The most general reliability number: Krippendorff’s α works for any number of coders, any measurement scale, even with missing data — one minus the ratio of the disagreement you SEE to the disagreement you’d expect by chance. Slide from noise to consensus.
Krippendorff’s α = 1 − Dₒ/Dₑ: observed disagreement over expected (chance) disagreement, computed over all codable value pairs. α=1 perfect reliability, 0 chance-level, <0 systematic disagreement. Unlike percent-agreement it corrects for chance and handles any scale via a difference function (here nominal: 0 if equal, 1 if not). A fail-loud self-check throws unless perfectly-matched coders give α=1. Convention: α≥0.80 is reliable, 0.667 tentative. ◆ real statistics, node-verified.
Nominal difference on a small table is the illustration; the Dₒ/Dₑ ratio is the exact α formula. The 0.80 threshold is Krippendorff’s own convention, not a law.