If I got it right, in an eventual consistent system you would accept and confirm events from a user, but really just keep the event in a queue, until it gets completely processed, including the effects becoming visible for the user.
But since in between by the very definition the system isn't 100% consistent, processing of the event might run into problems. As an example consider the event consisting of a change to an object which already got deleted.
What are patterns to present this kind of failure to the user?
The only thing I can come up with right now is a list of events, that still need processing, with markers about failures, but this seems to be a rather technical approach, that would probably by rather confusing for most users.