[address-policy-wg] 2023-04 Are anonymised assignment objects valid?

Brian Storey wrote on 30/09/2023 00:03:
> For me it comes down to this. As a community & considering the permitted
> exceptions (individuals & P2P assignments):-
> 
> 1) Is the publication of End User entities for an assignment important?
> 2) Is the publication of a prefix assignment boundary between end users
> important?

Brian,

thanks for this thoughtful analysis of this proposal. These two 
questions are indeed the crux of this issue.

I'm also surprised at Angela's answers, but let's deal with the intent 
of the policy for the time being, namely whether the RIPEDB should or 
shouldn't include information about end user assignments and the 
entities to which they are assigned.

My take is that, in theory, there would be a good deal of merit in 
publishing both if there were some way of ensuring that this data was 
accurate and well-maintained. However, from a practical point of view, 
there's a large amount of inaccurate and stale data in the RIPE 
database. A lot of effort has been expended over the last 30 years to 
attempt to address this problem, but no good solutions have been found 
and the outcome

If we haven't solved this problem in 30 years, and if there are no 
proposals on the table to address data quality issue, I'd speculate that 
it's unlikely we'll solve the RIPE DB's data quality issues any time 
soon, and maybe now it would be appropriate to have a good, hard think 
about whether it's worth keeping this policy at all.

But, speculation is not an appropriate substitute for fact. A better 
starting point for addressing the underlying data quality issue might be 
to formally measure it. There's ~4.7m inetnum/inet6num PA assignment 
registrations in the database. A quick check with a sample size 
calculator suggests that examining around 500 random inetnums / 
inet6nums for an a/b data quality outcome would give a result with 98% 
confidence, ~5% error margin.

Is there an appetite for examining this in the DB-WG + the RIPE NCC?

Nick