I have been looking at belief revision when belief states are
represented as default theories. Both my PhD dissertation and my ECAI-98
paper discuss how this approach enables us to implement persistent
contraction, context-sensitive choice functions and mechanisms for
retaining beliefs which would otherwise be discarded in anticipation of
future re-use. It makes it easy for belief change operations to be
reversible/recoverable (I'm not convinced that this is undersirable, as
some maintain). Non-prioritized belief change operations are easily
defined in this approach and the problem of iterated belief change finds
a simple solution (with provably minimal change over iterated steps
being an additional bonus).