To evaluate institutional requirements through AI Vigilance, Policy Objectives expressed in laws, regulations, standards, and organizational policies must be transformed into Operational and Evaluable Criteria. Institutional requirements are generally written for human interpretation and implementation and therefore cannot always be applied directly as Evaluation Criteria for individual AI Systems or operational environments.

The Translation Hub transforms Policy Objectives into Machine-Readable Policy and connects them with Contextual Structures such as Authority, Scope, Conditions, and Exceptions. This converts Institutional Requirements into structures that Governance Functions can reference. However, Translation alone does not determine all Evaluation Criteria required for a specific Operational Environment.

In a small or relatively simple Organization, translated Requirements may be applied relatively directly to AI Vigilance. In larger Organizations, specialized Domains, or multi-Region Environments, Context Design and Local / Domain Contextualization are required. Even under the same Policy Objective, specific Evaluation Criteria relating to Risk, required Evidence, and Thresholds may differ according to Domain, such as Healthcare, Finance, or Education, as well as Country / Region, Organization Structure, User Population, and Use Purpose.

However, Contextualization ≠ Modification of the Policy Objective. Adapting Evaluation Criteria to a particular Context must remain distinct from changing the underlying Institutional Requirement itself. Without a mechanism for verifying this boundary, Institutional Requirements could be unintentionally weakened or transformed into a different meaning during Operationalization.

AI Vigilance therefore performs two distinct forms of Evaluation. Alignment Evaluation assesses whether Evaluation Criteria produced through Contextualization remain aligned with the original Policy Objective. In other words, it verifies whether the Institutional Requirement has been appropriately contextualized.

Conformance Evaluation, by contrast, assesses whether actual AI Behavior conforms to the Evaluation Criteria established for that Context. It verifies whether the AI operates in accordance with the contextualized requirements.

Accordingly:

Policy Objective → Translation → Contextualization → Evaluation Criteria → AI Behavior

Within this flow, Alignment Evaluation verifies the relationship between the Policy Objective and Evaluation Criteria, while Conformance Evaluation verifies the relationship between the Evaluation Criteria and AI Behavior.

This separation makes it possible to distinguish between a situation in which AI Behavior conforms to Criteria that have themselves diverged from the Policy Objective, and one in which the Criteria are appropriate but AI Behavior fails to conform.

Context and Evaluation Criteria may legitimately differ across Regions, Industries, and Organizations. What should be standardized is not the Context itself, but the structure for verifying Alignment with the Policy Objective and evaluating the Conformance of AI Behavior. This enables a common Verifiable Evaluation Structure for AI Vigilance while preserving diverse Institutional Environments.