This takes output as input. Would you go with a self-consistent result, make it time-inconsistent, or cut it off at one tier?
A better solution, I think, would be to weight the karma change by the ‘information’ that this new vote provided if the order of votes was irrelevant—i.e. multiply it by min(1, log2(P)) with P being the fraction of the voter’s votes that are of that vote type. So if Bob likes Alice’s post when Bob likes everyone’s posts, Alice doesn’t get much from it. If Bob like’s Alice’s post when Bob likes half or fewer of all posts he votes on, Alice gets 1 full karma from it.
Time-inconsistent. Nothing you do after you upvote should change the results. This risks having karma determined almost exclusively by how many posts are upvoted by the top elite.
Perhaps a better solution could be found if we could establish all of the goals of the karma system. Why do we track users’ total karma?
This takes output as input. Would you go with a self-consistent result, make it time-inconsistent, or cut it off at one tier?
A better solution, I think, would be to weight the karma change by the ‘information’ that this new vote provided if the order of votes was irrelevant—i.e. multiply it by min(1, log2(P)) with P being the fraction of the voter’s votes that are of that vote type. So if Bob likes Alice’s post when Bob likes everyone’s posts, Alice doesn’t get much from it. If Bob like’s Alice’s post when Bob likes half or fewer of all posts he votes on, Alice gets 1 full karma from it.
Time-inconsistent. Nothing you do after you upvote should change the results. This risks having karma determined almost exclusively by how many posts are upvoted by the top elite.
Perhaps a better solution could be found if we could establish all of the goals of the karma system. Why do we track users’ total karma?