To bring another data point to the issue the OP is bringing (personally I don't use the "Hot" order, I'm accustomed with using chronological ordering; just helping the OP with additional data point from a slightly older account to the glitch they're observing): to me, "Hot" correctly sorts the entries by the number of upvotes (edit: and it also takes into account the amount of replies)... Except it only does so for the items that were fetched for the current scrolling page, which is 15 entries. When the user scrolls down to the end of the list, Catodon fetches another 15 entries, but these will be sorted independently from the previous entries, so it gets a stairway-like effect when it comes to upvote counting across entries.

It seems to be just the "hot" order, though, and maybe the "Active" (I didn't test the "Active", it likely experiences the same phenomenon given it probably does chunked fetching too). The "new" sorting order is working as intended.

Edited to correct my oversimplification of "hot = upvotes" and to bring additional screenshot from the "staircase effect".

I will look into what you're reporting for Hot - the rating should be for all posts from the last 3 days. Perhaps what happens might be that in the meantime some ratings changed, since some time has passed. Active shouldn't have the same issue in any case - it's sorting by last reply. Worst that could happen should be to lose a post that went from say 30th position to the 1st because of a new reply while you're scrolling.

@radio4392 @dsilverz please refresh the page and try it again. When we implemented this we just adopted Lemmy's algorithm, because why reinvent the wheel, but as it seems in practice it was boosting newer posts much. So I have adjusted it on Catodon based on your feedback. Should be better now.
BTW, replies count doesn't matter now, but I will further tune it in the next release (after 26.10.0) and it will also be affected by the replies number. I didn't want to make a bigger change now since we're already in the RC phase for 26.10.0.

Please tell me if it's better now!

@radio4392 @dsilverz I also took forum posts out of Hot in timelines outside Forums, because they had a big advantage and drowned regular posts out

@panos
I can hardly tell the difference. Sometimes a post with several likes pops up. However, the posts are the same as in the normal feed.

@radio4392 I made more adjustments, please test again. For transparency, how recent a post was counted x1.8 in the initial algorithm, x1 in the previous fix, and x0.5 now. This should be enough to help even in low engagement timelines. @dsilverz

I’ve been keeping an eye on this over the last few days. It’s got a bit better. But mostly I only see three posts with lots of likes in the ‘Hot’ feed, and the rest are just sorted chronologically.

I'd like to suggest this: Do you notice the same thing happening in the global ("All servers") timeline? If not, then I think that the culprit is probably that your feed mostly has low engagement posts, and I don't think that we can do something about it, given that the algorithm takes into account likes/reactions, reposts, and time passed. If there are not enough reactions and reposts for a post to be bumped (like the three posts you mention) then getting a more or less chronological order is natural. I have reduced the effect of time passed to almost 25% of the Lemmy algorithm (x0.5 from x1.8) so I don't think there's much more to do on our side... Open to ideas, of course.

it works in the global timeline.
From when, and for how long, is something considered ‘hot’?

The Hot timeline gets posts from the last 3 days for the microblogging timelines, and from the last 7 days for forums.

OK, I tried something else: Now the number of replies is also taken into account. Hope this helps a little.