AR Runtime Validator

Overview

At runtime you validate an LLM answer against a guardrail that has an AR policy attached, read the findings, and (if the answer isn't VALID) feed the contradicting/insufficient rules back to the LLM to rewrite, then re-validate. Repeat until VALID or a max-iteration cap (Valid@N).

Reference: ../../shared/references/ar-api-context.md (runtime section + qualifier rules), findings-reference.md (result types + actions), references/rewrite-templates.md (per-type prompts).