The idiotypic network has a long and chequered history in both theoretical immunology and Artificial Immune Systems. In terms of the latter, the drive for engineering applications has led to a diluted interpretation of the immunological models. Research inspired by theoretical immunology has produced compelling models of self-organised tolerance and immunity, but currently fail to have any practical engineering applicability. In this paper, we briefly discuss the engineering applicability of "self-affirming" idiotypic networks, leading to a suggestion that the "Third Generation" network models represent a way forward in this respect. Results obtained by implementing and extending a discrete model of this type of network suggest that the extended prototype is capable of two context-dependent modes of immune response, readily applicable to unsupervised machine-learning.