The assumption A((n)), proposed by Hill (J. Amer. Statist. Assoc. 63 (1968) 677), provides a natural basis for low structure non-parametric predictive inference, and has been justified in the Bayesian framework. This paper embeds A((n))-based inference into the theory of interval probability, by showing that the corresponding bounds are totally monotone F-probability and coherent. Similar attractive internal consistency results are proven to hold for conditioning and updating. (C) 2003 Elsevier B.V. All rights reserved.