Defining Baconian Probability for Use in Assurance Argumentation

The use of assurance cases (e.g., safety cases) in certification raises questions about confidence in assurance argument claims. Some researchers propose to assess confidence in assurance cases using Baconian induction. That is, a writer or analyst (1) identifies defeaters that might rebut or undermine each proposition in the assurance argument and (2) determines whether each defeater can be dismissed or ignored and why. Some researchers also propose denoting confidence using the counts of defeaters identified and eliminated-which they call Baconian probability-and performing arithmetic on these measures. But Baconian probabilities were first defined as ordinal rankings which cannot be manipulated arithmetically. In this paper, we recount noteworthy definitions of Baconian induction, review proposals to assess confidence in assurance claims using Baconian probability, analyze how these comport with or diverge from the original definition, and make recommendations for future practice.