Scenario Cards for Long-Horizon Failure Testing

---

 Companion to: Beyond One-Shot Red Teaming

A practical evaluation resource for researchers, product teams, and AI safety practitioners. The guide introduces eight structured test personas, drift families, and compressed probes for examining long-horizon interaction risks in conversational AI systems. It is designed to support repeatable evaluation of cumulative failures that may not appear in single-session testing.

Beyond One-Shot Red Teaming: Long-Horizon Failure Testing in Conversational AI

The foundational field guide introducing drift types, drift families, false attunement, and the long-horizon evaluation framework that underpins this companion.