Defines the core trait system for VRP features: FeatureConstraint, FeatureObjective, FeatureState, and FeatureBuilder. This pattern allows composing arbitrary constraints and objectives into a GoalContext, enabling extensible VRP variant support. The design is clean, generic, and reusable for any combinatorial optimization problem requiring pluggable constraints and objectives.

Implements a generic multi-armed bandit using Thompson sampling with Normal-Inverse-Gamma conjugate prior for non-stationary rewards. Used for dynamic operator selection in the hyper-heuristic. The implementation is self-contained, well-documented, and can be extracted for any RL-based selection problem.

Implements a Growing Self-Organizing Map (GSOM) network with node insertion, weight adjustment, error distribution, and min-max normalization. Used in the Rosomaxa population algorithm to maintain diversity. The implementation is isolated, generic, and reusable for any unsupervised clustering or dimensionality reduction task.

Provides macros like custom_dimension!, custom_solution_state!, custom_tour_state!, and custom_activity_state! that generate type-safe accessor functions for storing arbitrary metadata on domain objects (jobs, routes, solutions). This pattern eliminates string-keyed maps and provides compile-time safety, reusable in any Rust project needing extensible property storage.

Provides parallel iteration utilities (map_reduce, parallel_into_collect, parallel_collect, parallel_foreach_mut) with a WASM-compatible fallback using single-threaded execution. The abstraction is clean, reusable, and solves the common problem of writing parallel code that works both on native and WebAssembly targets.