Glossary

Agent adapter — the integration layer that invokes a customer-owned agent and returns normalised evidence.

Acceptance criterion — a business-authored expectation used to judge whether the agent behaved correctly.

Business journey — a business-owned file containing customer turns and acceptance criteria.

Canonical evidence — the normalised transcript, tool calls, tool results, final output and evaluator results captured from an agent run.

Fixture — optional synthetic test data. Useful for demos and deterministic mocked tests, but not required from business users.

Rubric — generated criteria used by the acceptance evaluator.

Trajectory — optional engineering-owned checks for tool calls, tool order, arguments or backend evidence.

Workflow — executable conversation flow generated from business turns.