Planning the Unplanned Experiment: Assessing the Efficacy of Standards for Safety Critical Software

Safe use of software in safety-critical applications requires well-founded means of determining whether software is fit for such use. While software in industries such as aviation has a good safety record, little is known about whether standards for software in safety-critical applications 'work' (or even what that means). It is often (implicitly) argued that software is fit for safety-critical use because it conforms to an appropriate standard. Without knowing whether a standard works, such reliance is an experiment; without carefully collecting assessment data, that experiment is unplanned. To help plan the experiment, we organized a workshop to develop practical ideas for assessing software safety standards. In this paper, we relate and elaborate on the workshop discussion, which revealed subtle but important study design considerations and practical barriers to collecting appropriate historical data and recruiting appropriate experimental subjects. We discuss assessing standards as written and as applied, several candidate definitions for what it means for a standard to 'work,' and key assessment strategies and study techniques and the pros and cons of each. Finally, we conclude with thoughts about the kinds of research that will be required and how academia, industry, and regulators might collaborate to overcome the noted barriers.