C3.2 Honest scales

Probability and impact literature; Cox's critique [12]

What it requires. Every likelihood and consequence label maps to a numeric band, and the bands are versioned, so that a trend across two assessments is a change in the risk and not in the ruler. Consequence is assessed per objective and the highest severity governs the row, not the average. Every estimate carries a precision field saying how it was obtained and how good it is. Probability words carry their bands. A matrix may be used for communication only after its cells have passed a consistency test, and it is not used to decide.

Where it comes from. Fifteen years of practitioner papers in the project literature [12] accumulated these repairs one at a time: versioned bands, impact per objective with the maximum rule, a precision field, a calibrated vocabulary after surveys showed “possible” meaning anything from one chance in four to one in two. Cox's 2008 paper supplied the theorem: a matrix ranks correctly only if its cells satisfy consistency conditions, and one that does not will sometimes place a larger risk below a smaller one.

What it reconciles. The matrix is the most widely used analytical device in the set and the one its own community has criticized most honestly. The component keeps it, because boards read it, and demotes it, because Cox showed what it cannot do. It fixes the fifth and sixth clusters of chapter 11 as fields: consequence is the description, severity the score, per objective; likelihood is the word, the band the number. The precision field is where a workshop guess and a hundred observations stop being the same data, and it is what C5.5 reads first.