There is no lack of mediocre code. Advances in Artificial Intelligence give us a vast supply of it. To ensure that systems built from it improve rather than degrade, this work focusses on correctness and maintainability. Traditional wisdom says less code is more maintainable. Code becomes maintainable by removing unused code, generalizing functions to reduce code duplication, and occasionally even dropping features. Maintaining code is traditionally like maintaining a restaurant menu, less like maintaining a recipe book. We aim to change the traditional approach to maintainability by introducing a recipe book.

A recipe book can be maintained in the way that scientific knowledge is maintained: By adding to it. For example, new algorithms do not invalidate old algorithms. Similarly, new definitions do not replace old ones. We proceed with multiple, possibly interchangeable, definitions and algorithms.