My point 1 was specifically phrased to clarify how it could have been more reliable before migration to FB because it did not have to deal with the same load back then, you said nothing to show it was not a correlation.
Your point 2 sounds like there were additional factors that could have influence the reliability besides the load (they didn't simply migrate to FB infra but also switched to Signal).
> The broadcasting is important, yes, but it is very heavily biased towards reads over writes, so something like Cloudflare would solve 99% of the load.
There's push-notifying millions of devices within seconds after a celeb or a major news source tweets. There's tracking view and engagement stats on that in realtime. There's making sure a tweet is not available to any of those within seconds after it's been deleted or moderator. There're separate back-office apps for moderating that firehose of content. And that's just what I can see from the outside. An e2ee instant messenger with size-limited chat groups doesn't even come close.
Please don't say "just stick a CDN on top of it and you are 99% there", it's embarrassing (and not to twitter). This will maybe get you 80% there if your goal is "a microblogging platform" but not even 20% if your goal is being both the go-to news source and shitpost forum for people worldwide reliably working even in sensitive times and emergencies. Twitter used to be a microblogging platform back when it had much fewer employees and you'd see a fail whale regularly even as it had much fewer active users, in recent yeas it's a completely different beast and saying increased headcount is unrelated is amusing.
