Model Interface & Signature Contract¶

Why model contracts matter¶

Serving systems require stable and explicit interfaces. Implicit assumptions lead to runtime failures.

Input contract¶

Each model defines: - input feature names, - data types, - required vs optional fields.

The input schema is enforced via: - MLflow model signature, - Pydantic schemas in the API.

Output contract¶

Outputs include: - prediction values, - optional confidence or probability scores, - metadata where applicable.

Versioning¶

A model contract is versioned together with: - the model artifact, - feature definitions, - training configuration.

Breaking changes require a new model version.