Use when tests should be generated from acceptance criteria, user stories or requirements with an LLM. Scores confidence per test case (self-assessment plus assertion, naming and completeness checks), batches uncertain cases for human review and emits tests in the team's existing framework.