2. Applicability of distribution of "exact" policy table is too restricted.
An implementation may want to have their own policy, or more attributes
(probably, in addition to ifindex), e.g. traffic class or whatever.
The ifindex should be assumed one of extensions to the "standard"
policy table, and the details should be left to implementations.
The policy announced from network cannot be set directly.
I know that conflicts are common, but, I would say, the distribution
should not (or cannot) be an exact one, but a "hint", "suggestion" or
"recommendation". I do think it is much better to have information as
"relative" representation, but at least, we should make the interpretation
clear.
Of course, an implementation may assume such information an order from
network, but the network policy can only be enforced by the network.
If the interpretation of the "policy" is relaxed, we will have more
chances to use such framework.