At Twitter scale, arbitrary hard-coded rules are almost always useless; they make life more difficult for authentic posters while nefarious actors can work around them with varying degrees of triviality.
With hand written (not arbitrary) rules it's easier to understand the intent of the attacker and build a system that they can't work around because we're blocking them at their source of income. Sure they can figure out how to post messages but unless they can include their link/payload/etc it's not worth their time.
Machine learning defences are definitely a part of what we did, but they're slower to respond to attacks and generally easier to work around.
I disagree; just pointed out how it's not hard to get pure spam by using the filtered stream rules. If I can reliably identify & filter for spam on my creaking desktop with limited compute power and technical/coding skills, I would be happy to operate a silicon backhoe for a modest fee.
https://developer.twitter.com/en/docs/twitter-api/tweets/fil...