Legal AI discourse has concentrated on answer correctness, post-hoc explainability, and output verification. This note argues that those efforts, while necessary, address an incomplete layer of the governance problem. Reliable legal AI requires three elements that current frameworks treat as one: legitimacy formation, the processes through which legal interpretations acquire force over time; legitimacy execution, the governance question of whether the runtime conditions for admissible commitment have been satisfied; and runtime admissibility, the transition mechanism between the two, which determines whether an interpretation is admissible for reliance in a specific execution context before action is taken. The note draws on the CriptoIus/JurisRank research architecture for legitimacy formation, engages with the VEIP framework for legitimacy execution, and proposes runtime admissibility as an explicit, inspectable, and computable primitive currently absent from legal AI governance. The central claim is that future governance architectures will be judged less by the quality of their answers and more by the quality of their transitions.
Ignacio Adrián LERER (Mon,) studied this question.