· Attribute measurements (pass/fail, defect grades) have no numeric value for R&R; instead several people judge the same parts repeatedly and we ask whether they reach the same conclusion — that is attribute agreement, and Kappa (κ) is its core metric
· What κ means: agreement after removing the "guessing right by chance" component. κ=1 perfect agreement, κ=0 about the same as guessing, κ<0 worse than guessing. Industry rule: ≥0.75 good / 0.40–0.75 acceptable with conditions / <0.40 unacceptable
· Compared with the standard (effectiveness): the proportion of each appraiser's judgments that match the standard; AIAG convention ≥90% acceptable, 80–90% borderline, <80% needs improvement. Then look at error direction — one-sided errors usually mean a misunderstood standard; random errors usually mean criteria/training are weak or boundaries are fuzzy
· The same part can be judged 2–3 times per appraiser (one row each) to further assess repeatability within a single person
· Recommended 20–50 samples, and mix pass and fail — if all parts pass, Kappa cannot be computed (no discrimination). It complements Gage R&R, Bias/Linearity and Stability: variable studies ask "is the number accurate?", Kappa asks "is the judgment correct?"