The model declined the claim, and it was the right call. Then a customer asked why, and the honest answer was that nobody had written down what gave the system the authority to decide that in the first place.
An architectural discipline for consequential decision systems. When an organisation builds AI into a consequential product, the hard part is not getting an answer. It is being able to stand behind it.
A correct answer is not the same as the right to give it. Authority has to be derived, not declared.
When someone asks why the system decided that, you can produce the structure rather than an assurance.
Legitimacy should not depend on everyone remembering the rules.
Use it when: When AI is placed inside a decision that someone has to answer for afterwards.