LERA defines risk by the consequence of execution, not merely by how intelligent the system is. L0 — Ordinary Judgment Contexts: Everyday decision contexts where outcomes are reversible, errors are limited or recoverable, and responsibility remains local or informal.
At this level, judgment mainly functions as learning or ordinary decision-making, not as a governance constraint. L1 — Assistive Judgment Contexts: Systems assist human judgment and influence operational outcomes, but consequences remain bounded and correctable.
Judgment begins to shape system behavior, but responsibility is still relatively loose. L2 — Structured Judgment Contexts: Judgment begins to determine whether execution is allowed.
Human responsibility must be explicitly anchored, and actions may carry significant cost, disruption, or institutional consequence. L3 — Irreversible Judgment Contexts: Actions are irreversible, failure cannot be tolerated, or consequences extend beyond local correction.
Here, judgment is not optional. It functions as a hard gate, not an optimization layer.
The higher the consequence of execution, the stronger the judgment and governance required before action proceeds.
LERA risk levels are execution-consequence categories, not model-size categories.