.len(), 2);

        let nid = page0[0];
        let detail = array.get_memory_neuron_detail(nid).expect("detail");
        assert_eq!(detail.neuron_id, nid);
        assert_eq!(detail.cortical_area_idx, 7);
        assert!(detail.pattern_hash.is_some());
    }
}