Scott’s π (1955) and Krippendorff's α (1980) have been among the most-often used or recommended general indicators of reliability. This article presents paradoxes showing that neither can be a general indicator. We show that π or α should be used only when (a) coders enforce a predetermined quota as the first priority and (b) coders maximize chance coding as the second priority. Because the two conditions rarely hold, π or α should rarely be used.