Technical reports · 2015.01
Possibilities and limits of statistical software-safety evidence
Determining Software Safety: Understanding the Possibilities and Limitations of International Safety
Overview
Julia V. Bukowski examines Annex D statistical testing through input independence/distribution, internal states and correct output classification. Her analysis of why failure-free hours alone are insufficient offers background for scrutinizing sampling assumptions in AI assurance.
Publication and version information
Public publisher PDF copy; cover date, posting date and later-copy copyright years are distinguished.