← Glossary
Glossary · AI and agent workflows
Acceptance criteria
Statements, written before a workflow runs, of what its result must satisfy to be accepted, each checkable against something observable.
A criterion settles meanings the request left open, such as how a foreign currency is treated, and names a condition the result must meet. Completeness criteria are bounded to something checkable, such as a register of expected inputs plus the records a run finds. A completion message alone does not establish satisfaction: it is the action record, the output read by a person, arithmetic redone, or the state of a destination outside the harness. Criteria written after a result tend to describe that result, which is why they come first; criteria written first can still be missed, and make the miss visible.