Case in point, I just edited this message to change all "_" to "*" to make italics work. That's another use: fixing wrong assumptions about usage of technology itself.
No, my position is that if 5% of a problem (5 out of 100 people seeing it) is equally catastrophic as 100%, then it makes no sense to differentiate, which is a very true dichotomy.
The fact that just 1 person that's not supposed to see it seeing it can do damage, makes the problem a binary and not a spectrum.
Which one occurs more often than the other, that's another question, but I'd argue that the later is more common, and there is no doubt that people knew, in the event they wrote a very catastrophic secret, the limitations of the edition/deletion of messages (as long as most people nowadays seem to know about screenshots being a thing)
There's a difference between offering corrections and retractions that others may or may not accept and actually enforcing deletions. This is both socially and information-theoretically true.
Obviously on Discord, one is tied to the official client's behaviour since the ToS discourages third party clients. However, for Slack, Mattermost, Matrix etc. clients, the possibilities widen up.
By this time, it's too late, anybody could have seen it, copied it, etc.
Deleting it after the fact is security theater.