Re: [Eligibility-discuss] On 3797 alternatives

[Eric Rescorla <[e-mail]>]

On Wed, May 31, 2023 at 8:51 AM Salz, Rich <[e-mail]> wrote:

> > That's right. I'm also assuming that others have access to that
> information (Robert, the previous chair, people writing software for the
> datatracker, etc...) I'm not implying anything about the character of those
> individuals, only saying that a stronger system wouldn't rely so much on us
> trusting those people ... or chance.
>
> Can you tell me what attacks you see as possible?  I assume that Akamai
> (for example) could tell its employees to not volunteer which is effective
> immediately. If it wanted to get many volunteers, they would either be
> already qualified, or there is a more-than-one-year timeline. This was
> discussed in London, and the consensus was that watching the growth of
> eligible volunteers was good enough.
>
> What am I missing?
>

The attack I have in mind is that you have a pool of (say) 10
nomcom-qualified people.
They all volunteer at the beginning and so are on the list. Then once the
rest of the list
is set, you run the numbers and find the optimal set of people to withdraw
to get the
most favorable distributional outcome for the selection process.

It should be obvious that this is worthwhile if the randomness is known in
advance: you
get to choose between 1024 different nomcoms. I haven't run the numbers and
so am
not really sure how many bits of randomness you need to make this attack
ineffective
with a pool of attackers of size N [0]

-Ekr

[0] Note that on average you'll reduce your representation in the pool by
N/2 but that
may be worthwhile because there are people you don't control who
nevertheless have
known preferences.