A premoderation system would be strongly biased in favor of publishing tweets—in fact, if it worked right, Twitter would rarely, if ever, actually get involved. This wouldn’t be like a commenting system in which every utterance needs to be approved before being published. It would be much more passive. There would be, say, a 30-second delay on every tweet (so a bit longer than a live TV delay), during which time someone from Twitter’s team would have to look at it. This would not be a task for algorithms—we would need actual, smart humans on the job. (Twitter would have to pay these people awfully well and rotate them out frequently, of course—spending all day staring at a screen waiting for a handful of Trump’s tweets would be terrible.)
If nothing is clearly dangerous, the tweet would go live as normal. But if there were a major issue—like a threat of military action, or something suggesting that the account had been compromised—then the Twitter representative would make sure that the tweet doesn’t go live. Twitter would need a detailed social media triage and crisis communications plan, one that outlined how things should be escalated and managed.