Structural Audit for Autonomous Driving Safety Metrics

A Public-Data Review of Disengagement, Crash Reporting, ODD Comparability, and Edge-Case Risk 

Public autonomous-driving safety metrics such as crash counts, miles driven, disengagement rates, and safety-driver interventions are useful but structurally incomplete. Without normalization by ODD, scenario type, road condition, reporting capability, and edge-case exposure, these metrics may support misleading safety conclusions. 

What Is Being Audited? 

crash counts；

miles driven；

miles per disengagement；

safety-driver intervention；

incident reporting；

ODD coverage；

scenario coverage；

edge-case exposure；

simulation-to-road gap。

Core Audit Questions 

Are crash rates comparable across different ODDs?

Are disengagement events defined consistently?

Does total mileage reflect exposure to difficult scenarios?

Are rare but high-risk edge cases visible in average metrics?

Are L2 ADAS and ADS data being mixed improperly?

Are reporting capabilities consistent across companies and systems?

Can the public metrics support strong safety claims?

Main Failure Modes 

Method Name : Structural Safety-Metric Audit  (S-SMA Framework )

Suggested Conclusion :

The purpose of this audit is not to rank autonomous-driving companies or certify system safety. Rather, it demonstrates that public safety metrics require structural normalization before they can support strong conclusions. A structurally valid safety-data audit should account for ODD, scenario exposure, reporting capability, disengagement definition, tail-risk visibility, and reproducibility. 

1. structural audit；

2. safety metrics；

3. public data；

4. disengagement；

5. crash reporting；

6. ODD comparability；

7. edge-case risk.