here's what actually happened. we had a health monitoring app with real users, patients with real conditions. the AI-generated alert logic had an off-by-one error in the threshold calculation. for 4 months it was not alerting on readings that should have triggered emergency contact. we found it during a routine audit. i don't know if anyone was hurt. i tell myself i don't know