The 150 baseline adversarial cases Picaroon runs against ecommerce refund agents — every sample scored against the OWASP LLM Top 10, the CSA Risk Rubric v2, and the Picaroon vertical catalog. Public, unauthenticated, and written so you can read what a retainer re-runs every cycle.
Every row below is a redacted adversarial case from the catalog. Each carries its OWASP LLM Top 10 and CSA Risk Rubric v2 tags inline; open the family for the full redaction snippet, reproducer, and vertical rubric.