Reduce jurorsForCourtJump parameter in translation courts

Proposal
#1

Proposal

Update/change the jurors for jump parameter in each of the following courts:

  • Spanish-English: 17
  • French-English: 5
  • Portuguese-English: 3
  • German-English: 8
  • Russian-English: 3
  • Korean-English: 1? (currently no stake)
  • Japanese-English: 2
  • Turkish-English: 3
  • Chinese-English: 5

Rationale

Jurors for jump determines the point at which an appealed case goes to a parent court, see: Explaination of “Jump” and “Alpha” parameters Currently all of these subcourts are set to have jurors for jump at 128. The proposed values were made based on current amounts staked in each court, based roughly on the principle that if an appeal within a subcourt is unlikely to draw in new unique staker addresses compared to the previous round, there is little value to having that appeal in the subcourt and one should jump to the parent court. The heuristic used here is to simply set the jurorsForCoutJump parameter to the number of addresses currently staking in the court. Please don’t hesitate to tweak the numbers to something cleaner and more sensible/principled when submitting the proposal to snapshot. As the number of disputes and jurors in these subcourts grows, the jurors for jump can be further updated.

(note: format and text copied partially from Jurors for jump Parameters Updates Proposal)

KIP-39: Update jurorsForCourtJump Parameters in Translation Courts
#2

Jurors are doubled + 1 for every appeal, I think the values you proposed are extremely low and disputes would benefit of additional attention by jurors with those linguistic skills before disputes being sent to a higher court where most jurors would not speak the language.

1 Like
#3

Sure, but aren’t the values I proposed equivalent to:

  • Spanish-English: 31
  • French-English: 7
  • Portuguese-English: 3
  • German-English: 15
  • Russian-English: 3
  • Korean-English: 1
  • Japanese-English: 3
  • Turkish-English: 3
  • Chinese-English: 7

i.e. the next 2^n-1? And if I haven’t misunderstood, this should mean that a court jump would occur only after an appeal with 2^(n+1)-1 seats. So for instance, in the case of the Spanish-English court that would mean an appeal would only occur once 1+3+7+15+31 = 57 juror positions have been selected in total.

But I do agree that many of these values could benefit from an increase. So how about (increased values in bold; total votes before court jump / number of juror addresses in (parentheses)):

  • Spanish-English: 31 (57/17)
  • French-English: 15 (26/5)
  • Portuguese-English: 7 (11/3)
  • German-English: 15 (26/8)
  • Russian-English: 7 (11/3)
  • Korean-English: 3 (4/0) (currently no stake, but let’s anticipate)
  • Japanese-English: 3 (4/2)
  • Turkish-English: 7 (11/3)
  • Chinese-English: 15 (26/5)
#4

Taking a jurorsForJump equal to the number of addresses staked in the court is not necessarily a good heuristic because addresses can be drawn more than once taking up more than one spot. (For more information on why this point is like this, see the questions under “proof of humanity” here: https://blog.kleros.io/learning-and-improvement-in-kleros-mechanism-design/) Typically an additional appeal in a child court is not so useful when you get past the point where the previous rounds already give good statistical evidence that the a large, randomly drawn panel of jurors from that child court would decide a given way. How large a panel you need for this depends somewhat on how close the votes are: if 18 out of a panel of 20 jurors vote for Alice, that might give you stronger evidence that a large panel of jurors from the court would at least vote 50% for Alice than 51 out of a panel of 100 jurors voting for her.

Nonetheless, there is certainly little sense in increasing the jurorsForJump past the point where every staked juror is likely to be drawn before you jump to the next court. For these courts, where most of the jurors in the parent court don’t speak the language it seems particularly worthwhile to try to err on the side of drawing in more or less everyone that is staked in the court, to have as many qualified jurors as possible paying attention. One might hope that if a controversial dispute is appealed a few times it might catch the attention of other qualified people who would then stake in the court and be eligible to be drawn in appeal.

The (second set of) values you propose in most courts would be enough so that the jurors with the smallest stakes/at the minstake would mostly have a greater than 50% chance of being drawn at least once over the rounds, but several of those chances are in the 50-80% range for being drawn at least once, so a fair number of those low-stakes jurors would probably still get unlucky and not be drawn. As an example, if you take the smallest staked juror currently in the Turkish-English court who has roughly 2600 PNK staked out of roughly 30000 PNK in the court, his chance of getting any given draw are 8.7%, so as the number of draws follows a binomial distribution, his odds of being drawn at least once out of 11 chances are roughly 63%. (One can compute these values with calculators such as https://stattrek.com/online-calculator/binomial.aspx)

So, while I agree that it might make sense to decrease some of these values somewhat based on our observations of current use and staking, these propositions might still be rather low.

