Implementation: IMPLEMENTED (contract) · ENABLED (nonce fencing in the work runtime) · Exposure: INTERNAL. See status labels. Two labels travel with content. Trust says how far to believe it. Sensitivity says who may see it.

​
Architecture contract

Sensitivity uses PUBLIC, INTERNAL, RESTRICTED, SECRET. Eligibility is a policy function applied before utility, so a component that may not see a chunk never has it ranked. Untrusted content is fenced and compiled, never concatenated into control instructions.

​
Current implementation

On main, untrusted text passed between nodes is wrapped in a fence with a random per-render nonce. The compiler-level enforcement across all context is not on main.

​
Availability

​
Developer relevance

Label everything you emit. Treat unlabeled input as UNTRUSTED.