I have a kind of systemic idea of what you're noticing, if you'll indulge me.

If people are using Active Sort, then most of the front page and the next couple of pages will be dominated by posts that are a day old or more at most (or, more accurately, are more old than new). Active Sort resets a posts timestamp to the most recent comments timestamp until the post is at least two days old, then it switches back to its original timestamp. The delta between a posts timestamp and the date when rank was recalculated heavily impacts a posts rank. The score (upvote - downvote = score) then influences the rank some. So, this means that posts that generate comments have a higher likelihood of remaining on the front page, since each comment makes the post appear "newer" to the algorithm. This makes the post more likely to be on the front page, garnering it more chances for upvotes and comments. So long as the topic is "chatty," it remains on the front page.

There is a minimal window of time that determins when a post lives or dies. It's, I think, the first 5 hours of the thread existing. If the post can't generate any comments in the first 5 hours, it's effectively off the front page (contingent on volume of new threads), and most people likely won't see it.

Scaled Rank is supposed to solve this by considering statistics about the community the post comes from; the smaller the community is or the lower its engagement, it gets an inverse proportional boost to its rank calculation (lower = more boost, higher = less boost or no boost). It doesn't consider new comment timestamps.

Hot Rank, only considers the post's original timestamp and its score, which can make its rank results look very similar to Scaled Rank but also means that posts typically fall off the front page in at least 5 hours, moving newer threads up in the ranks faster. A newer thread that gains substantial votes in a small window of time hits the front page faster as the others decay. Those higher-scoring posts will fall off, even if they're generating more "conversation." (I put this in quotes because, that "conversation" could also be drama, arguments, struggle sessions, whatever you want to call it. All comments are "positive" to the algorithm regardless of its content)

The "reddit model" Lemmy is emulating requires a network effect to work properly. On a larger scale, open community creation is good since it lets more niche interests have a space to collectivize, and that list of interests is crowdsourced. However, when you do not have the correct scale for that model, you don't have enough people to support or sustain those niche interests. In theory, the network effect generated by federation should allow for niche interests.

In practice, however, the LemmyUI and the nature of its federation mean that to the end user, what is listed in the "Communities" tab are all the communities that exist. You can switch the view to All, instead of Local, but that list is only a list of "all remotely subscribed communities" and not "all communties on all connected instances". So discovery of those external communities is basically contingent on users A) knowing where to find external communties, and B) subscribing to them. The LemmyUI has made strides in making that process easier (being able to subscribe to an external community, from that instances page, to your local account on your instance). Due to the barriers to community discovery, the network effect that should be generated by federation is extremely hampered. This leads to nearly every instance having its own version of a generalized niche, like gaming, TV, movies, music, news, etc. What makes those groups unique then is the moderation and the underlying ideological presentation of the instance.

There are numerous communities on Lemmygrad, and that's due to community creation being open to everyone; however, the number of communities, I think, is out of proportion to the local user base. Lemmygrad gets, according to its front page, 524 unique active users a month, and from the /site/ API endpoint, it has 694 communities (I double checked this by using community/list?type_=Local and got 556 local communties, not sure why there is a discrepancy). That's just over 1 community (1.3) per user a month. That definitely isn't enough people to support all the different niche communities that exist on Lemmygrad. To put it into contrast, Hexbear has 1649 unique active users per month and 137 communities (again, double checked this and got 115 local communities). Because of the nature of community discovery, there isn't a good way to expose the wider network to all of Lemmygrad's niche communities to try and build an external subscriber base.

How many of these communities have active moderators, I wonder? Are these communities the result of a handful of users creating them, or are they almost all owned by unique people? I know I'm guilty of this. I have a community that only I can post to called !wizardstower@lemmygrad.ml. My use of it ranges between using it as a personal blog or an archive for comments I make. Probably not the best usecase for a Lemmy community, frankly.

I wonder if a more directed and curated approach to communities, in the way that Hexbear curates its list of communities, would be beneficial to Lemmygrad.