rapid unscheduled torment nexus disassembly expert — nonbinary, she/they — works on CI platforms at Google — writes Racket elsewhere — admin of @claude.notjack.space
Scale. There are lots of people with tens of thousands of followers who post a lot. And people really want to find out about new posts quickly. It's a many-to-many high throughput low latency write delivery problem. And it's the only one: everything else bluesky does is either many-to-few, few-to-many, low throughput, or high-latency, and can thus be way cheaper to scale.
This is my favorite thing about atproto too, and it really bummed me out when I learned that Tangled only pushes pull request records, they don't ingest them and turn third-party-authored pull request records into commits on new branches. I was so looking forward to hooking my automated code refactoring tools up to a PDS.
I chatted with one of the devs though and they stated it's on their roadmap, so that's something. I really think people haven't grasped what's possible yet.
The tricky bit is you can't just limit Tap to ingesting only the repos of accounts in your PDS, you have to ingest the accounts of anyone that someone in your PDS follows. With even just a few dozen users that can quickly become thousands or tens of thousands of accounts.