I am still unclear of the real benefit of having an 'active' choreography
mechanism that is required to manage its own state - is there a clear
business/use case that warrants this, over and about the passive monitoring?

The problem I see is that the logic within the choreography, related to
decisions, would end up having to replicate the same logic/state that is
within the participant, in order to determine which route the participant
will take - but what if the decision made by the participant is based on
private data (e.g. from a customer database) that is never exposed in any
message? Or does a participant implementing this choreography mechanism need
to expose this state information, so it is just shared by the choreography
mechanism?

Then there are the problems of keeping this logic in the participant and the
choreography in step.

If the choreography simply represents the alternate flows that could be
taken, and the actual flow is deduced by the observed message exchange, this
simplifies the monitoring, aswell as the maintanence of the choreography.
(G. Brown)