1 Like
#5

Thanks for your patient explanations and your informative article. I do disagree with some of the points you made however…

On the contrary, I would find a sudden increase in jurors during appeal to be very suspicious and a possible indication of jury stacking. I feel that if people were qualified, they should have staked long before the case started, and one might even argue that the Kleros contract should assign a reduced weight to new jurors. Of course, this wouldn’t help for long-premeditated attacks, but it would perhaps help neutralize the most egregious attempts at jury stacking. And there’s also the issue of code complexity and therefore gas costs…

Anyway, I get your arguments but applying the principle that all jurors should have a high chance of getting picked at least once leads to (what is in my opinion) an absurdly prolonged appeal process in many cases. Take the French-English court for instance. It has one juror with an absolute majority (60000 vs 2000+1542+1000). Now, in order that all jurors would be likely to have a say (and that’s all it would be) (~79% for the 1000PNK staker), you would need jurorsForCourtJump=63 (total votes: 120), i.e. 6 rounds of Mr. 60000PNK winning over and over and over. And since we don’t know if the small value jurors are three distinct individuals or the same person with different addresses anyway, even if they all consistently voted against the 60000PNK staker, their plurality is not something that should be taken into account by the higher court anyway. Furthermore, if all you want is the opinion of the low-stakes jurors (and given how useless their votes are, that’s all you can hope to get anyway), surely they could provide a signed message with their opinion (although of course they have no financial incentive to do so).

There are however two kinds of evidence a higher court could still take into account:

  • Expert testimony, where the higher court would have to evaluate whether the person having made the testimony is trustworthy.
  • Evidence of subversion in the lower court (see my comments about case 546 [1] which I believe to be the target of jury stacking).

All this being said, I think we should be able to agree that the current court jump parameter (127) is too high for the translation courts in their current state (except perhaps for the Spanish one), but since I’m not sure exactly what would satisfy you or the community, could you make a proposal yourself?

[1] Likely jury stacking in case 546 & court jump impossible

#6

I feel that if people were qualified, they should have staked long before the case started, and one might even argue that the Kleros contract should assign a reduced weight to new jurors.

This is an interesting idea. Arguably it could make the court more vulnerable to long term attacks as a long-term attacker might need some stake that is less than 51% to have a high probability of being able to carry out a 51% attack. (With the current structure, at least someone who has an amount of PNK that is a majority of the current general court stake but not a majority of the overall PNK in circulation that might consider attempting a 51% attack would have some risk of failing due to new stakers joining that they would have to take into account.) However, the idea certainly has some arguments in its favor.

Anyway, I get your arguments but applying the principle that all jurors should have a high chance of getting picked at least once leads to (what is in my opinion) an absurdly prolonged appeal process in many cases. Take the French-English court for instance. It has one juror with an absolute majority (60000 vs 2000+1542+1000).

To be clear, I was advancing the “jurors have a high chance of being drawn at least once” as a sort of heuristic. One certainly wouldn’t want to apply it blindly in cases where courts seem to be particularly whale dominated. (Indeed, if one imagines coding this reasoning into the smart contract to give values of jurorsForJump that auto-adjust to current stakes, that would be sort of manipulable by a whale, who by staking more would mean that a higher jurorsForJump would be needed to get a good chance of drawing everyone. There are elements here of Goodhart’s Law that if a measure becomes a target it ceases to be a good measure. A heuristic based on setting the jurorsForJump around the number of distinct addresses in the court is, of course, similarly manipulable as a whale could split up their holdings to make it seem like a higher jurorsForJump is justified.)

With all of that in mind, how would people feel about the following values?:

  • Spanish-English: 63
  • French-English: 31
  • Portuguese-English: 31
  • German-English: 31
  • Russian-English: 31
  • Korean-English: 31
  • Japanese-English: 31
  • Turkish-English: 31
  • Chinese-English: 31

In a couple of these courts, it is admittedly the case that with current stakes even a jurorsForJump of 31 would imply a couple of rounds where you are likely to effectively be drawing the same people. However, with panels of 31 jurors, one at least has a reasonable chance of the panel being relatively reliably statistically representative of the current stake of the court. (A panel of 7 in particular is still small enough that you could imagine weird draws due to chance.) So these choices might be reasonable, particularly if one is somewhat optimistic about the growth of stakes in some of these courts in the mid-term.

1 Like
#7

For my part, I’m okay with these new values.

#8

Absent further discussion here, I created a post here: KIP-39: Update jurorsForCourtJump Parameters in Translation Courts with the “Vote” tag proposing those specific values. So if anyone else has any thoughts between now and when it goes to Snapshot for the vote, they can comment on that post.