There's a decent write-up of how to do this here: https://www.spinellis.gr/blog/20090805/

As you said, it largely relies on reverse engineering and bruteforcing Twitter's ID generation.

Bot author here. That post was before twitter made changes on the id generation. Now it consists of four parts that you need to guess (the datacenterId, the workerId, a sequence and a timestamp). For more details, the bot repo is here: https://github.com/pomber/escher-bot

This tweet (or one like it) caused a massive Twitter outage.

Yeah, "If you type 'google' into google.. you can breakthe internet, so please no one try it" IT Crowd /s

Nothing like finding a loop in a simple graph...

Citation?

Used to work there :D

Was it because Twitter tried to recursively load the previews?

Was there a post-mortem about it?

Is that because some algorithm assumed link cycles weren't possible?