Some suggestion for finetuning the Certification parameters V2

Choosing the right certification parameters could be very tricky, too high requirements could hinder growth, too low requirements could encourage the creation of double accounts.

Here are some thoughts to that:
instead of focusing on choosing the right parameters we could algorithmically define the maximum growth rate per time.

How could this look like?
for example we could limit newcomers per year on x% (for example 100%) of the current members with minimal y (for example 1000) newcomers

The question would be then which newcomer can come in first if the limit is reached.
the simplest solution would be, first come first served. that means the newcomers who at first fulfilled the web of trust requirements would come in first.

With such an limit in place the allowed certificated per time could be chosen more freely, like for example:

  • at max 30 certifications per 30 days per member

Is there any argument why member growth should be limited? Just asking because I can’t think of one.

Sybil cancer. If you allow any number of certifications, even worse if you also allow them instantly, you allow machines to invade the currency before humans can even react.

The limit of 1 certification / week is a solution to address both problems:

  • because we need time to make certifications, so cancer won’t occur instantly
  • because we have a limited stock (we cannot have more than $certValidity / 53 weeks$ certifications made in our example of 1 certification a week)

Ok, I see. But is it neccessary to have an absolute growth limit, i.e. grow at max. x% per time frame? Or would something like “a single member can certify at max 150% of average certifications per member per time frame” also do? Just because I’d like to see this adapt more to demand over time. So when there’s a wave of newcommers comming from several directions/knowing many different existing members, max. certifications would get higher. After the wave is over, the max. certification would drop back to normal. On the other hand, if someone issues many sybills, these can only be certified by a small number of existing members and max. certifications would not rise very much.

1 « J'aime »

?
what is $certValidity$ (validity of one certification?)
where is 53 coming from? why not 42 :smile:
what is weeks$?
which example?

yes this would hinder ¨spam¨ attacks.
yes this would help in bubbles to certify more people
this would not help to predict an certain growth per time, but could be combined with other methods like outlined above
the algorithm would need to guess the right balance between active certifying people and non active people
for example one solution would be to allow higher percentages then average like up to 1000% (normally you have round about 10% active people in an community at normal times), or you have to count only members for the average that make more then x lets say at least one or two certifications in one month. otherwise we could easily end of in stagnation if many people do not actively take part in certifying.

At our semestrial meetings (FMM/RML), we imagined that a possible value for certification validty would be 1 year. We also though about a possible value of delay between each certification that could be 1 week.

In this example of values, and givien we have approx. 53 weeks in a year, we have a mechanical stock of 53 certifications by member, without even defining explicitely that a member would have a limited stock of 53 active certifications.

That’s what I meant :smile:

ah now i understand :smile:

year 53 seems like enough to have.

but instead of one week we should use something like 4 per 30 days, then one has not to look all the time if one week is over :smile:

No because this rule alone allows Sybils to certify themselves instantly, you need a minimum delay.

1 « J'aime »

yes thats true ;/

maybe max 2 certifications per 2 weeks could still have little bit more flexibility and still have the same sybill protection. in both cases after two weeks he could make his 3. certification.

1 certification per week feels very slow. but i understand, that if we have no possibility to revoke a certification before one year it could be tricky to have more certifications.

just some thoughts:
maybe we could limit the lifetime of the first certifications to one month, then we could also increase certifications per week of members that are older then one month.

You know, I said 1 certification a week, but we could also set 1 certification per 3 days. This would double the potential quantity of certifications, so approx. 123 certifications max / year.

yea its hard to guess any right parameter. maybe we should really think more about kacees average certifications per member suggestion?

UPDATE:
another idea would be to not count certifications if the member already certified you. these re recertifications do not increase the Sybil attack risk.
these would also encourage ¨recertifications.¨

on top of that newcomers could be limited to only two non re certifications for one month.

in general certifications to newcomers have a total different risk to certifications to members and could be treated differently.

My opinion is we should rather make simulations to understand the impacts of all what wee say :slight_smile:

Because we say a lot of things (I included) without deeply seeing the consequences. Also I am not focused on rules yet, and this problem requires a lot of concentration. So I will give clues of my work on this problem later with few studies I will let you appreciate at that time :smile:

1 « J'aime »