Hey all!

I assume its a bug, but Im not sure, but some posts seem to get nuked from the hot page despite being up in likes. This is unfortunate because they dont get seen as much.
Here's the proof:

1 Post to the left has 52 likes and my post 72
2 Here you can see the left post in the second row. My post should be right next to it, however, it's not there, despite having more likes than the one that is on the hot page.

Anyone know why that is? It happened to another artist too, Nat127. He messaged me about it and asked if I knew anything.
Is it some E6 policy that im not aware of? Is it some sort of shadowban?

Kind regards
Syn-6

I've noticed this 2 times with my work.
The ambassadors who were sent earlier than mine and scored fewer up arrows end up in the hot spot, but my work is simply not there.
moreover, after a couple of hours, they appeared for 15 minutes in the right place and then magically disappeared again.

I've seen a similar algorithm, and according to its values, these posts should have been at least certain, but they just weren't in the hot section.

relation = relation.order(Arel.sql("log(3, posts.score) + (extract(epoch from posts.created_at) - extract(epoch from timestamp '2005-05-24')) / 35000 DESC"))

Yeah looking at the way the hot page works, it should be impossible for posts with less points posted hours or DAYS before Syn-6s art to be ahead of it, and yet they are. Also others have seen the posts appear in the right spot for about 15 minutes before suddenly being dumpstered way down the list suddenly, another thing that should be impossible with an impartial system. So, what the heck is going on? Why are certain posts having an arbitrary penalty applied?

It simply isn't possible to weigh the rank order in any meaningful way
https://github.com/e621ng/e621ng/blob/851cd5c68a922a39b7b300670b0d988b4c41693e/app/logical/elastic_post_query_builder.rb#L303

The code for it is hardcoded and only based on the post's creation date and score

is the scoring function

This same scoring function has been used completely unchanged for years

donovan_dmc said:
It simply isn't possible to weigh the rank order in any meaningful way
https://github.com/e621ng/e621ng/blob/851cd5c68a922a39b7b300670b0d988b4c41693e/app/logical/elastic_post_query_builder.rb#L303

The code for it is hardcoded and only based on the post's creation date and score

log(score) + log(3) + (created_at_millis / 1000 - 1116936000) / 35000

is the scoring function

This same scoring function has been used completely unchanged for years

well that's kind of exactly the point, how can a post that is both older and lower scored be ranked higher than a newer post with many more points? according to the e621 source code like you showed, that should be impossible, but that's what's happening.

trashy_panda said:
well that's kind of exactly the point, how can a post that is both older and lower scored be ranked higher than a newer post with many more points? according to the e621 source code like you showed, that should be impossible, but that's what's happening.

It would definitely be better to actually plug numbers into the equation and see what the result is rather than assuming conspiracy

donovan_dmc said:
It would definitely be better to actually plug numbers into the equation and see what the result is rather than assuming conspiracy

I can't provide screenshots now, but I uploaded the work just yesterday and it wasn't in the hot ones, while the work uploaded earlier and with 3 times fewer up arrows was on the second page.

I don't think there's any conspiracy here, it seems to me that there's some kind of error on the site that's causing some of the work to simply not be displayed in the hot.
(and it was still 15 minutes when my work suddenly appeared in the second line and then just as suddenly disappeared)

donovan_dmc said:
It would definitely be better to actually plug numbers into the equation and see what the result is rather than assuming conspiracy

this is easier to solve at the moment of the event.
so could you tell me any contact or resource where you can contact if this situation arises again?

donovan_dmc said:
It would definitely be better to actually plug numbers into the equation and see what the result is rather than assuming conspiracy

