Engineering Principles

1. Make invalid states unrepresentable. Use immutable validated domain objects at component boundaries.
2. Prefer polymorphism to conditionals. New models, schedules, controllers, solvers, and decoders implement contracts rather than modifying the core loop.
3. Fail loudly and locally. Raise typed exceptions at the boundary where a bad value is introduced.
4. Keep state ownership explicit. No hidden RNG state and no mutable global extension registry.
5. Use Google Python style. Public APIs have complete annotations and Google-style docstrings. Lines are limited to 100 characters.
6. Avoid semi-private APIs. Do not use underscore-prefixed members as a convention. Public extension points are documented; implementation details are isolated in private modules or concrete classes.
7. Test behavior, not coverage numbers. Contract, property, numerical, integration, and failure-path tests are required alongside coverage.
8. Be honest about research status. Educational approximations and claims of paper fidelity must be labeled clearly.