Explainability

Explainability is the ability to say, in the terms of the underlying litigation or claim, why a model's output moved. It depends on inputs that are inspectable and a method that can be traced from a feature to a result.

Underwriters and regulators ask why a number is the number. A model built on structured features such as forum, case type, procedural history, party configuration and timing can answer in those terms, and a reviewer can check the answer against the record.

Explainability is a design choice made early. Methods whose internals cannot be traced, or that rely on text embeddings of what documents happen to say, make the question harder to answer and are harder for a model-risk team to approve.

Explainability is not a narrative written after the fact to justify an output. It is a property of the method that lets the reasons for a result be reconstructed.