Knowledge base evolution covers not only the maintenance of an existing KB [\protect\citeauthoryearCoenen and Bench-Capon1993], but also the continous improvement of the KB, its structure and content. Knowledge base evolution operates on the KB of a knowledge-based system. Thus, for an overall description of knowledge base evolution in the RPPP context we distinguish two main units (Fig. 5): the knowledge-base itself (RPPP) and the knowledge-evolution system (KES).
The KES operates as a meta-level system on the object level KB. Reasoning in the knowledge evolution system is performed by the exploration and verification components.
The iteration cycles can be arbitrarily interleaved, permitting evolution to consist of dual verification and exploration processes. Together they form a heuristic, approximative process that alternates focusing and processing phases and improves the KB any time a sufficient amount of knowledge for an update (i.e., assimilation or repair) is accumulated within the KES or provided by the user. For example, assume that the verifier has identified a rule whose premises cannot be satisfied in a given KB. The explorer could then try to generalize that particular rule or to complete the missing knowledge reachable from its premises. Conversely, after the explorer has discovered a pattern (e.g., a new or generalized rule) the verifier may be asked to verify the KB, focused on the assimilated pattern.