Oh, I have a good example.
1 here is the SYN work: https://e621.net/posts/5702005
2 It's the second work: https://e621.net/posts/5701748
the second work is uploaded earlier, has fewer arrows at the top, and is on the second page of the hot (At least for me)
https://postimg.cc/N591FTgc
The second work is uploaded later, has fewer arrows at the top, and is hot at the end of the first page.
I don't see the SYN work at all! not on the first page, not on the second, not on the third anywhere, although logically it should be higher

donovan_dmc said:
It would definitely be better to actually plug numbers into the equation and see what the result is rather than assuming conspiracy

you should be able write a pretty simple userscript to have what the ranking values should be displayed below the posts' thumbnails, both of the needed variables are available as attributes.

donovan_dmc said:
It would definitely be better to actually plug numbers into the equation and see what the result is rather than assuming conspiracy

not assuming anything, just stating an observation. you're absolutely right though so I'll give it a shot.

Syn-6's post here: https://e621.net/posts/5702005 I get

log(3,87) + (1752266509 - 1116936000)/35000 = 18,156.3653

another post ranked much higher: https://e621.net/posts/5701165 I get

log(3,16) + (1752237975 - 1116936000)/35000 = 18,151.9847

so it should be ranked lower, but it isn't. If my math is wrong please tell me, I'm just going based off what I see in github.

There is something goin on. If I go look at the hot page right now it is missing a couple of posts from the top of popular page. From what I understand, a post with 2000 favourites within 12 hours used to show up on the hot page, but now it seems random wherher it shows up or not.

This has been going on at least since April. To me it seems like if a post is too popular is gets kicked out of "hot" and only shows up on "popular".

kaitzuwu said:
This has been going on at least since April. To me it seems like if a post is too popular is gets kicked out of "hot" and only shows up on "popular".

Thank you, Ive noticed it too! I never understood the difference between popular and hot too much. But of course this issue is annoying since E621 is an important portal to get some eyes on one's work. I owe a plethora of commissions I did to my E6 presence! I know the folks coding the site do it voluntarily and im sure its a moderation nightmare at times, but I do hope they fix or adjust the hot system to accurately reflect a posts well... "hotness".

Some other commentor asked for evidence: look at my initial post. As you can see, posts with fewer upvotes somehow make it to hot even though they are posted at the same time as another post (with more upvotes).

dba_afish said:
you should be able write a pretty simple userscript to have what the ranking values should be displayed below the posts' thumbnails, both of the needed variables are available as attributes.

I've done just this, it can be installed via this link (presuming tamper monkey is installed), and its code can be viewed here
Note that this script requires disabling Re621, if you have that installed

From what I'm seeing in the elasticsearch code this should be a 1-1 match, and the numbers do seem a bit all over the place
There is some semblance of order, but it is not consistent

donovan_dmc said:
I've done just this, it can be installed via this link (presuming tamper monkey is installed), and its code can be viewed here
Note that this script requires disabling Re621, if you have that installed

From what I'm seeing in the elasticsearch code this should be a 1-1 match, and the numbers do seem a bit all over the place
There is some semblance of order, but it is not consistent

Small error in your userscript, it's dividing the milliseconds by 100 instead of 1000, but even with that fixed I'm seeing the same thing. The numbers do overall trend down as you go down the page, but it's like there's multiple rank lists being merged together for some reason? Whatever the pattern is it isn't obvious.

trashy_panda said:
Small error in your userscript, it's dividing the milliseconds by 100 instead of 1000, but even with that fixed I'm seeing the same thing. The numbers do overall trend down as you go down the page, but it's like there's multiple rank lists being merged together for some reason? Whatever the pattern is it isn't obvious.

Apparently the raw link I copied earlier linked to only the initial version, I fixed that hours ago

donovan_dmc said:
Apparently the raw link I copied earlier linked to only the initial version, I fixed that hours ago

appreciate you putting that userscript together because it makes it really easy to see that it's a problem with order:rank in general. any idea of what the issue might be? has anything like this happened before?

donovan_dmc said:
Apparently the raw link I copied earlier linked to only the initial version, I fixed that hours ago

well, whatever it is, its still happening with my current posts! :/