Generates a plain-language risk narrative from structured check results. Designed for non-technical users reviewing AI agent output. Supports KYB and invoice fraud contexts.