Fact is not generated through Evaluation. A Fact Tag represents Externally Grounded information obtained from Reality or Observed Evidence that becomes the subject of Evaluation. An AI System may interpret Facts, relate them to other information, and use them in Evaluation, but it must not rewrite the Facts themselves to conform to an Evaluation outcome. This separation is essential for preserving Evidence that can later be used to verify Decisions and Executions.

The information represented by Fact Tags generally originates from two types of Sources. The first consists of Declared / Pre-registered Facts provided and registered in advance by an Actor or Organization. These may include Name, Identifier, Organization, Role, Qualification, License, Country, and Nationality. Such attributes are not inferred or generated by the AI. They are registered through Governance-defined Methods and serve as foundational information referenced in subsequent Authentication, Authorization, and Evaluation.

The second consists of Event-captured / System-observed Facts obtained by the System when an Event occurs. These may include Timestamp, Location or IP, Device, Authentication Result, Session ID, Event ID, and other information observed or recorded in connection with the actual Event. Unlike pre-registered attributes, these Facts provide Evidence for reconstructing what occurred at a particular point in time.

A Fact Tag may contain more than a single Value. It can include a Structure that clarifies what the Fact refers to, where it originated, and the point in time to which it relates. Who / What identifies subjects such as Actor, Organization, or Role. Authentication records Credential, Method, and Result. Provenance preserves Source and Origin, while Temporal Context preserves Timestamp and Event Time. Related Facts may also be associated, including Attributes such as Country, Project, and Qualification, as well as Data, Content, and Event Details.

Importantly, a Fact is Externally Grounded, and a recorded Fact must not be retrospectively altered by subsequent Evaluation. This does not mean that real-world states never change. If a Role or Qualification changes, for example, the updated state can be recorded as a new Fact. However, the Fact referenced at the time of a past Decision must not be rewritten to reflect the current state.

This Temporal preservation makes it possible to reconstruct not only the current state, but also which Facts were referenced at the time of a particular Decision or Event. The Fact Tag therefore functions as an Evidence Layer and, through its association with subsequent Policy Tags, Evaluation Tags, and Process Tags, forms part of the foundation for Decision-State Evidence.