The ICYMI (ADN) bot has not been seeing a fairly large set of possible inputs for some time, meaning that it has only seen the accounts which it follows directly and essentially worked as an aggregator of a (small) fixed set of accounts, instead of also being able to look at a wider range of possibly-relevant toots and amplify those as well.
Why is this happening?
The ICYMI (ADN) bot was first set up on the botsin.space instance. botsin.space is very bot-friendly, was easy to register with, and seemed a good home. It is full of lots of other interesting and innovative bots – and if you’re looking for some quirky or useful bots to follow, it’s a good place to start.
This ICYMI bot is more of a lurker than many other bots – it mainly works to boost other toots, but it will also boost toots from people it doesn’t follow. It does this by listening on hashtags, but here I hit the problem that botsin.space is an abnormal instance populated by abnormal accounts.
Look at the algorithm for toot visibility – here I’ll use Cassotl’s diagram of toot visibility, available on wikicommons:
Critically, if no one else on the instance follows the author of a toot, it doesn’t matter if it has a hashtag which the bot follows, the toot will never be loaded to the instance and so will be invisible.
The ICYMI (ADN) bot only ever followed a small number of accounts because (unlike it’s law, philosophy and urbanist cousins) the community of accounts active on the topics of Arms Control, Disarmament and Non-Proliferation on Mastodon is small. The code was adjusted to also work from hashtags (and use hashtags to judge the relevance of toots).
What is the fix?
I think the fix should be fairly straightforward. I am going to move the bot to another instance, one which is more representative of the broader Mastodon community.
@icymi_law@esq.social and @ICYMI_urban@urbanists.social are both on special-purpose instances. @icymi_philosophy@botsin.space is also on botsin.space but follows about ten times as many accounts.
When will this happen?
Because there is a cooling-off period for transferring accounts I want to be relatively sure that I choose a new instance correctly. I don’t want to move to a new instance and then find that the bot doesn’t comply with the instance’s bot policies or cannot be made to comply. I’m looking at three possible new host instances and will choose in the next few days.
I’ve not done an instance transfer before but it feels like everyone I follow on Mastodon has done it at least once. I hope the bot should be settled in its new home, with nearly all its followers, and working better, by the middle of January 2024.
Other comments?
The bot has now been running for almost a year, and running with an essentially unchanged algorithm for about ten months. I have not put much effort into it recently but had been quite content that it was working fairly well (I thought) without intervention.
The proposed fix here is a quick and light touch adjustment. I have found the bot useful for what I do (it helps me find toots which are relevant to my work) and I want it to keep going, and to work better.
I have ideas of how to improve the logging and tracking of the bot so I can better analyse things with less after the fact traffic to the server. (Currently my monthly data analysis takes an hour or so to pull data from the server, data which the bot had on hand when it first made decisions about boosting or not boosting toots, but which it then throws away.) I hope that moving to the new instance will improve the relevance and performance of the bot.