The problem is that censorship is in most cases hard to identify.
They don’t need to announce “you’re banned”.
Delaying your messages, not returning them on topic queries, removing them from global feed, etc. Also simply blocking your posts to some geographic areas is easy without alerting users in other areas.
Like a Twitter employee said “it’s not whether you are allowed to post that matters, it’s who actually gets heard”.
Manipulating discoverability is very easy for relays and very hard to detect.