107 sats \ 2 replies \ @Brunswick 15 Jul 2022 \ parent \ on: Daily discussion thread
That depends on what percentile of what. I dont know what the distribution looks like. I assume there are three or four groupings of scores and they aren't following a single gaussian distribution, probably at least three clumps of scores.
If I were to try to make it algorithmic, I'd probably try to collate the ranks into 100 bins, smoothe the bins through a moving average or high order polynomial fit, then search starting at a score of zero and look for a local minimum between new/inactive users and active users, and set the threshold there.
write
preview
0 sats \ 1 reply \ @k00b 15 Jul 2022
Fancy. Okay, mostly trying to gauge how new/untrusted you meant to filter out
reply
100 sats \ 0 replies \ @Brunswick 15 Jul 2022
TBH I was originally thinking you'd give the user a spinner or pulldown to set it to whatever you like (either on the recents page or in settings) but then I thought that would be too complicated but maybe setting a fixed threshold is less optimal. It's possible I just dont get your trust system and am thinking it's useful where its not
reply