Machine learning model optimization explainability

This invention describes a software system that helps people understand why an optimization model made certain decisions. When someone asks a question about a specific issue with the model's output, the system builds a map of how the model works, called a dependency graph. It then traces the relevant steps on this map and uses that information to create an explanation, often in plain language, detailing the decisions that led to the identified issue. The claims specifically detail that this map includes elements like the model's goals, variables, and constraints, and the explanation is generated by reviewing the traced path.