Relatedly, accounts that engage in followback-inviting behavior (follows many people, and then unfollows most of them in order to maintain a misleading resemblance of equilibrium for followers/following numbers)

You unfollow 87% of the people that you initially start following (after they receive the notification).

That's the whole point of this list.

For some reason it looks like you've been following/unfollowing one specific account (@retrodotcards.com) repeatedly multiple times per second, on November 2nd.
No idea if this was a bug or done manually.

I agree that it's misleading to count them as separate follows, will fix
pastebin.com/t2R0yAR2

Publicly available/OSS, not that I'm aware of.
But it's not complex per se, just record user follows and deletions, join, and calculate the unfollow ratios.

The Bluesky firehose data is public.
With some custom code anyone can compute this kind of statistics.

Of the people you follow, you eventually unfollow 58% of them, almost as if you followed them only to solicit followbacks and not out of actual interest.

You follow people (so they get a notification and maybe follow you back) and then unfollow 60% of them.

The problem is not bots and automation, the problem is behavior. I don't care if you spend 2 hours a day manually following and unfollowing random people, or if you automate that. The effect is the same.

That's because they conceal these unsuccessful followback solicitations by deleting them, so that they don't look like desperate spammers to the average user (followers: 70, follows: 10,000).
Not to me, since I track follows and subsequent unfollows.

"Who don't do these things"
Did you run some data analysis on the Bluesky Firehose? I doubt so.

Cool story.

You followed TENS of THOUSANDS of users, and then unfollowed HALF of them.
Why, if not to solicit followbacks?

Said the guy who unfollows 97% of his followees

You unfollowed 64% of the followers you added during the past few days.

You unfollowed 62% of the accounts you followed just a few days before.

Yes they will expire.
I looked into labelers, but they seem to require a lot more infrastructure to set up (a server, with websocket connections).
Lists on the other hand only support muting (not warning), unfortunately.

It's updated dynamically based on the data from the firehose.
Turning it into a label is on the roadmap.