-spec add_global_bonus(float(), map()) -> float().

Add global bonus to a silo's reward based on global health improvement.

Each silo gets +0.1 * (current_global - previous_global) as bonus. This encourages silos to cooperate for overall system improvement.

-spec compute_distribution_reward(map()) -> float().

Compute reward signal for Distribution Silo L0.

Input metrics map should contain: - load_balance_score: How evenly distributed load is (0-1) - migration_effectiveness: Do migrations improve diversity? (0-1) - network_efficiency: Low overhead, successful transfers (0-1) - resource_pressure_caused: Pressure on Resource Silo (0-1) - task_diversity_contribution: Helping Task Silo diversity (0-1)

-spec compute_global_health(map()) -> float().

Compute global health metric from all silo rewards.

Global health is a weighted average of silo rewards. This provides a holistic view of system performance.

-spec compute_resource_reward(map()) -> float().

Compute reward signal for Resource Silo L0.

Input metrics map should contain: - throughput: Evaluations per second (normalized 0-1) - gc_triggers: Number of GC triggers in period (0 = best) - pauses: Number of evolution pauses (0 = best) - memory_utilization: Current memory usage (0-1) - cpu_utilization: Current CPU usage (0-1) - task_silo_blocked: Was Task Silo blocked? (0-1) - distribution_blocked: Was Distribution blocked? (0-1)

-spec compute_task_reward(map()) -> float().

Compute reward signal for Task Silo L0 TWEANN.

Designed to reward LC controllers that: 1. MAXIMIZE fitness improvement velocity (faster learning) 2. MINIMIZE training time to convergence (efficiency) 3. PUNISH premature convergence (getting stuck at low fitness) 4. MINIMIZE resource usage (memory/CPU efficiency)

Input metrics map should contain: - improvement_velocity: Fitness improvement rate per 1000 evals (-1 to 1) - best_fitness: Current best fitness achieved (0-1) - fitness_target: Target fitness to reach (0-1) - evaluations_used: Total evaluations consumed - evaluations_budget: Total evaluation budget - stagnation_severity: How stuck evolution is (0=progressing, 1=stuck) - diversity_index: Population diversity (0-1) - memory_pressure: Current memory usage (0-1) - cpu_pressure: Current CPU usage (0-1)

-spec compute_task_reward_breakdown(map()) -> map().

Compute reward breakdown for Task Silo L0 - returns individual component scores.

Returns a map containing: - total_reward: The combined weighted reward (-1.0 to 1.0) - velocity_score: Fitness improvement velocity score (40% weight) - efficiency_score: Training time efficiency score (20% weight) - convergence_score: Premature convergence penalty score (20% weight) - resource_score: Memory/CPU efficiency score (20% weight) - velocity_weight, efficiency_weight, etc.: The weights used

This allows visualization of which factors are contributing to the reward.

-spec compute_trial_reward(map()) -> float().

Compute cumulative reward for an LC TWEANN trial.

Called at the end of a trial period to evaluate the LC controller's performance. Uses fitness improvement achieved during the trial as the primary metric.

Input metrics map should contain: - start_fitness: Fitness at trial start - end_fitness: Fitness at trial end - evaluations_used: Evaluations consumed during trial - evaluation_budget: Evaluation budget for trial - peak_memory: Peak memory usage during trial (0-1) - avg_cpu: Average CPU usage during trial (0-1) - stagnation_events: Number of stagnation events during trial - convergence_reached: true if fitness target was reached