pull down to refresh

Brainstorm: No Trust November
7101 sats \ 11 comments \ @k00b 3h meta
November has come and gone, but I'm beginning work on this experiment.
For anyone that wasn't deep in the threads where we teased this, we'll be running an experiment where trust isn't used in ranking anymore.
I promised to provide a venue for feedback before introducing this, so here it is.

history

When SN was started, we didn't have sybil resistance for our zaps. This forced us to introduce a web of trust ranking system. Not long after, we added sybil resistance to our zaps by sending 10% of any zap to the rewards pool. Today, 30% of any zap is sent to the rewards pool and territory founders.

hypothesis

This 30% sybil fee may be enough to make SN's ranking trustless, easier to comprehend, and funner to participate in.

what this means in terms of ranking

  1. 100 sats zapped from any stacker is equivalent to 100 sats zapped from anyone else
  2. 100 sats zapped from any stacker is equivalent to 1 sat zapped by 100 different people, or 2 sats zapped by 50 people, and so on
  3. 100 sats zapped from any stacker, and any number of stackers, is equivalent to 100 sats boost by the OP

more detail

  • boost behaves exactly like a zap of the same amount
    • we'll remove the boost minimum, ie you can boost 1 sat or 10 or 1m and it affects ranking like one or many stackers zapped you that much
    • top boosts can still be pinned, but pinning will only last a week rather than a month
  • meme monday will no longer have a bounty, but the meme with the most sats will still be included in the newsletter
  • 100% of downzaps go to the rewards pool
    • also, because downzaps have ~3x the provable sacrifice of a zap or boost of the same amount (ie downzaps send ~3x the amount to the rewards pool), they will have ~3x the effect on ranking
    • something becomes outlawed when 0.3*(zap_sats+boost_sats) - downzap_sats <= -1000 sats
  • 100% of rewards will go to top zappers of posts and comments, but rewards will still be distributed based on a WoT ranking system (the incentives of a trustless rewards pool is very tricky and this shouldn't influence the experiment too much anyway)
That's what is planned for the experiment currently. My suspicion is that our ranking algo is overindexed on trust currently and we'll either find out we don't need trust at all or learn that we need a lot less than we think.
If you'd be interested in seeing the above modified, speak now and forever hold your keys.
write
preview
related posts
100 sats \ 4 replies \ @SimpleStacker 3h
so in addition to no trust, there will also be no log transformation?
reply
0 sats \ 3 replies \ @k00b OP 3h
Good question. We'll probably keep the log transformation but it'll apply something like as follows: log(0.3*(zap_sats+boost_sats) - downzap_sats).
The log transform won't be applied per zapper, but for the entire item.
reply
0 sats \ 2 replies \ @k00b OP 1h
The more I think about it, it might be worth removing the transform. It’s less transparent and discourages zaps on posts that have already earned a lot.
reply
100 sats \ 1 reply \ @SimpleStacker 1h
That's a pretty interesting hypothesis. I can see it going both ways. With the log-transform, someone who really wants to boost the post and is sats-rich is incentivized to make larger zaps, since they need to in order to influence the rankings.
Similarly, with log-transform, sats-poor users are more incentivized to make small zaps on posts that already have a lot of zaps, since each unique user zapping is relatively more impactful.
reply
0 sats \ 0 replies \ @k00b OP 1h
The fear without the log transform is we end up with rich-rank or promotional-motive-rank. (I’m more afraid of the latter.)
The problem with the log transform is that we don’t know that unique zappers are actually unique (and need trust to make a better guess).
reply
100 sats \ 0 replies \ @Scoresby 3h
Trust has always been opaque to me.
This:
the relationship between sats contributed and a stacker's influence on item ranking is not linear, it's logarithmic: the effect a stacker's zap has on an item's ranking is trust*log10(total zap amount). This basically means that 10 sats equal 1 vote, 100 sats 2, 1000 sats 3, and so on ... all values in between and above 0 are valid as well.
And this:
The only consideration that factors into a stacker's trust level is whether or not they are zapping good content. Zap amounts do not impact stackers' trust scores.
from the faq are pretty clear, and yet I've struggled to keep them in mind.
It fells like no-trust will make things a little easier to think about.
reply
21 sats \ 1 reply \ @winteryeti 3h
I'm usually sus of everyone anyways...lol
reply
0 sats \ 0 replies \ @k00b OP 3h
Then you'll love this! :)
reply
0 sats \ 2 replies \ @noknees 2h
so that means new items will be more visible easily somehow on the top page even if it persists for a small time?
reply
0 sats \ 1 reply \ @k00b OP 2h
No, it means what is top will be determined based on sats and nothing else
reply
0 sats \ 0 replies \ @noknees 2h
ohhh
reply