RoboSafe Simulation Corpus v1.0

The RoboSafe Simulation Corpus is a standardised 30-scenario test battery for evaluating the safety governance of physical AI character deployments. Scenarios are organised into six categories: adversarial input (jailbreak and prompt injection), gray zone (context-dependent ambiguity), cultural mismatch, scope escalation, emotional manipulation, and off-topic escalation. Each scenario specifies a deterministic pass condition expressed as a boolean function of the character's safety configuration (SafetyHarnessConfig), enabling automated certification without requiring LLM inference. Results are fully reproducible across platforms. This is the first publicly released safety scenario corpus specifically designed for social robot and physical AI deployment contexts. It accompanies the preprint: Governing Physical AI Characters: A Regulation Mapping, Simulation Corpus, and KPI Framework for Social Robot Deployments (arXiv cs.RO / cs.CY). Released under CC BY 4.0.