What Kartik demonstrated

Built a perception validation dossier that defined detection rigorously at IoU ≥ 0.5, computed aggregate and VRU-specific precision/recall from a synthetic scored table, and then sliced recall by lighting, visibility, and class rarity to surface the unsafe night/occlusion failure modes. The final recommendation was evidence-based and internally consistent, using the student’s own metrics to justify a daylight-restricted release with targeted mitigations, and the real Uber Tempe case analysis correctly mapped the crash to the night-time pedestrian miss pattern.

• Leakage-Free Detection Definition
• Safety-Critical Metric Selection
• Per-Condition Failure Triage
• SOTIF Hazard Framing