@stripey @gravepapaya @molly0xfff That deletion issue is even more noticeable with fediverse: I’ve had software ignore my (test) attempts to undo a fav. I run Pleroma, and I think it was Friendica ignoring the undo. This was years ago; presumably that particular glitch is fixed.

This made me seriously rethink whether it’s any safer to post on a federated instance than on a centralized walled garden. Even self-hosted instances propagate a post widely and out of author’s control; with walled gardens, at least I have no illusions that I am not the admin and that the admins may keep a copy of my deleted posts (which, if they’re scrutinized by regulators, they likely do not actually keep — but they might). With self-hosted instances, I personally felt an illusion of data control that’s not really there.

Therefore I concluded that it’s not safe for me to post short opinions of sort that I might regret 1min, 1h or 1d later. (I sometimes would hit the send button, then rethink, then delete.) So I rarely post on fediverse despite believing it’s morally better than the walled gardens that I stopped using for other reasons.

And there’s no real sync mechanism with fediverse: just shutting down the instance for long enough means you might never be re-sent the delete requests.

Neither of these two examples was software intentionally breaking a social contract — these are bugs or system failures. No need to modify the instance software to actively ignore delete requests, take regular snapshots or write new software that just accepts create activities while ignoring delete activities.

There’s at least a chance of Matrix software catching up (they have the sync mechanism for room which I don’t really like for other reasons, but it’s there). With fediverse, the sending instance would have to keep the delete activity around long enough so that recipient instance (which may be something like 6mo down) catches up. Recipient instance won’t be continuously polling “is this deleted” for all posts it hosts a copy of — that would be just continuous useless traffic — and the sending instance will give up to save resources.

Receiver can catch up on posts by polling the posts feed; will it catch up on activities operating on those posts? Not sure what’s in the feeds, honestly, and the lookback will also only take you so far anyway.

XMPP gets closest to what I feel somewhat safe, with “a chat message and pubsub post will likely not get replicated widely, and it will be fetchable on demand”. This experience with XMPP is where my illusion of self-hosted data control originates from.

With fediverse and its replication/caching and software bugs and potential for third party scrapers to archive everything, that expectation is not really there.