optimization

Reusable genetic and evolutionary optimization for FlossWare.

This repository provides optimization algorithms, beginning with a genetic algorithm implementation for evolving parameter sets. It is a reusable capability, not an agent runtime, task router, or orchestration layer.

Boundary

OptimizationStrategy contract
          |
          v
    optimization
          |
          v
candidate parameters -> fitness evaluation -> next generation

The algorithms can be used by Loom or other compatible systems without coupling them to Loom, a model provider, or a particular execution runtime.

Responsibility split

• optimization owns reusable genetic and evolutionary optimization algorithms.
• strategy owns reusable decision strategies such as Thompson Sampling and epsilon-greedy selection. Strategies may use optimization algorithms where appropriate, but the two responsibilities are not interchangeable.
• evaluation owns reusable assessment of results against intended outcomes.
• Loom owns execution semantics and orchestration, not optimization algorithms.

Optimization must operate within externally enforced security, authorization, safety, and budget constraints. An algorithm cannot make an otherwise forbidden operation permissible.

Scope

The public package contains reusable genetic optimization algorithms and the optimize decorator. Task classification and execution-blueprint generation are not part of this repository; callers should use Loom-domain contracts or a separately owned task-classification capability for those concerns.

Naming

The repository is named optimization; documentation and repository links use that name. The distribution is optimization and the canonical Python import package is optimization.