You should always be aware that anything you post online might not be deletable after you post it, even if the platform itself has a delete button.

On that platform, the delete button might actually be a hide button (or a hide and flag button, since deleted comments might be more likely to be interesting, since the author changed their mind about saying it).

But even if the platform does honour the delete action in good faith, crawlers could have already saved it. There were sites specifically for showing deleted comments for Reddit before the API change, and I wouldn’t be surprised if there are some today that just parse the html.

With the way the fediverse works, someone could create their own implementation that specifically highlights the parts that most servers hide. Like deleted posts, voter identities, I think even private messages that cross from one instance to another.

Safest to assume that once you post something, it’s out there forever and might not be as private as you’d like. And, depending on the platform, even those posts that get written but then the author decides to close instead of post could be saved somewhere. If the text has any real-time interactions with the server, the question is if the client or server is responding to those. For example, I don’t think that Google sends all search suggestion possibilities to all clients so they can locally suggest searches based on text entered.

Also try typing @ followed by random letters here. The server is open source, which means you can look at what it does. But it also means instance admins can edit that code to do whatever on their server.