pull down to refresh
102 sats \ 32 replies \ @k00b 19 Nov \ on: Stacker Saloon
working out the game theory for removing trust from ranking is simple, but tricky for rewards because when you sacrifice enough (via the sybil fee) you may be able to recoup your sacrifice in rewards ... which makes it no sacrifice at all when you self-zap. downzaps are some defense against this if they're incentivized but would lead people to downzap rivals potentially without sacrifice too.
one solution to this would be randomizing rewards. meaning we'd randomly skew the reward distribution away from the top content by just enough to make recouping one's sacrifice unlikely. but if top content is defined by sats, even skewing away from top content would still slightly reward self-zapping.
another option is to continue using trust but only for rewards.
reply
Would another solution be to randomly select one zapper each day who gets the whole days rewards?
This would incentivize creating a multitude of accounts and zapping a little from each, so perhaps there needs to be another qualification: zaps at least a certain amount, or zaps and comments.
I remember when we had the somewhat random rewards function that varied the criteria. Maybe it's that sort of model, but with a random winner function, like a lottery.
reply
Would another solution be to randomly select one zapper each day who gets the whole days rewards?
I still remember million sat madness where the whole rewards of a month were given to just 64 stackers, including 1 million to the top.
I remember not enjoying that. My philosophy would be that rewards should feel smooth, not lumpy, and that it should be a reliable, and fairly predictable, incentive for consistent participation.
reply
MSM wasn't pleasant, I agree. And probably i agree with you that rewards ought to be aimed at being predictable. Yet I wonder if there isn't a short enough time frame, and a small enough reward sum, that makes the random winner mode feel more like fun and less like dreadfulness.
reply
reply
That difference makes sense.
The truly random option you describe in another comment (#1286158) probably makes more sense. But the dopamine lover in me thinks it would be pretty awesome to take home the whole pot.
I suppose it would be possible that someone figures out an optimal zapping strategy that would lead to a reliable gain of sats (thereby defeating the purpose of Sybil fees).
This could happen with a reward spread out over some random segment as well, but not for a single big winner.
reply
Can you point me to the code that defines the reward function?
reply
why do you still keep rewards?
reply
reply
The alternative would be to pocket the sybil fees and downzaps. I'd be fine with it, but I think many people would find it aesthetically displeasing.
What happens to the model if you take "reward for top post/comment zaps" out?
reply
People have no direct incentive to zap
reply
Wait there is an incentive to zap?
reply
A user must pay to post and they must pay to comment, but they don't necessarily have to pay to see more of what they like on the site. Of the three actions one might do on SN, zapping is the least directly tied to the outcome one might want (more of this kind of content).
Rewards being tied to zapping seems like a way one might encourage more zapping and if there is more zapping, the hope is that the connection between zapping content you like and seeing more of the content you like might become more clear.
reply
reply
reply
reply
I wouldn't call it hacky and lazy.
I'm not sure it's possible to have both sybil resistance and make people whole on zaps, on average.
Thus, the real net incentives to zap must come from either:
- A desire to shape the content on SN
- Altruistic desire to support content that you like
- Build your trust on SN to enhance your ability to do the above two
Since (1) and (2) cannot easily be controlled by SN, the best lever SN has to play with net incentives to zap is (3). Thus, I don't think the trust system should be removed, though its influence on post rankings could be tinkered with.
The balancing act with trust, I think, is to have a good new user UX, to avoid platform insularity and dominance by a handful of trusted users, while simultaneously providing new users enough incentives to zap content and develop a positive reputation on SN.
We talked about this, but one thing that might help is using a rolling window of zaps to calculate trust instead of the entire history, since it may ossify trust levels too much on a few users.
reply
Fair.
For our No Trust experiment, I'm leaning toward removing trust from ranking but continuing to use trust in rewards. I'll also tweak trust to have this sliding time window.
The hypothesis being that sybil resistance of zaps should keep ranking healthy and using trust in rewards should prevent this sybil recoup by not having the sybil fee/sacrifice be a direct input to rewards.
reply
reply
Posters, commenters and referral program. I think!
reply
I said there's no direct incentive to zap quality content. There's an indirect, warm and fuzzy feeling for zapping quality content otherwise, and an increase in the possibility that OP posts more, but no direct incentive.
Not having a direct incentive isn't that bad in isolation but there's direct incentive for self/sybil zapping (I want people to see my content more so I zap it), so the goal is to create a direct incentive for zapping other people's content that rivals the incentive to sybil.
reply