Targeting: Level 3+ ADAS, Driver Monitoring (DMS), and AI-based Perception Stacks.

Section A: Article 12 – Automated Logging (The "Black Box")

Goal: Ensure the system provides enough data to reconstruct an event.

• [ ] [ ] Inference Context Capture: Does the system log the "Confidence Score" for every safety-critical object classification?
• [ ] [ ] Decision-Path Traceability: In the event of an OEDR (Object and Event Detection and Response) failure, can you trace which specific neural network layer or logic gate triggered the action?
• [ ] [ ] Immutable Time-Stamping: Are logs synchronized across the domain controller and sensors with a tamper-proof timestamp (PTP/TSN)?
• [ ] [ ] Resource-Efficient Logging: Have you defined the "Trigger Events" (e.g., emergency braking, system override) that initiate high-fidelity logging to prevent SSD/Flash wear?
• [ ] [ ] Log Retention Architecture: Is there a defined path for off-boarding logs from the vehicle to the OEM cloud for the required duration of the system’s lifecycle?

Section B: Article 19 – Conformity Assessment (The "CE Mark")

Goal: Prove that your "Design-to-Production" pipeline is compliant before the car hits the road.

• [ ] [ ] Annex IV Alignment: Is your Technical Documentation mapped to the specific "General Description" requirements of Annex IV?
• [ ] [ ] System Integration Evidence: Can you prove how the AI component interacts with the non-AI "Safe State" logic (ISO 26262)?
• [ ] [ ] Post-Market Monitoring (PMM) Plan: Do you have a technical process to update the "Conformity Assessment" if the model is retrained via Over-the-Air (OTA) updates?
• [ ] [ ] Risk Management Interoperability: Are the AI Act risks mapped directly into your existing HARA (Hazard Analysis and Risk Assessment) and FMEA (Failure Mode and Effects Analysis)?
• [ ] [ ] Quality Management System (QMS) Update: Has your ASPICE process been updated to include "Data Quality Management" (Article 10)?

💡 The "LXD" Professional Insight

"Most teams fail Article 12 because they log hardware signals, not AI reasoning. If your auditor asks 'Why did the car think that shadow was a pedestrian?' and your logs only show the braking pressure, you are non-compliant."