<snip>
> > - The assumption in Section 3.2 that one must find an old policy in
> > order to determine if an algorithm was valid at a point in
> the past is
> > too complicated. Suitability definitions should accumulate in a
> > single policy definition. An enterprise could maintain several
> > policies. For example, one complete, one current and one
> past policy could be maintained.
> >
> In our notion, the policies are published by specific
> institutions (e.g.
> annually) and one policy represents the evaluations based on
> current knowledge (e.g. on current findings, RSA with 1024
> bit key length could be valid until end of 2007, but next
> year a new policy could be published which states that RSA
> 1024 is valid until end of 2008).
> To expect, that a policy also contains all past evaluations
> of an algorithms could be error-prone.
> In our opinion, the question, if the evaluation of an
> algorithm in an old policy is different from that in the
> current policy is primarily important in law cases. And there
> you cannot trust, that the current policy correctly quotes
> past evaluations, instead you will have to look in the old policy.
I wasn't suggesting that a policy contain all past evaluations. I was suggesting that the current policy would contain the current position with regard to an algorithm's life span, even if the algorithm is no longer viable, and that's the only thing that matters. I don't see the value in referring to an old policy since the position on an algorithm can change over time. Using your example, in my opinion, the only thing that matters to the verifier is what the current position is with regard to an algorithm. The fact that an algorithm's expected life span at the time a signature was generated was thought to be shorter than it turned out to be isn't important.
Annual publication makes sense and makes it easy to perform the accumulation. Someone verifying an evidence record will need to know when an algorithm stopped being suitable. It'd be nice if this information could be contained in one location instead of having to collect old policies, i.e., verifiers always use the current edition. My comment that there could be multiple policies was probably confusing. The idea there was that for apps that don't use any expired algorithm, there could be a very brief policy statement with just the currently viable algorithms. Having one full statement is probably easiest though.