RE: Horn clauses (formerly...access control information)

My basic point is that separation of policy specification
from mechanisms for implementing the policy is a good practice -
in this particular instance, the mechanism I would have
"worried" about would be how to represent the policy
in terms of Horn clauses for a specific interpretation system.

On a related note, I am still not convinced that an authorization
model is strictly necessary to write the policy. I am suspecting
our need to use an authorization model is stemming
from a sense of "in practice we need PDP to inject new policy statements
unstated in the authorization policy." Is this true?

Cheers,
-Suresh