Once information has been structured into a machine-readable form at the Fact stage, it becomes possible to analyze “what the information means.” In I2EA, this stage of analyzing meaning is defined as “Evaluation.”

Whereas Fact primarily addresses verifiable attributes associated with the information itself, Evaluation uses analysis by AI or other systems to assess the content and meaning of that information. For an image, for example, this may include analyzing whether it contains a person, whether it may contain a minor, whether it contains violent content, or whether it may contain elements protected by copyright.

The Evaluation generated at this stage is also recorded as structured tags through Umecohming. For example, results of AI analysis can be represented as “containsPerson: Yes,” “containsViolence: No,” or “containsCopyright: Possible,” making them available for reference by subsequent systems.

Evaluation, however, differs in nature from Fact. Unlike creation timestamps, identifiers, and other attributes recorded as Fact, semantic analysis by AI involves uncertainty. It is therefore important to preserve states such as “Unknown” and “Possible,” rather than relying solely on “Yes” and “No.” Where appropriate, Confidence can also be assigned to record the degree of certainty associated with an evaluation.

For example, an AI system may determine with high confidence that an image contains a person while being unable to determine from the image alone whether that person is a minor. In such a case, “containsPerson: Yes” may coexist with “containsMinor: Unknown.” There is no need to force a classification into Yes or No when sufficient information is unavailable.

Here again, Evaluation does not make the final governance decision. An evaluation indicating that an image “may contain elements protected by copyright” does not mean that use of the image is prohibited. Evaluation represents a semantic state that subsequent Policy and Decision processes can reference.

Accordingly, Evaluation is the stage in which information made machine-readable as Fact, together with its content, is analyzed and its semantic characteristics and uncertainties are structured as machine-readable evaluation information.

Whereas Fact describes “what is known about the information,” Evaluation describes “what is recognized when the information is analyzed.” Separating these two stages makes it possible to pass information to subsequent Policy processes without conflating observed information with AI-generated evaluation.